Classic Microfinance Model

American Economic Review 2013, 103(6): 2196–2226
http://dx.doi.org/10.1257/aer.103.6.2196
Does the Classic Microfinance Model Discourage

Entrepreneurship Among the Poor? Experimental
Evidence from India†
By Erica Field, Rohini Pande, John Papp, and Natalia Rigol*
Do the repayment requirements of the classic microfinance contract

inhibit investment in high-return but illiquid business opportunities
among the poor? Using a field experiment, we compare the classic
contract which requires that repayment begin immediately after
loan disbursement to a contract that includes a two-month grace
period. The provision of a grace period increased short-run business
investment and long-run profits but also default rates. The results,
thus, indicate that debt contracts that require early repayment
discourage illiquid risky investment and thereby limit the potential
impact of microfinance on microenterprise growth and household
poverty. (JEL A21, G32, I32, L25, L26, O15, O16)
Lending to entrepreneurs is a risky proposition in the best of cases. In developing

countries, where borrowers often do not have collateral to seize in the event of a
default, this risk is even higher. Somehow microfinance, which has expanded rap-
idly from its roots in Bangladesh in the late 1970s (Daley-Harris 2006), has struc-
tured debt contracts so as to limit the risk of lending to poor entrepreneurs and for
that reason is considered an important tool for helping the poor.1 Early initiation of
repayment is widely considered an important means by which the classic “Grameen
model” limits lending risk.2 Yet there is growing evidence that microfinance, despite
its success in achieving high repayment rates, has had little impact on microenter-
prise growth and poverty (Banerjee et al. 2009; Karlan and Zinman 2011; Kaboski
and Townsend 2011). This is particularly surprising given substantial evidence that
* Field: Department of Economics, Duke University, 213 Social Sciences Building, Box 90097, Durham, NC
27708 (e-mail: field.erica@gmail.com); Pande: Harvard Kennedy School, Harvard University, Mailbox 46, 79 JFK
Street, Cambridge, MA 02138 (e-mail: rohini_pande@harvard.edu); Papp: Highbridge Capital Management, 40
West 57 Street, New York, NY 10019 (e-mail: johnhpapp@gmail.com); Rigol: Department of Economics, MIT,
50 Memorial Drive, Cambridge, MA 02142 (e-mail: nrigol@gmail.com). We thank Emmerich Davies, Sitaram
Mukherjee, and Anup Roy for superb field work; the Village Financial Services (formerly known as Village Welfare
Society) and Center for MicroFinance for hosting this study; and Yeunbee Jeanette Park for exceptional research
assistance. Theresa Chen, Annie Duflo, Nachiket Mor, and Justin Oliver generously enabled this work. We thank
ICICI Foundation, the Exxon-Mobil Foundation, the International Growth Centre, and the South Asia Initiative at
Harvard for funding. We thank Abhijit Banerjee, Tim Besley, Dominic Leggett, and numerous seminar participants.
†
Go to http://dx.doi.org/10.1257/aer.103.6.2196 to visit the article page for additional materials and author
disclosure statement(s).
1
The importance of microfinance as a tool for helping the poor was recognized in 2006 when Muhammad Yunus
and the Grameen Bank were awarded the Nobel Peace Price.
2
In 2008, microfinance institutions had an estimated 130–190 million borrowers worldwide and outstanding
loans exceeded $43 billion (Gonzalez 2010).
2196
2197
VOL. 103 NO. 6 THE AMERICAN
field et al.: classic ECONOMIC
microfinance REVIEW
model and entrepreneurship
october 2197
2013
credit constraints inhibit small business expansion (Evans and Jovanovic 1989;
Banerjee and Duflo 2012) and that returns to capital in this sector are high (de Mel,
McKenzie, and Woodruff 2008, 2012).
This paper examines whether the immediate repayment obligations of the classic
microfinance contract inhibit entrepreneurship, and therefore blunt the potential
impact of microfinance, by making high-return but illiquid investments too risky
for poor borrowers. To shed light on this question, we conducted a field experiment
with poor urban borrowers in India that evaluates the short- and long-run effect of
relaxing the liquidity demands early in the loan cycle. We randomly assigned 169
loan groups of five clients to one of two debt contracts: Clients assigned to the
control group received the classic contract that required them to initiate repayment
two weeks after receiving their loan, as is standard practice in microfinance (hereaf-
ter, regular contract). Clients assigned to the treatment group received a two-month
grace period before repayment began (hereafter, grace period contract). All loans
were individual liability contracts, and once repayment began, clients repaid at an
identical frequency.
Survey data on loan use and long-run business profit showed that the intro-
duction of a grace period led to a significant change in economic activity:
Microenterprise investment was approximately 6.0 percent higher and the like-
lihood of starting a new business was more than twice as high among clients
who received the grace period contract relative to those on the regular contract.
Furthermore, nearly three years after receiving the loan, weekly business profits
and monthly household income for grace period clients were, on average, 41.0 and
19.5 percent higher and these clients reported roughly 80 percent more business
capital. Taken together, the profit and capital increases correspond to a monthly
return of capital of 11.0 percent.3
These large effects of debt structure on investment behavior cannot be reconciled
with perfect credit markets. Rather, they suggest an environment where clients face
borrowing constraints and where illiquid investments yield higher returns, an inter-
pretation that is supported by case study evidence. In the presence of borrowing
constraints, illiquid investments are likely to be riskier since they reduce clients’
ability to deal with shocks.
Consistent with this interpretation, we find evidence of heightened risk-taking
among grace period clients. In the short run, they were more than three times as
likely to default than regular clients. In the long run (three years after they received
their loans) they reported riskier business practices which reflect a greater willing-
ness to reduce their access to liquid funds and experiment with product and client
diversification: relative to regular clients, they were more likely to extend credit
to customers through loans and pre-orders and to offer a wider array of goods and
services.
Thus, by limiting illiquid investment choices, the immediate repayment obliga-
tions of the classic microfinance lending model may simultaneously limit default
3
All estimates refer to the top coded sample. A simple accounting exercise verifies that these differences, though
large, are consistent with a return differential of 2 percent per month (6 percent for grace period clients and 4 per-
cent for regular clients) generated by the initial difference in investment behavior followed by compounding of these
returns over the next three years.
2198
microfinance REVIEW
october 2198
2013
and income growth. A heterogeneity analysis shows that more risk-averse clients
and those with fewer means of dealing with short-term liquidity needs benefit more
from a grace period contract. These results help reconcile experimental estimates of
high returns to capital among micro-entrepreneurs in developing countries with the
low estimated impact of microfinance on business growth of the poor. Our results
indicate that, while access to credit places a binding constraint on microenterprise
activity, successfully relaxing those constraints remains sensitive to the mode of
credit access.
Given the higher business profits among grace period clients, it is natural to ask
why most microfinance institutions (MFIs) do not offer a grace period contract at
a higher interest rate as part of their loan portfolio. Indeed, survey data indicate a
high willingness to pay for a grace period among a substantial portion of our study
clients. To shed light on this question, we calibrate a simple model of MFI profits
when it offers both a regular and a grace period contract and clients self-select across
contracts. Our calibration utilizes survey data on client preferences and experimen-
tal data on their default risk. The calibration suggests that asymmetric information
in credit markets is an important reason for the absence of grace period contracts: in
the absence of moral hazard MFIs can break even when they offer existing clients
the choice between the regular contract at 17.5 percent interest and a grace period
contract at 37 percent. However, no such zero profit separating equilibrium exists
if, in addition, study clients exhibit a modest amount of moral hazard and/or there
is adverse selection into the grace period contract by new clients (from outside the
study client pool). Thus, consistent with a large literature on asymmetric informa-
tion in credit markets, it appears that the contract that maximizes MFI profits also
creates inefficiencies via underinvestment.
To the best of our knowledge, this is the first paper to demonstrate how early ini-
tiation of repayment for microfinance loans may prevent high return investment by
poor entrepreneurs. A small and predominantly theoretical literature examines the
role of repayment frequency in reducing default in MFIs, but focuses on channels
other than investment choice.4
In contrast, the idea that the structure of debt contracts influences entrepreneur-
ial risk-taking and investment exists in many corporate finance models. One line
of reasoning argues that longer term debt reduces risk-taking because lenders cap-
ture part of the returns of new growth opportunities (Myers 1977). On the other
hand, if shareholders use risky investments as a way to capture debt-holder wealth
(Jensen and Meckling 1976; Tirole 2005), shorter term debt can reduce entrepre-
neurial risk-taking (Barnea, Haugen, and Senbet 1980; Leland and Toft 1996). The
empirical literature presents support for both mechanisms (Barclay and Smith 1995;
4
Fischer and Ghatak (2010) show with present biased borrowers, the optimal contract (in terms of loan size)
requires frequent small repayments. Fischer (2013) identifies peer monitoring as an alternative reason for low
risk-taking by microfinance clients. On the empirical front, observational studies of how greater repayment flex-
ibility affects default report mixed findings (possibly reflecting selection bias): Armendáriz and Morduch (2005)
reports that more flexible repayment is associated with higher default in Bangladesh, while McIntosh (2008) finds
that Ugandan MFI clients who choose more flexible repayment schedules are less delinquent. A few recent papers
circumvent the selection issue by providing experimental evidence on the effect of changing repayment frequency.
More frequent meeting improved new clients’ informal risk-sharing arrangements and, therefore, ability to repay
moderately sized loans (Feigenberg, Field, and Pande 2012). However, no change in default was observed when
either loan amounts were very small (Field and Pande 2008) or group members were well-acquainted (Field et al.
2012).
2199
microfinance REVIEW
october 2199
2013
Brockman, Martin, and Unlu 2010). Consistent with our findings, these papers, and
the corporate finance literature more broadly, emphasize the role of informational
asymmetries in determining the optimal debt contract.
Section I describes the experimental intervention and predicted effect of a grace
period contract on client investment behavior. Section II describes the data, client
characteristics, and empirical strategy. Section III reports our experimental findings
and compares the long-run increases in profit and returns to capital implied by our
model with the existing literature. Section IV examines the viability of MFIs offer-
ing grace period contracts, and Section V concludes.
I. Background
With perfect credit markets, introducing a grace period contract should not alter
borrowers’ investment patterns. Microfinance, however, targets credit-constrained
households. As background, we describe our experimental design and then, using
insights from client case studies, identify the likely effect on investment of a grace
period contract and determine which borrowers are most likely to be affected.
A. Experimental Design
Our study was conducted with Village Financial Services (VFS), an MFI that
makes individual-liability loans to women in low-income neighborhoods of Kolkata.
Between March and December 2007 VFS formed 169 five-member loan groups
designated for inclusion in the study, giving us a study sample of 845 clients. Each
client received an individual-liability loan varying in size from Rs 4,000 (∼ $90) to
Rs 10,000 (∼ $225) with a modal loan amount of Rs 8,000.
After group formation and loan approval, but prior to loan disbursement, groups
were randomized into one of two repayment schedules. Eighty-five groups were
assigned to the regular VFS debt contract with repayment in fixed installments start-
ing two weeks after loan disbursement, and 84 groups were assigned an analogous
contract that also included a grace period of two months. Other features of the loan
contract were held constant: Once repayment began, all groups were required to
repay fortnightly over the course of 44 weeks. Repayment occurred in a group meet-
ing conducted every two weeks by a loan officer in a group member’s home (on
group-meetings also see Feigenberg, Field, and Pande 2012). Both groups faced the
same interest charges. However, longer debt maturity (55 as opposed to 44 weeks
before the full loan amount was due) combined with the same total interest charges
implied that grace period clients faced a slightly lower effective interest rate on the
loan (on the potential income effect see Section IIID).
Treatment status was assigned within batches of 20 groups, determined by tim-
ing of group formation (the final batch was smaller with nine groups). No clients
dropped out between randomization and loan disbursement.
B. Should a Grace Period Influence Investment Behavior?
Roughly three years after clients entered the experiment (and just prior to our final
survey) we interviewed a random sample of grace period clients to better understand
2200
microfinance REVIEW
october 2200
2013
their experience. Here, we describe interviews with a sari seller and a tailor, repre-
sentatives from the two largest occupations in our sample.
Both clients were second-time VFS borrowers and experienced business owners,
and neither had non-VFS formal loans. Only the sari seller had a savings account.
The sari seller repaid her loan on time while the tailor was delinquent and repaid the
full loan only 24 weeks after the due date.
When asked directly how the grace period had influenced loan use, both said that
the two-month delay had provided security to invest the entire amount into their
businesses as opposed to setting aside a portion for initial repayment installments.5
A two-month delay provided a sufficient buffer to expect a large enough return to
cover the first installment. Expanding their investment, in turn, enabled economies
of scale in purchasing inputs. For instance, the sari seller explained that she accessed
larger wholesale discounts.
Both clients voiced concern about variability in market demand. Over half of the
sari seller’s clients bought on credit and repaid in small monthly installments. While
on average she could sell Rs 3,000 worth of merchandise for Rs 3,800– 4,500, dur-
ing low seasons she earned as little as Rs 300 per month, which was insufficient to
cover her monthly payment of Rs 500. She felt that the grace period combined with
higher investment return reduced default risk during low months. In contrast, if she
had invested her entire loan while on a regular contract then a low sales month soon
after loan receipt might have required her to liquidate stock. Such liquidation would
imply a loss, reduce subsequent earnings, and increase default risk. The tailor gave
a similar account of the grace period reducing default risk during a low season.
Finally both clients stated that the grace period encouraged experimentation with
new business opportunities, and increased their willingness to take on entrepreneur-
ial risk. For instance, in addition to increasing her stock of saris, the sari seller
expanded the variety of saris she was offering. Prior to the second loan, the tailor
had operated his business with a borrowed sewing machine or sewing by hand. He
invested the VFS loan in a sewing machine as well as in raw materials that allowed
him to expand into the ready-made market. This expansion also prompted him to
establish business connections in Assam, a neighboring state.
A key hypothesis that emerges from these case studies is that a grace period con-
tract allows credit-constrained clients to access high-return but lumpy and illiquid
investment opportunities. More formally, consider a three-period loan contract. A
client receives a loan with principal b at t = 0 and owes payments of amount p1 in
t = 1 and p2 in t = 2. There are two contract types: a grace period contract with
p1 = 0 and p2 = b and a regular contract with p1 = F ∈ [0, b), p2 = b − f.
Suppose the client has access to an illiquid high-return investment with minimum
investment size i min. An investment of i > imin ∈ (b − f, b) pays r × i in t = 2
with r > 1. Clients can save at zero interest rate and default incurs a fixed penalty
d ≥ b. Hence, default is never optimal if the client has sufficient cash on hand.6
5
Both respondents affirmed that they had saved a portion of their previous VFS loan (which had no grace period)
to pay their first few installments. While the sari seller had taken the loan out for her own business, the tailor’s wife
had borrowed for her husband’s tailoring business. Hence, for the tailor, household responses about loan use and
business activity were given by the husband.
6
We implicitly assume that outside income is unavailable or will be needed for purposes other than paying the
loan in t = 1. Put differently, consistent with recent evidence (Dupas and Robinson 2013), we assume that the
2201
microfinance REVIEW
october 2201
2013
In this setup, only the grace period client can invest in the high-return asset. Hence,
the grace period client makes positive profits (of r × b − b ) while the regular
client makes zero profits. Moreover, the profit difference is independent of loan
size.7 While a larger loan size also allows a regular client to invest in the high- return
project, at any loan size the grace period client invests more.
Which borrowers are most likely to change investment decisions when provided
a grace period contract? We consider two categories of factors that are frequently
discussed in the entrepreneurship literature (Hurst and Lusardi 2004). First, if high-
return investments are illiquid and risky (as suggested by our case studies), then the
effect of a grace period should be more pronounced among the more risk-averse and
those who lack alternative income-smoothing mechanisms to buffer against short-
run income shocks (such as a savings account).8 Second, the effect of the grace
period should be lower for clients that lack the skills and access to complementary
markets required to succeed in entrepreneurial activities.9 Proxies for this include
business ownership at baseline and household specialization in business activity.
Similarly, grace period contracts may be poorly utilized by present-biased clients
(Fischer and Ghatak 2010). In our data, we unfortunately do not measure present
bias and will, therefore, restrict attention to client discount rates.
To summarize, this framework raises three empirical questions. First, do poor
credit-constrained entrepreneurs have access to high-return but illiquid and lumpy
projects, such that a grace period changes their investment choices and profit? If
yes, then does the effect of a grace period vary with observable project and client
characteristics? Finally, does the introduction of such a contract increase risk for the
MFI? Below, we use data from our field experiment to investigate these questions.
II. Data and Empirical Strategy
As a precursor to the empirical analysis, we describe the data sources and relevant
characteristics of our sample, and provide a randomization balance check.
A. Data
We tracked clients for roughly three years from when they entered the study. We
gathered information on household business activities, socioeconomic status, and
demographic characteristics at three points in time: shortly after they entered the
study (Survey 1), when they completed the experimental loan cycle (Survey 2), and
two years after the experiment ended (Survey 3).
regular contract leads clients to invest some loan money in short-term informal savings that provide negligible or
even negative returns.
7
To see this, suppose loan size b is increased with p1, p2 scaling proportionally until b − f > i min. Regular clients
can now invest b − f in the illiquid investment for a profit of (b − f )r − (b − f ), but this is still below grace period
clients’ profit of r × b − b.
8
Suppose the illiquid investment is risky and succeeds with probability 1 − p, reflecting either project failure
or greater uncertainty as to the timing of returns. Grace period clients get a payoff of prb − pb − (1 − p)d if they
make the illiquid investment or a payoff of 0 if they make the liquid investment. They invest in the illiquid invest-
ment if and only if the return is high enough such that d < _
p
1 − p (rb − b), in which case they have higher profits
and also default more often. Note that risk attitudes and income-smoothing mechanisms are likely to be correlated.
Madajewicz (2011), for example, argues that client project choice and attitude to risk will vary with wealth.
9
In our model, this translates to only a fraction of clients who have access to a high-return investment.
2202
microfinance REVIEW
october 2202
2013
We use Survey 1, which was conducted an average of eight weeks after loan dis-
bursement, to check the balance between treatment and control groups along time-
invariant characteristics and to measure the fraction of clients who invest their loans
into new businesses. This survey also provides data on client characteristics that
we use in the heterogeneity analysis. The loan use module in Survey 2, which was
completed roughly one year after loan disbursement by 93 percent of clients, is used
to study differences in short-run investment behavior. Clients reported loan alloca-
tion across the following categories: business, human capital (health and school),
housing repair, food expenditure, savings, relending, and other. Finally, Survey 3
administered in 2010 (almost three years after loan disbursement) to 91 percent of
clients provides long-run data on household income and microenterprise profits and
assets for up to five household businesses. It also includes information on client
business practices.10
To study delinquency and default, we tracked client repayment behavior using
VFS administrative data in which repayment date and amount paid were recorded
by loan officers in client passbooks and then compiled into a centralized database.
Data are available through January 2010, by which date at least 52 weeks had passed
since the loan due date for all groups.
As a check on VFS administrative data, we also collected repayment data from a
separate logbook on meeting activities maintained by loan officers for the purpose
of our experiment. The logbook recorded date of meeting, number of clients present,
and names of clients who repaid at the meeting. We find similar default rates across
this measure and VFS administrative data (4.9 percent compared with 5.4 percent).
B. Client Characteristics
The majority (75 percent) of our sample are second-time VFS borrowers and the
remainder are first-time borrowers. Since VFS clients rarely drop out between the
first two loan cycles, the two groups have similar demographic characteristics.11
A key baseline variable of interest is business ownership. A large fraction of
India’s urban poor rely on (informal) household enterprises. The nationally rep-
resentative 2009–2010 Employment and Unemployment Survey (National Sample
Survey Office 2011) reports business ownership among urban households with a
female aged 20 to 60 as 57.6 percent and 60.7 percent for the poorest and second
poorest quartiles respectively (the subpopulation most closely corresponding to
the MFI client population). In urban Hyderabad, Banerjee et al. (2009) report that
roughly 50 percent of MFI borrowers had businesses of some kind. We construct a
baseline measure of microenterprise activity (“Has business”) using Survey 1 data
on the duration of existing household business activities to exclude businesses that
were formed after loan receipt. When asked without probing, slightly more than
10
Survey 2 was completed between January and November 2008; average time between Surveys 1 and 2 was
12 months. For Survey 3, the first wave (of 88 percent) clients were surveyed between April and July 2010, and the
second wave (which tracked temporarily out-of-town clients) was completed between October and November 2010.
11
Feigenberg, Field, and Pande (2012) find no VFS client dropout between the first two loan cycles, and that
VFS client demographics, such as income, home ownership, and home size, are comparable to those of clients with
similar MFIs operating in other Indian cities (Spandana in urban Hyderabad and SEWA in urban Ahmedabad).
However, consistent with cross-city differences in MFI penetration, VFS clients report relatively lower rates of bor-
rowing outside of the MFI (online Appendix Table 1, Feigenberg, Field, and Pande 2012).
2203
microfinance REVIEW
october 2203
2013
three-quarters of households reported some kind of microenterprise at the time

they entered the study. When business activities were probed more extensively in
Survey 3, virtually all households (97 percent) reported being engaged in some type
of business activity around the time they entered the experiment (“Has Business
(broad measure)”).12
The lower rate (78 percent) is most likely to correspond to business activity elic-
ited in other surveys. Still, business ownership in the VFS population is relatively
high, which likely reflects the fact that VFS, more than many MFIs, targets entrepre-
neurial households.13 In this respect, our study results most readily apply to the large
subset of MFIs that screen clients on loan purpose or client occupation.
In terms of employment diversification, roughly half the households report at
least one wage earner. The households also face significant risk: over the last month
62 percent report a shock to household income and 16 percent report having missed
days of work due to a household shock (birth, death, or exposure to heavy rain or
flood). Levels of chronic sickness—defined as having at least one household mem-
ber with an illness that lasted more than three months in the previous year—are
relatively high at 19 percent.
At the same time, access to savings and informal sources of credit or insurance
to finance shocks is relatively limited. Only 19 percent of households have a formal
savings account. Household ability to smooth shocks by liquidating assets is simi-
larly limited. The long-run survey probed respondents on liquidation costs of selling
business stock. On average households said that they could only retrieve 37 percent
of their inventory’s original value if they had to sell their stock in a one-day period.
Paralleling this, clients report a high rate of business closure: over 35 percent of
businesses that were active at baseline are reported as shut down three years later.
Roughly a third of these (11.5 percent of businesses) were closed due to household
member illness.
Our baseline survey provides two psychological measures relevant to entrepre-
neurial decision-making: risk attitudes and discount factors. Discount factor is
measured from client responses to a lottery game in which she was given a choice
between receiving 200 rupees now or receiving between 210 and 250 rupees in a
month. The average rupee amount between the point at which the client prefers the
money up front and the point at which she prefers to delay consumption is used to
calculate her discount rate. The mean discount rate is 18.9 percent. A client with dis-
count rate above the median is defined as impatient. Risk index is constructed from
client responses to eight questions regarding client willingness to choose a lottery
over a safe return under different levels of expected returns to the lottery. A client
with risk index above the median is defined as risk-loving.
12
The difference in reported rates of business activity across the baseline and follow-up surveys is due to addi-
tional effort we put into capturing microenterprise ventures and self-employment in the follow-up, which we believe
was underestimated at baseline. We observe very significant female participation in household business—roughly
80 percent of business owners report that the female client closely manages and can answer detailed questions about
at least one household business.
13
VFS rules require that loans be used for business, and loan officers ask for business descriptions when evaluat-
ing potential clients. Such targeting is reasonably widespread among South Asian MFIs.
2204
microfinance REVIEW
october 2204
2013
C. Randomization Check
Online Appendix Table 1 reports a randomization balance check using Survey 1

data. In panel A, which considers only time-invariant client characteristics, treat-
ment and control groups are imbalanced in only one out of 11 baseline characteris-
tics (married), and a joint test of significance (χ2 ) of mean differences demonstrates
overall balance. Balance checks (available from authors) show that Survey 2 and 3
data remain balanced on all covariates other than marriage.
Panel B reports client characteristics that were either potentially influenced by
treatment assignment or were only collected for a subset of clients due to change
in survey format. Only one panel B variable—risk-aversion—is imbalanced across
treatment arms. This variable is not included in the set of controls, but is used in the
heterogeneity analysis. We have, therefore, verified that adding risk-aversion to the
set of controls does not alter the magnitude or significance of our results and that all
empirical results are robust to excluding the one randomization strata with imbal-
anced risk attitudes (results available from authors).
III. The Effect of a Grace Period on Business Activity
We start by describing our specification and then report average treatment effects
for short-run investment decisions and long-run profits. Next, we use administra-
tive data on client default and survey data on client business behavior to examine
investment risk and long-run business practices. Finally, we examine heterogeneous
treatment effects and compute returns to capital.
A. Empirical Strategy
Randomization of contract-type implies that differences in average outcomes

across clients assigned to different contracts has a causal interpretation. Since no
client dropped out after assignment, Intent To Treat estimates are the average treat-
ment effects of being on a grace period contract. We, therefore, estimate for client
i in group g:
(1) y ig = βGg + Bg + δ Xig + ϵig ,
where y ig is the outcome of interest and Gg is an indicator variable that equals one
if the group was assigned to the grace period contract. All regressions control for
stratification batch (Bg ) and cluster standard errors within loan groups. We report
regression estimates both without and with the controls (Xig ) listed in panel A of
online Appendix Table 1 and loan officer fixed effects.
In online Appendix Table 2 we use group-level data collated from VFS transac-
tions records and group meeting data records (collected by loan officers) to ver-
ify compliance with experimental protocol. All clients in a loan group received
their loans on the same day, at which point their first repayment meeting date is
announced. Consistent with the grace period contract stipulating a period of eight
weeks before the first payment, groups that received the grace period contract made
2205
microfinance REVIEW
october 2205
2013
500
450
400
350 Business use
Nonbusiness use
300
Loan expenditure
250
200
150
100
50
0
Inventory and Equipment* Operating Home Human Money for Savings Food and Utilites, taxes,
raw materials* costs repairs capital relending durable and rent

consumption
Loan use category

(*Scaled by a factor of 10)
Figure 1. Category Wise Loan Use (in Rs) (Survey 2)
their first loan installment an average of 52 days after groups assigned to the regular
contract (column 1).
Column 2 shows that, once repayment starts, the average time lapsed between two
consecutive meetings is identical across the two contracts (14 days). Finally, the
average repayment meeting lasted 18 minutes and was not influenced by contract
type (column 3).
B. Effects on Short-Run Investment
Figure 1 shows category-wise spending of the VFS loan by our study clients. Over
91 percent of the clients spent at least some of their loan on business-related expen-
ditures, and on average, a client spent 83 percent of her loan on business-related
activities. Home repairs is the second largest category of loan expenditure, but only
6.2 percent of clients report spending on home repairs.
With respect to business expenditures, close to 70 percent of clients report spend-
ing on inventory and raw materials, which include the three most common expen-
ditures: saris, wood, and sewing materials. Notably, as brought out by case studies,
all of the above are relatively illiquid investments once they have been transformed.
In the face of demand shocks, it may take several months to realize the returns on
embroidered saris or pre-ordered goods such as tailored clothing. Consistent with
this, clients who are sari sellers and tailors report that they would be able to recover
an average of 34 percent and 10 percent respectively of the value of raw materials if
forced to liquidate in 24 hours.
Did receipt of a grace period loan influence client expenditure decisions? Online
Appendix Figure 1 plots the distribution of business spending for regular and grace
2206
microfinance REVIEW
october 2206
2013
period clients. There is more concentration of spending by grace period clients in

the right tail of the business spending distribution.14 Table 1 further investigates
loan use differences by estimating equation (1).15 The left-most column of the table
shows the outcome variable of the regression. Column 1 of the table reports the mean
of the outcome variable for the control group.16 The second column of Table 1 reports
the coefficient on the treatment variable in a regression without controls. The last
column shows the coefficient of the treatment variable in a regression with controls.
In the first row and second column of panel A we observe that, relative to regular
clients, the average grace period client invests roughly 6.0 percent (Rs 364.9) more
of her loan in her business. In rows 2 through 4 we divide total business spending
into inventory and raw materials, business equipment, and operating costs. While
the estimates are noisy, grace period clients predominantly shift loan use toward
inventory and raw materials.
Our expectation is that grace period clients finance additional investment out of
money that would otherwise be set aside for initial loan payments. In row 5 we
observe a significant decline in reported nonbusiness spending by grace period cli-
ents, with the decline driven by a reduction in spending on house repairs (which
accounts for 58.6 percent of the difference in nonbusiness spending, row 6) and not
savings (row 10). However, in-depth interviews with a random sample of six VFS
clients on saving practices suggest that purchasing construction materials is actu-
ally a form of informal savings that is considered safer than keeping cash on hand.
The majority (five) of clients reported incremental purchase of housing materials
(including bricks and bags of concrete) which they stored for up to a month prior
to construction.17 Housing materials were preferred to cash on hand both because
they are harder to steal, and because liquidation for personal consumption imposes
transactions costs on household members (or clients with self-control problems).
Meanwhile, compared to other investments, housing materials are very liquid (cli-
ents reported that unused materials could be readily liquidated) and highly divisible
(bags of concrete vary in size and bricks are sold individually). The spending pat-
terns are consistent with regular clients investing more of their loan in a zero-interest
safe asset.
In row 12 we examine grace period effects on business formation within six
months of loan receipt. New business formation is low: only 2.0 percent of regular
contract clients start new businesses within the six-month period surrounding loan
disbursal. However, the likelihood of starting a new business is almost thrice as
14
A Kolmogorov-Smirnov Test shows that we can reject equality of distributions at the 0.002 level.
15
The regressions include loan size controls (to account for loan amount) in panel A and the full set of controls in
panel B. When we estimate these specifications without loan size controls, standard errors for estimated coefficients
increase. The business spending measures increase in size and significance, while home repairs falls in absolute
size and become insignificant.
16
This means that the regular contract clients spent on average Rs 6,142.4 on business expenditures.
17
Our research manager conducted these interviews in July 2012. Clients were asked whether they purchase
materials for home repair incrementally and in advance of using them. If they answered yes, they were asked why
they preferred this to saving money and making a bulk purchase at the time of construction. All clients reported
long-term home improvement plans and all but one said that they purchase building materials incrementally as
money becomes available. No households had access to a formal savings account (other than a cooperative savings
account).
2207
microfinance REVIEW
october 2207
2013
Table 1—Impact of Grace Period on Loan Use and Business Formation
Coefficient on grace period dummy (SE)

Control group OLS OLS
mean (SD) (no controls) (with controls)
(1) (2) (3)
Panel A. Total business spending 6,142.4 364.9** 383.9**
(162.4) (180.1) (185.2)
Component-wise business spending
Inventory and raw materials 4,521.4 337.1 367.6
(226.3) (279.9) (272.8)
Business equipment 1,536.5 8.786 −14.4
(172.4) (234.1) (227.1)
Operating costs 84.46 19.01 30.75
(36.91) (48.37) (49.38)
Panel B. Total nonbusiness spending 1,149.1 −356.1** −371.6**

(149.1) (172.4) (178.7)
Component-wise nonbusiness spending
Home repairs 557.2 −208.8** −222.1**
(116) (105.1) (110.4)
Utilites, taxes, and rent 25.95 −8.214 −9.657
(15.66) (19.9) (20.66)
Human capital 237.9 −34.97 −33.06
(76.88) (90.26) (91.99)
Money for relending 197.6 −27.42 −30.13
(56.74) (70.61) (69.51)
Savings 131.6 −15.02 −10.75
(35.97) (47.12) (47.48)
Food and durable consumption 151 −91.79 −94.73
(76.21) (94.11) (97.86)
Panel C. New business 0.02 0.0268** 0.0258*

(0.00648) (0.0135) (0.0139)
Notes: The cells in columns 2 and 3 of the table present the coefficient estimate of OLS regressions which regress
loan use and business creation variables on the grace period coefficient, estimated without and with controls respec-
tively. All regressions include stratification fixed effects and loan size, and standard errors are clustered by loan
group. Column 3 regressions include all controls reported in panel A of online Appendix Table 1 and loan officer
fixed effects. Missing controls are set to zero and a dummy variable included for whether the variable is missing.
The dependent variables in reported regressions differ by row and are as follows: the rows under panel A include:
all business expenditures and then its components (inventory and raw materials; business equipment; and operating
costs). Panel B rows include all non-business expenditures and then its components (home repairs; utilities, taxes,
and rent; human capital; money for relending; savings; food and durable consumption).The dependent variable in
panel C is an indicator variable which equals one if the household reported having started a business up to 30 days
before or up to 180 days after loan disbursal. Data in panels A and B comes from Survey 2. The Loan Use mod-
ule of the survey was completed by all clients, hence the number of observations in all regressions is 845. Panel C
data come from Surveys 1 and 3, and the number of observations is also 845 (see the online Appendix for variable
construction details).
*** Significant at the 1 percent level.
** Significant at the 5 percent level.
* Significant at the 10 percent level.
2208
microfinance REVIEW
october 2208
2013
Distribution of profits for grace and nongrace period clients

0.32
0.30
0.28
0.26
0.24
0.22
Regular
0.20 Grace period
Density
0.18
0.16
0.14
0.12
0.10
0.08
0.06
0.04
0.02
0.00
Long-run profits
Figure 2. Distribution of Long-Run Profits Divided into Rs 500 Bins
Notes: Profits are top coded at Rs 12,000. Data come from Survey 3.
high among grace period clients.18 These differences in business formation dem-
onstrate treatment effects not only in the quantity of business expenditures but also
on the nature of business investment since new ventures presumably entail greater
risk. They also serve as a consistency check on the business spending results: since
business creation was measured independently of how a client reported loan expen-
ditures, it avoids the concern that receiving a grace period contract may influence
mental accounting regarding loan use but not actual expenditures.
C. Long-Run Effects
Microenterprise Growth.—Does this change in investment influence microenter-

prise activity in the long run? We measure long-run business profits in Survey 3
by asking about both business outcomes and household income with single survey
questions: “Can you please tell us the average weekly profit you have now or when
your business was last operational?” and “During the past 30 days, how much total
income did your household earn?”19
Figure 2 graphs the profit distribution for the two client groups. Relative to regular
clients, those on a grace period contract are less likely to report very low (or zero)
profits and more likely to report high profits.20 In Table 2 we examine this difference
18
For clients administered Survey 1 more than one month after loan disbursal but before the loan cycle ended,
this variable is measured close to business formation. For those administered Survey 1 less than four weeks after
loan disbursal, we also use retrospective data on business formation collected in Survey 3, making the indicator
more noisy. Note that Survey 1 timing is balanced across treatment arms, and our result is robust to excluding cli-
ents surveyed fewer than four weeks after loan disbursal. The online Appendix describes variable construction, and
online Appendix Figure 2 shows that new business formation reflects an increase in vendor business.
19
de Mel, McKenzie, and Woodruff (2009) provide evidence that simply asking about profits provides a more
accurate measure of small entrepreneur profits than detailed questions on revenues and expenses.
20
A Kolmogorov-Smirnov Test rejects equality at the 0.077 level.
2209
microfinance REVIEW
october 2209
2013
Table 2—Impact of Grace Period on Long-Run Profit, Income, and Capital
Average weekly profits log of monthly HH income Capital

OLS OLS OLS OLS OLS OLS
(no (with (no (with (no (with
controls) controls) controls) controls) controls) controls)
(1) (2) (3) (4) (5) (6)
Panel A. Full sample

Grace period 906.6** 902.9** 0.195** 0.199** 28,770.2** 35,733.1***
(373.8) (370.2) (0.0805) (0.0782) (11,291.0) (13,020.6)
Observations 752 752 749 749 766 766
Control mean 1,586.8 1,586.8 20,172.71 20,172.71 35,730.2 35,730.2
(121.8) (121.8) (55,972.25) (55,972.25) (5,056.0) (5,056.0)
Panel B. Top coded sample

Grace period 645.0*** 640.9*** 0.195** 0.202** 23,594.1*** 29,068.9***
(214.6) (208.1) (0.0801) (0.0778) (8,849.6) (9,432.4)
Observations 752 752 749 749 766 766
Control mean 1,579.3 1,579.3 18,110.65 18,110.65 35,535.9 35,535.9
(117.9) (117.9) (26,962.41) (26,962.41) (4,951.8) (4,951.8)
Panel C. Top coded sample and trimmed at 1 percent

Grace period 503.8*** 486.5*** 0.190** 0.199** 15,266.2** 19,010.0***
(182.8) (176.8) (0.0798) (0.0770) (6,825.5) (7,067.9)
Observations 748 748 744 744 761 761
Control mean 1,514.7 1,514.7 17,160.57 17,160.57 33,030.8 33,030.8
(102.7) (102.7) (23,571.94) (23,571.94) (4,238.4) (4,238.4)
Panel D. Top coded sample and trimmed at 5 percent

Grace period 440.5** 452.6** 0.198** 0.207*** 15,266.2** 19,010.0***
(175.9) (175.3) (0.0795) (0.0768) (6,825.5) (7,067.9)
Observations 747 747 743 743 761 761
Control mean 1,514.7 1,514.7 16,692.76 16,692.76 33,030.8 33,030.8
(102.7) (102.7) (21,739.62) (21,739.62) (4,238.4) (4,238.4)
Notes: The outcome variables are “Can you please tell us the average weekly profit you have now or when your
business was last operational?” (columns 1 and 2); “During the past 30 days, how much total income did your
household earn?” (columns 3 and 4). Columns 5 and 6 report the value (Rs) of raw materials and inventory plus
equipment across all businesses in operation at the time of the survey. All data comes from Survey 3. Variation in
number of observations for a given sample reflects missing data. The panel-wise sample is as follows: Panel A uses
the full sample. In panel B the the top 0.5 percent of the cumulative distribution of the dependent variable is top
coded to the 99.5th percentile value. Panels C and D use the top coded sample and exclude the top 1 percent and
5 percent of dependent variable respectively. We report OLS regressions which include stratification fixed effects
and standard errors are clustered by loan group. Regressions reported in even number columns include controls pre-
sented in panel A of online Appendix Table 1 and loan officer fixed effects. If a control variable is missing, its value
is set to zero and a dummy is included for whether the variable is missing.
in profit outcomes in a regression framework. To address the concern of noise in sur-

vey responses to questions that require a high level of aggregation, we also examine
top coded and trimmed specifications. In columns 1 and 2 of panel A we see that
grace period clients report 57.1 percent higher weekly profits and the difference is
statistically significant. In panels B through D we present additional specifications
to test the sensitivity of this result to outliers. We first top code the top 0.5 percent of
the cumulative distribution of profits to the value at the 99.5th percentile (panel B).
Next, we trim the top coded sample to drop the top 1 percent and 5 percent of values
2210
microfinance REVIEW
october 2210
2013
of the household profits distribution (panels C and D). While the estimates retain
significance, since much of the gain in profits for grace period clients occurs in the
right tail, the magnitude of the estimate falls as we trim (to 29.1 percent at 5 percent
trimming).
An alternative measure of business profitability is total household income, which
may be measured with less noise. Consistent with the profit distributions, online
Appendix Figure 3 shows a rightward shift of income among grace period clients.21
Columns 3 and 4 of Table 2 report the difference in a regression framework (again,
with the same top coding and trimming rules as with the profits variable). In
column 3 we see that household income is an estimated 19.5 percent higher for
grace period clients three years after loan disbursement (∼ two years after the loan
was due). The income results are highly robust across alternative specifications.
Reassuringly, the magnitudes of the income and profit estimates are consistent, with
our weekly profit estimate accounting for 92.2 percent of the estimated increase in
monthly household income.22
Finally, we examine an important measure of business size: business capital,
defined as the sum of raw materials and inventory and assets. The value of raw
materials and inventory is computed from survey questions on the value of materi-
als clients currently stock and which are used for production. We compute the value
of equipment from clients’ valuation of their durable assets that are used in busi-
ness. Online Appendix Figure 4 shows the distribution of business capital— once
again, we see a rightward shift in the distribution for grace period clients (relative
to regular clients). Columns 5 and 6 of Table 2 test the significance of this differ-
ence in a regression framework. The column 5 estimate for the full sample suggests
that microenterprises in grace period households are 81.0 percent larger in terms
of assets and inventory and raw materials. The effect on business capital falls to
46.2 percent in the top coded and trimmed sample (panel D).23
Entrepreneurial Risk-Taking.—Three years after entering the experiment, grace

period clients have larger and more profitable businesses. Next, we examine whether
default outcomes and reported business practices suggest that higher profits entailed
greater risk-taking.
The most direct measure of differences in investment risk is default rates, which
reflect income realizations in Year 1 of clients’ loan investment. Figure 3 graphs the
fraction of clients who had not repaid in full relative to the date of first installment.
The vertical bars indicate the loan due date and eight weeks after the loan was due.
Relative to regular contract clients a substantially lower fraction of grace period
clients had repaid in full two months after the due date.
21
A Kolmogorov-Smirnov Test rejects equality of distributions at the 0.034 level.
22
To calculate the increase in profits as a fraction of the increase in income, observe that column 3 shows
the control mean income is 20,172.71. The treatment effect is therefore 20,172.71 × 0.195 = 3,933.67. In col-
umn 1, the treatment effect on monthly profits is 906.6 × 4 = 3,626.4. Thus the increase in profits accounts for
3,626.4/3,933.67 = 92.2 percent.
23
Grace period households also report more workers relative to regular contract households (2.89 versus 2.53
workers), though the difference is statistically insignificant. The fact that scale of business operations adjusts more
rapidly than size of the microenterprise workforce is consistent with imperfect substitutability between outside and
in-family labor.
2211
microfinance REVIEW
october 2211
2013
0.8
Grace period
0.6
Regular
0.4
0.2
0 50 100 150 200 250 300 350 400 450 500 550 600
Days from rst meeting to nished repaying
Figure 3. Fraction of Clients Who Have Not Repaid in Full Relative to the Date of
First Installment
Notes: The vertical bars indicate the loan due date and eight weeks after the loan was due. The
figure is constructed using administrative data and loan officer data collected at group meetings.
Table 3 examines the statistical significance of these patterns. We consider client

default at three time intervals: 8, 24, and 52 weeks after loan due date (defined as the
date when the final installment was due). Throughout, we observe a robust differ-
ence in default patterns between regular and grace period clients. Grace period cli-
ents are, on average, six to nine percentage points more likely to default than regular
clients. Twenty-four weeks after the loan was due, 2 percent of the regular clients
and 9 percent of the grace period clients have failed to repay (column 2). Even after
one year, the experimental difference is roughly the same (column 3). Column 4
shows that, on average, grace period clients have roughly three times (Rs 220 versus
70) higher fraction of unpaid debt 52 weeks after their loan was due.
These patterns are consistent with differences in client investments causing grace
period clients to experience a higher fraction of low business income months dur-
ing the loan cycle which, in turn, increased default. A slightly different explanation
that is also consistent with our conceptual framework is that, independent of income
realizations, grace period clients had less incentive to liquidate assets in the event of
a low profit realization in order to repay the loan because doing so entailed higher
losses (both in terms of lower resale value and higher expected future returns on
their assets). Unfortunately, in the absence of monthly profit data during the experi-
mental loan cycle, we cannot directly tease these two stories apart. However, we gain
some insight into this from examining patterns of business activity reported in
Survey 3. As reported in column 1 of Table 4, grace period clients are less likely to
report a business closure between loan disbursal and the three-year follow-up:
38.6 percent of the regular clients but only 31.4 percent of grace period clients report
2212
microfinance REVIEW
october 2212
2013
Table 3—Impact of Grace Period on Default
Full loan not repaid Repayment history

Amount
outstanding Repaid at Made first
Within 8 Within 24 Within 52 within 52 least 50 half of loan
weeks of weeks of weeks of weeks of percent of repayments Made first
due date due date due date due date the loan on time payment
(1) (2) (3) (4) (5) (6) (7)
Panel A. (No controls)

Grace period 0.0901** 0.0696** 0.0614** 148.7* −0.0137 −0.00842 0.0288
(0.0349) (0.0280) (0.0251) (83.61) (0.0151) (0.0613) (0.0261)
Panel B. (With controls)

Grace period 0.0845** 0.0642** 0.0609** 149.0* −0.0156 −0.0246 0.0244
(0.0333) (0.0262) (0.0249) (83.55) (0.0159) (0.0534) (0.0240)
Observations 845 845 845 845 845 845 845
Control mean 0.0424 0.0212 0.0165 69.65 0.988 0.501 0.953
(0.0142) (0.0101) (0.00899) (40.15) (0.00774) (0.0427) (0.0231)
Notes: The outcome variables are default rates measured at increasing number of weeks after due date (columns
1–3); the outstanding balance on the loan by clients who had not repaid within 52 weeks of the due date (column 4).
The outstanding amount is defined as the loan amount plus the interest minus the 10 percent security deposit given
by clients prior to loan disbursal. Columns 5 and 6 report whether clients paid at least 50 percent of their loan bal-
ance and paid the first half of their payments on time (updated as recently as January 2010) and whether they were
able to make their first loan payment on time (column 7). Data from columns 1–7 comes from VFS administrative
data and from data collected at group meetings by loan officers. We report OLS regressions with stratification fixed
effects and standard errors are clustered by loan group. Panel B regressions include all controls presented in panel
A of online Appendix Table 1 and loan officer fixed effects. If a control variable is missing, its value is set to zero
and a dummy is included for whether the variable is missing.
a business closure.24 They also experience higher within-business variance of profits

(column 2), defined as the difference in reported profits in months of high and low
profits (averaged across all household businesses). Relative to regular clients, the
average difference in profits between high and low months is over Rs 600 higher for
grace period clients. This combination of results suggests that grace period clients
were less inclined to shut down businesses when short-run profits were low or nega-
tive, either because this occurred during the grace period or because liquidation is
costlier due to either the nature of assets they hold or their expectations for long-run
returns. Consistent with this, grace period clients are also less likely to report having
ever sold goods or services at a discount in order to meet loan repayment obligations
(column 3), though the result is sensitive to the inclusion of controls.
An alternative explanation for the default result is habit formation: a grace period
may prevent clients from acquiring regular payment habits or, by leading them to
believe that prompt payment has fewer consequences, it may increase strategic
default. However, differences in habit-formation would presumably be starkest at the
onset of regular repayment when grace period clients have just had two months
24
We constructed an alternative measure of business closure from an open-ended survey question that asked
households to report changes in each business they had operated since loan disbursement. We constructed a dummy
variable indicating whether a household reported having closed its business. This measure of business closure yields
a similar effect (−0.04) which is significant at the 5 percent level.
2213
microfinance REVIEW
october 2213
2013
Table 4—Impact of Grace Period on Business Size and Business Behavior
Average differ- Sold goods

ence in profits or services Customers Number of
between high- at a discount Customers pre-order goods and
Business and low-profit to make loan buy goods services
closure months payment on credit or service provided
(1) (2) (3) (4) (5) (6)
Panel A. (No controls)
Grace period −0.0718** 686.6* −0.0232* 0.0972** 0.0989*** 5.543**
(0.0324) (375.7) (0.0128) (0.0373) (0.0356) (2.467)
Panel B. (With controls)

Grace period −0.0669** 713.9* −0.0166 0.113*** 0.107*** 6.051**
(0.0334) (396.6) (0.0122) (0.0371) (0.0358) (2.566)
Observations 766 751 764 769 769 769
Control mean 0.386 2,361.6 0.0468 0.432 0.395 5.607
(0.0243) (242.0) (0.0112) (0.0270) (0.0236) (0.475)
Notes: The outcome variables are: whether a client reported having closed a household business that was operating
at the time of loan disbursement (column 1); whether clients reported having sold their goods or services at a dis-
count to make a loan payment (column 2); the averaged difference in profits between high- and low-profit months
across all household businesses (we counted a seasonal business as a business currently in operation) (column 3);
whether clients report that they had customers who bought from them on credit (column 4) ; whether clients report
that they had customers who pre-ordered goods or services from them (column 5); and the number of types of goods
or services clients offered to their customers (column 6). Data in columns 1– 6 come from Survey 3. Variation in the
number of observations reflects missing outcome data. We report OLS regressions which include stratification fixed
effects and standard errors are clustered by loan group. Panel B regressions include all controls presented in panel
A of online Appendix Table 1 and loan officer fixed effects. If a control variable is missing, its value is set to zero
and a dummy is included for whether the variable is missing.
off. Likewise, strategic default should be concentrated early on in the loan cycle
when benefit of defaulting is highest. In contrast, columns 5–7 of Table 3 indicate
that grace period and regular clients were equally likely to make their first payment
and just as likely to repay at least half of the loan.25
Corroborating evidence on entrepreneurial risk comes from Survey 3, in which cli-
ents were asked about three types of risky business practices.26 First, clients were asked
whether they sold to clients on credit, a practice which increases business scale but
without enforceable contracts. As is evident in Table 4, over 43 percent of regular cli-
ents reported in the affirmative, and this number is nine percentage points higher among
grace period clients (column 4). Another risky practice that arguably makes a business
more vulnerable to hold-up is allowing clients to pre-order items. Roughly 40 percent
25
The grace period could also restrict social networking among group members and thereby increase default by
lowering informal insurance. However, a group-level index of network ties between group members (constructed
using survey 2 data on social and financial interactions) shows no difference across contract types. Another concern
is that a grace period prevents monitoring of client activities by loan officers’ ability early on in the loan cycle.
However, we do not consider this to be an important channel since loan officers did not undertake monitoring activi-
ties during loan meetings or discuss clients’ business activities. It is also unlikely that the difference reflects loan
officer effects: Each loan officer serviced groups on both the regular and grace period contract (throughout, panel B
regressions include loan officer fixed effects).
26
Due to the occupation-specific nature of risk-taking, there were very few risk measures general enough to
apply to all or most businesses. Hence these were the only direct measures of risk that were included in Survey 3.
2214
microfinance REVIEW
october 2214
2013
of regular clients allow pre-ordering, and again we see a significant increase of approxi-
mately 10 percentage points among grace period clients (column 5). Finally, in col-
umn 6 we observe that grace period clients offer a wider array of business goods and
services, indicating that they are both expanding and diversifying business operations.
While these findings are consistent with the idea that grace period client busi-
nesses continued to include more illiquid investments, there is at least one other
possible explanation. Since the outcomes were measured three years after the loan,
we cannot identify whether risky business behavior in the long run is a direct conse-
quence of having a grace period or an indirect consequence of grace period clients
having larger and more profitable businesses in the long run. Likewise, while busi-
ness income is more variable for grace period clients in the long run, this may be
due to the level shift in business income, and does not necessarily mean that grace
period clients experience a higher incidence of critically low-income months.27
Importantly, this finding also suggests that a full welfare calculation for clients
would require knowledge of client ability to smooth consumption.28
Heterogeneity Analysis.—In Section I, we identified testable hypotheses regard-

ing which types of clients will be more likely to respond to a grace period contract
by increasing their investment in high-return but illiquid investments.
In Table 5 we examine these predictions. First, we hypothesized that there should
be larger treatment effects for clients for whom the risk reduction provided by a grace
period is particularly valuable, including those who are relatively risk averse and
those without a savings account. We see that while, on average, risk-loving clients
have higher profits, they do not benefit from a grace period contract (column 1).
In other words, consistent with the idea that the grace period reduces the risk
associated with illiquid investments, the grace period contract benefits relatively risk-
averse clients. Similarly, clients who have a savings account and therefore an
alternative means for addressing short-term liquidity shocks should benefit less
from access to a grace period contract. Consistent with this, we see that these clients
have higher profits but no marginal gain from access to a grace period contract (col-
umn 2), although the difference between clients with and without savings accounts
is large but not statistically significant.
Second, clients with short-term liquidity needs (proxied as having a chronically
sick household member) and impatient clients should benefit less since their likeli-
hood of responding to investment opportunities is lower. Columns 3 and 4 show
evidence consistent with this prediction, although the effect on impatient clients is
27
In terms of long-run profits, grace period defaulters look somewhat worse off after three years relative to
non-defaulters, with 14.6 percent lower profits. However, it is hard to infer much since we lack clear predictions
regarding the relationship between Year 1 default and profits in Year 3.
28
The lack of consumption and savings data precludes a full analysis of this issue. We have examined two partial
measures of consumption smoothing. First, as a summary measure of consumption the client survey conducted by
loan officers during loan repayment meetings asked if the client ate fish or meat in the previous day. We see no signifi-
cant difference across treatment and control clients. A second medium-term measure comes from a daily consump-
tion survey administered for seven weeks to a random subsample (134) of the clients who took out a subsequent loan.
Clients were surveyed every 48 hours on their income and expenditures (the surveying occurred roughly 15 months
after they entered the experimental loan cycle). We see treatment effects on business expenditures and continue to see
a decline in spending on house repairs. While the treatment group does report more savings, these estimates are very
noisy and we see no significant difference in the coefficient of variation in savings. Thus, taken together, these data
are consistent with treatment and control households facing similar consumption and savings volatility.
2215
microfinance REVIEW
october 2215
2013
Table 5—Heterogenous Profits
Monthly profit (in Rs)

Household No
Risk Savings member household Wage Heterogeneity
Characteristic loving account chronically ill Impatient business earner index
(1) (2) (3) (4) (5) (6) (7)
Panel A. Full sample; OLS (no controls)

Grace period 1,733.1** 1,097.1** 579.9* 1,547.5** 906.2** 1,428.5** 2,984.6**
(710.9) (518.2) (317.5) (718.8) (375.4) (647.7) (1,395.1)
Characteristic × grace −1,557.9* −1,420.3 −1,004.3* −1,208.8 −1,187.4** −1,243.2* −1,350.7*
period (795.4) (1,071.5) (512.9) (794.4) (568.9) (743.5) (715.2)
Characteristic 660.5** 1,247.6* 34.13 103.4 −1,366.1*** −543.5** 213.9
(284.1) (633.1) (384.7) (247.9) (284.6) (236.8) (213.4)
Treatment effect evaluated 175.2 −323.3 −424.3 338.6 −281.2 185.3 1,633.9
at characteristic = 1 SE (304.8) (822.6) (384.4) (286.9) (427.9) (1,633.9) (331.6)
Panel B. Full sample; OLS (with controls)
Grace period 1,543.6** 1,075.8* 531.5* 1,512.2** 900.6** 1,399.0** 2,913.6**
(623.8) (552.9) (295.8) (665.4) (373.4) (622.0) (1,293.9)
0
Characteristic × grace −1,243.5* −1,518.6 −831.1 −1,193.6* −1,195.8 −1,203.0* −1,310.5**
period (654.6) (1,184.5) (516.2) (700.0) (914.5) (702.8) (652.5)
0
Characteristic 512.0* 1,138.2* −134.7 −128.1 −472.9 −512.6* 124.5
(287.8) (634.0) (403.8) (354.1) (674.7) (275.4) (231.3)
at characteristic = 1 SE (311.5) (867.4) (420.1) (281.0) (989.0) (331.6) 663.4
Observations 713 683 489 721 752 751 752
Panel C. Top coded sample; OLS (no controls)

Grace period 1,207.7*** 690.3*** 589.3* 1,008.9*** 643.6*** 970.8*** 1,903.4***
(351.7) (252.4) (314.7) (346.0) (215.1) (304.6) (583.6)
Characteristic × grace −1,034.5** −815.5 −1,015.8** −649.7 −926.4* −806.6* −810.3**
period (478.6) (796.8) (509.2) (433.3) (484.6) (428.3) (318.0)
Characteristic 539.8** 1,269.5** 47.92 87.33 −1,495.9*** −554.6** 180.7
(245.6) (567.4) (379.4) (236.9) (238.9) (222.9) (196.5)
at characteristic = 1 SE (291.3) (718.7) (383.1) (263.9) (433.7) (296.0) (306.0)
Panel D. Top coded sample; OLS (with controls)

Grace period 1,096.4*** 625.9** 534.9* 1,020.9*** 633.9*** 967.1*** 1,919.5***
(339.8) (248.4) (294.9) (334.2) (209.1) (297.8) (564.1)
0 0 0 0 0 0
Characteristic × grace −880.6* −754.3 −839.0 −736.1* −1,476.1* −833.2* −827.6***
period (464.2) (823.4) (513.7) (429.8) (789.3) (438.0) (306.2)
0 0 0 0 0 0 0
Characteristic 489.1* 1,140.8** −125.5 32.31 −646.9 −507.4** 160.5
(257.1) (568.7) (400.8) (255.0) (551.1) (252.9) (203.0)
at characteristic = 1 SE (279.3) (738.8) (418.4) (260.6) (696.8) (298.0) (296.6)
Observations 713 683 489 721 752 751 752
Mean of characteristic 0.515 0.190 0.190 0.494 0.0221 0.482 1.632
(0.500) (0.393) (0.393) (0.500) (0.147) (0.500) (0.937)
Notes: Each panel in this table reports (column-wise) the coefficients on the regression of profits on a set of characteristic (indi-
cated in column heading), a dummy for grace period, and the interaction of grace period dummy with the characteristic. The char-
acteristics are a dummy for above median riskiness in a risk index (see online Appendix) (column 1); whether the household has a
savings account (column 2); whether any household member had been ill for over 3 months in the 12 months preceding the survey
(column 3); a dummy for having a monthly discount rate above the median (column 4); a dummy for not having any household
business (column 5); and whether any household member earned wages at the time of the survey (column 6). The heterogeneity
index (column 7) is the total number of characteristics (from columns 1– 6) that a household possesses. For example, a household
in which the client was risk loving and impatient, there was at least one wage earner, and there was at least one savings account gets
a value of 4 in the heterogeneity index. Panels A and B consider the full sample while panels C and D consider the top coded sam-
ple, where the top 0.5 percent of the cumulative distribution of profits is top coded to the value of profits at the 99.5th percentile.
Panels A and C provide the estimates without controls and panels B and D with controls. All characteristics, except for household
business (described in the online Appendix), were measured at baseline in Survey 1. The row with “Treatment Effect Evaluated at
Characteristic = 1” reports the F-test for whether the sum of the level treatment effect and the treatment effect interacted with the
characteristic differs from 0. The standard error is reported below the point estimate. The row total “Mean of characteristic”: reports
the mean of the characteristic presented at the top of the column. For example, 51.5 percent of the sample is classified as Risk loving
(column 1). We report OLS regressions which include stratification fixed effects, and standard errors are clustered by loan group.
Panel B regressions include all controls presented in panel A of online Appendix Table 1 and loan officer fixed effects. If a control
variable is missing, its value is set to zero and a dummy is included for whether the variable is missing.
*** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level.
2216
microfinance REVIEW
october 2216
2013
noisily estimated. Finally, clients with more business experience and skill (proxied
by having a business at baseline and having no wage earners in the household) should
benefit more.29 The estimates in columns 5 and 6 support this prediction.30
In all cases, our results are robust across the full sample and the sample where we
top code profits.
Thus, while not all clients are “responders,” treatment effects are distributed
across a wide fraction of the population and vary in magnitude in predictable ways
according to client characteristics. The largest average treatment effects we estimate
are on the subpopulation of risk-averse clients, which encompasses 48.5 percent
of our sample. The estimate in row 1 of column 1 implies a 92.0 percent increase
in profits among clients in this group.31 In Section IIID, we conduct a back-of the-
envelope exercise using estimated returns to capital to show that it is reasonable to
anticipate differences in profits of this magnitude after three years of compounding.
To summarize our heterogeneity analysis, we construct a simple index equal to
the number of “non-responder” traits a client possesses, and regress profits at end-
line on treatment and treatment interacted with this index of responsiveness. We
include all traits in columns 1 through 6 of Table 5 other than whether a household
member is chronically ill since health data are only available for 60 percent of the
sample. The estimates in column 7 suggest positive and significant treatment effects
among as few as 47 percent to as much as 82 percent of our sample.32 Accounting
for average profits in the corresponding subsamples, our treatment effects on the
treated imply an increase in profits of 20.9 percent for the 35 percent of the sample
with two non-responder traits, 70.9 percent for the 36 percent of the sample with
one non-responder trait, and 109.2 percent for the 10 percent of the sample with
zero non-responder traits.33
D. Returns to Capital
What do these results imply about returns to capital? Following de Mel,

McKenzie, and Woodruff (2008), we estimate a linear relationship between capital
and profits:
(2) PROFITSi = βCAPITALi + αHOURSi + ϵi .
We instrument total capital using the grace period treatment. Following de Mel,
McKenzie, and Woodruff (2008), we control for labor inputs by including hours
29
It is possible that the absence of wage earners increases responsiveness to the grace period because these
households have fewer alternative sources of income to address short-term liquidity shocks. However, as business
profits are significantly lower for households with wage earners, we think the absence of business expertise and
interest is likely to be the most important channel.
30
An important caveat is that only 3 percent of our clients do not have businesses.
31
The mean weekly profits for risk averse clients on the regular contract is Rs 1,313.81.
32
That is, the estimated treatment effect is large and highly significant for clients with an index value of 0 or
1, and reasonably large but insignificant for clients with an index value equal to 2, indicating significant treatment
effects for as few as 47 percent (clients with heterogeneity index of values 0 or 1) to as much as 82 percent of our
sample (clients with heterogeneity index values of 0, 1, or 2).
33
The mean weekly profits in the control are 1,742.29 (for zero non-responder traits), 1,540.742 (for one non-
responder trait), and 1,346.67 (for two non-responder traits).
2217
microfinance REVIEW
october 2217
2013
worked as a covariate.34 Online Appendix Table 3 reports the IV estimates. The

point estimates imply monthly returns of 13.0 percent for the full sample (column 2)
and 11 percent for the top coded sample (column 4). These estimates (together with
the corresponding 95 percent confidence interval which are 2.4 –19.7 percent for the
top coded estimate and between 1.0–26.0 percent for the full sample) are consistent
with existing estimates in the literature. For instance, monthly returns of 5.5 percent
are estimated by de Mel, McKenzie, and Woodruff (2008), while McKenzie and
Woodruff (2008) estimate returns of 20–33 percent per month for Mexican enter-
prises (with higher returns of over 70 percent for financially constrained firms).
Dupas and Robinson (2013) find implied median returns of capital of 5.9 percent
per month among Kenyan micro-entrepreneurs. Udry and Anagol (2006) report
average returns of 4 percent per month in the informal sector in urban Ghana and
Banerjee and Duflo (2012) estimate returns to capital for small and medium enter-
prises in India at 89 percent per month.35
Our heterogeneity analysis identified the subsample of risk-averse clients as exhib-
iting the largest change in profits. Using the same procedure as above, the returns to
capital for this group are between 12.0 percent and 15.7 percent per month.
As a consistency check, we undertake a simple accounting exercise to gauge
whether the capital accumulation rates implied by observed differences in invest-
ment are reasonable. Suppose the average regular and grace period clients invest
Rs 6,100 and Rs 6,500 respectively (as we observe from the control mean and row
1 coefficient estimate in column 1 of Table 1). Further, suppose that grace period
clients earn a net monthly return of (x + z) percent compared with a return of x
percent for regular clients. If all returns are reinvested, then the differential in capi-
tal stocks at endline three years later will be (1 + x + z)366,500 − (1 + x)366,100.
With x = 0.017 and z = 0.04, this yields a difference of Rs 23,000, which is com-
parable to the difference of Rs 23,600 that we estimate (column 5 of Table 2). More
generally, online Appendix Figure 5 shows that the observed capital stock differen-
tial is generated by a range of (reasonable) values for x and z, where we assume the
return differential persists for all three years.36 Furthermore, the required rates of
return are reasonably robust to alternative assumptions about rate of profit reinvest-
ment. For instance, if instead of reinvesting all profits in the business clients invested
only half, then the differential would have to be 3.4 percent (11.4 percent versus
34
Since de Mel, McKenzie, and Woodruff (2008) have a panel of observations on capital, profits, and hours
worked, they first estimate the coefficient on hours worked in the pre-intervention period, and then use that esti-
mated coefficient multiplied by post-intervention hours worked as an estimate of labor’s share of profits. Both
their approach and the approach used here rely on cross-sectional variation in hours worked to identify the causal
relationship between hours worked and profits.
35
One caveat is that our estimate is identified off the change in capital induced by the grace period contract, and,
according to our conceptual framework, marginal investments have a higher-than-average expected return. Thus, if
the composition of capital difference persists three years after loan disbursement then we are estimating the return
for the high-return capital (although that is also true for some of the existing estimates).
36
This is consistent with the fact that our evidence on differential investment composition is from the long-run
survey. In results not presented, we make the more conservative assumption that the return differential persists for
only three months. An initial return differential of 8 percent in the first three months followed by a return of 8 per-
cent for both grace period and regular clients yields a capital stock differential of Rs 26,000. Using as a benchmark
the returns estimated above, the required values of x and z are well within the interval of the estimates.
2218
microfinance REVIEW
october 2218
2013
8 percent) per month for grace period clients relative to regular clients to generate
the observed capital differential. 37
Finally, the grace period contract potentially encompasses two effects: a portfolio
effect which makes illiquid investments more viable and an income effect which
increases total repayment time by two months, making it easier for a client to accu-
mulate income needed for repayment. Up to now, we have solely focussed on the
portfolio effect.
The income effect is driven by grace period clients having a lower net-present
value of payments relative to regular clients. Assume a client receives a loan of size
b with a flat interest rate of 10 percent to be repaid fortnightly over a 44 -week period
starting either 2 or 10 weeks following loan disbursement. The client has access to
a perfectly liquid investment opportunity with monthly return on capital 1 + rL in
which she invests i. Each fortnight, the client pays the required loan payment and
reinvests remaining profits. We set b equal to the median loan size in our sample
(Rs 8,000) and set initial investment size equal to Rs 6,500 as observed in the data.
We assume that the remaining Rs 1,500 (8,000 − 6,500) are set aside to pay off the
first four installments. Using a return to capital of rL = 0.08, even if all returns are
reinvested, the endline capital stock differential will be Rs 10,000, which is roughly
half of the difference observed in the data. The subsequent monthly rate of return
would have to be over 10 percent to generate the observed capital differential. While
such a return is potentially plausible, it is much higher than the returns required when
we allowed that grace period clients have a higher return to capital. This, combined
with the facts that grace period clients exhibited higher default (which would not
follow from a pure income effect) and that in qualitative interviews, clients high-
lighted the grace period effect, lead us to emphasize the portfolio effect.
A different type of “income effect” arises from differences in default rates which
imply that, on average, grace period clients had Rs 149 more income in the form
of outstanding loan payments one year after the loan due date compared with regu-
lar clients. However, even if this money were invested with a monthly return of
8 percent, the resulting difference one year later would be less than 2 percent of
the observed capital stock differential (Rs 373). Hence, the difference in repayment
amounts can explain almost none of the observed differences in long-run profits.
IV. The Viability of Grace Period Contract
Given the combination of higher returns and increased default among grace period
clients, it is natural to wonder whether a grace period contract offered at a higher
interest rate is a viable loan product. Put differently, why do so few MFIs include a
grace period contract as part of their loan portfolio?38
37
If we only consider the sample of responders with a heterogeneity index equal to zero or one, then we need to
explain a larger capital difference (36,000 versus 23,600). However, returns to capital and the differential in initial
investment as computed from the loan use survey (Rs 900 using the top coded estimated) are also higher for this
group. Recall that for the full sample, we need a differential of 1.7 percent (5.7 percent versus 4 percent) in per
month returns to capital for grace period versus regular clients. For this responder subset we would need a differen-
tial of 2.2 percent (6.2 percent versus 4 percent).
38
The MFI-transparency database for 2012 shows that only 18 out of 144 MFIs (in 15 countries) offered a con-
tract with a length of greater than two months between disbursement and first repayment. The interest rate on these
2219
microfinance REVIEW
october 2219
2013
To examine whether adverse selection and moral hazard make higher interest rate
grace period loans unsustainable, we calibrate a model of MFI profits. We assume
clients differ in their contract preferences and risk of default, and self select across
loan contracts that differ in grace period and interest rate.
A. Model
Consider a regular loan contract with principal b paid to a borrower at t = 0.

Equal-sized loan payments of mreg are to be made at equal intervals from t = 1 to
t = T. Define the implied interest rate of this contract rreg as the interest rate r that
satisfies
T mreg
(3) b = ∑_. t
t=1 (1 + r)
Next consider a grace period contract in which payment of a possibly different fixed
amount mgp starts at t = 2 and ends at t = T + 1. The implied interest rate rgp is the
r that satisfies
T+1
mgp
(4) b = ∑ _. t
t=2 (1 + r)
Denote contract type as c = gp, reg. A one-to-one mapping exists between mc and rc.
Assuming full timely repayment, present-value of MFI revenue from the contracts is
T mreg
(5) PVreg(rreg) ≡ ∑ _t and
t=1 (1 + rk)
T+1
mgp
(6) PVgp(rgp) ≡ ∑ _ ,
t=2 (1 + rk)t
i
where rk is the cost of capital for the MFI. Let f reg ( · ) map the implied interest rate
of the regular contract into a fraction of total present value of loan paid by client
i. For any interest rate she faces, we can compute the fraction of the loan she will
repay without knowledge of loan size.39 Formally, in terms of realized repayment
stream m1, …, ∞,
∞
mt
∑ _ t
i
(1 + rc)
t=1
_
f reg(rreg) ≡ T .
mreg
∑_ t
t=1 (1 + rc)
2220
microfinance REVIEW
october 2220
2013
grace period contracts was higher (55 percent versus 48 percent; the average APR charged for a two-month grace
period (only offered by six MFIs) is 58 percent).
39
We take the fraction of loan to be repaid as non-random, but the analysis goes through even if f ireg( ⋅ ) is instead
the expected fraction of loan conditional on all client information known by MFI.
2221
microfinance REVIEW
october 2221
2013
We define f gpi
( ⋅ ) analogously for the grace period contract. To incorporate the notion
of moral hazard, we assume f ci ′( ⋅ ) ≤ 0, such that a higher interest rate reduces
repayment.
Let u ci (r) denote client utility from selecting contract c with interest rate r.
Normalize the outside option to zero. Expected profit per client from offering only
contract c is
πc(rc) ≡ E[ f ic(rc)PVc − B | ui c(rc) ≥ 0

],
where the expectation is taken over the distribution of clients.

If, instead, the MFI offers both contracts, then expected MFI profit per client is40
(7) πgp, reg(rgp, rreg)
≡ E[ f igp(rgp)PVgp − B | ui gp(rgp) ≥ 0 ≥ ui reg(rreg) ] × Pri[ u gp(rgp) ≥ 0 ≥i u

reg(rreg) ]
+ E[ fi gp(rgp)PVgp − B | ui gp(rgp) ≥ iu reg(rreg) ≥ 0 ]
× Pr[ui gp(rgp) ≥ u
i
reg(rreg) ≥ 0]
+ E[ fi reg(rreg)PVreg − B |iu reg(rreg) ≥ iu gp(rgp), 0 ] × Pri [ u reg(rreg) ≥i u gp(rgp) ]

.
The first term is profits from new clients who prefer the grace period contract to
no loan but will never choose the regular contract. Its value (positive or negative)
depends on the repayment rate of new clients. The second term is profit from exist-
ing clients who previously selected the regular contract but now prefer the grace
period contract and captures two separate effects. First, clients who select the grace
period contract may repay more or less compared with other existing clients. Second
if interest rates differ across contracts, then the standard adverse selection and moral
hazard effects apply. (For example, a higher interest rate may select riskier clients or
may cause clients to take on riskier projects.) The third term is profits from existing
clients who still prefer the regular contract (and again, depending on who opted for
the grace period contract, repayment rate for this pool may be higher or lower than
the original regular contract client pool).
Following the standard zero MFI profit assumption we restrict πreg = 0, assume
a unique interest rate r reg consistent with zero profits exists and set rreg equal to r reg.
∗ ∗
Next, we examine if there is an interest rate rgp such that πgp, reg(rgp , r∗reg) ≥ πreg(r ∗reg).
(To minimize clutter, we drop r∗reg from subsequent expressions and replace rgp
with r.)
The first term in equation (7) is profit from new clients who prefer the grace
period contract to no loan but who would never choose the regular contract. Our
experimental data cannot evaluate this expression as participation was conditional
2222
microfinance REVIEW
october 2222
2013
40
All probabilities Pr ( ⋅ ) are conditional on {u gp(rgp) ≥ 0 or u reg(rreg) ≥ 0}.
i i
2223
microfinance REVIEW
october 2223
2013
on a willingness to accept the regular contract. Hence, we begin by setting this term
to zero, and restrict our calculation to the change in profits from existing clients
when a grace period contract is introduced.
We denote profits from offering both contracts conditional on clients preferring
the regular contract to no loan as πgp, reg(r | ureg
i
≥ 0). In this scenario, the change in
41
profits from adding a grace period contract is
(8) πgp, reg(r | u reg ≥ 0) − πreg(rreg) = Pr[ugp (r) ≥ ureg ≥ 0]
i i i
( E[ f gp(r)PVgp(r) | u gp(r) ≥ u reg ≥ 0 ] − E[ f regPVreg | u gp(r) ≥ u reg ≥ 0

i i i i i i
×
] ).
See the Appendix for details. Equation (8) shows that to compute the change in
profit, we need to calculate the difference between amount repaid by grace period
clients who would select the grace period contract and amount repaid by clients who
prefer the grace period contract but receive the regular contract. Multiplying this by
the fraction of clients who prefer the grace period contract gives the total change in
profits per client.
B. Calibration
Our calibration examines whether, holding the regular contract interest rate fixed
at 17.5 percent, there is a grace period contract interest rate that makes a zero-profit
separating equilibrium viable. To identify client self-selection across contracts we
utilize data on clients’ contract preferences collected in Survey 3. Specifically, each
client was asked whether she (hypothetically) preferred a grace period to a regu-
lar contract at progressively increasing interest rate differentials.42 Forty percent
of clients are willing to pay an interest rate of 17.5 percent or higher for the grace
period (no client is willing to pay more than 59 percent).
Recall that the difference in profit between offering a grace period and regular
contract compared with offering only a regular contract is given by equation (8),
rewritten here:
(9) Pr[u igp(r) ≥ ui reg ≥ 0]

× ( E[ f gp | u gp(r) ≥ u reg ≥ 0 ] × PVgp(r) − E[ f reg | u gp(r) ≥ u reg ≥ 0 ] ×
i i i i i i
PVreg ).
Pr[ u igp(r) ≥ ui reg ≥ 0 ] is the fraction of clients who prefer the grace period
contract
to the regular contract when grace period contract interest rate is r. We compute
E[ f ireg | u igp(r) ≥ ui reg ≥ 0 ] using the fraction of present value of loan repaid by
clients
who received the regular contract but report preferring the grace period contract.
2224
microfinance REVIEW
october 2224
2013
41
All probabilities Pr ( ⋅ ) are conditional on uireg ≥ 0.
42
Our protocol was as follows: if the client stated a preference for the grace period contract at 17.5 percent, we
progressively increased the hypothetical interest rate on the grace period contract until she reported preferring the
regular contract.
2225
microfinance REVIEW
october 2225
2013
80
Difference in profits per client 60

40
20
0
−20
−40
−60
−80
No selection and no moral hazard
−100 Selection and no moral hazard
Selection and moral hazard
−120
−140
15 20 25 30 35 40 45 50 55 60
Percent APR
Figure 4. Model Calibration Results
Notes: This figure shows calibrations, under different assumptions, of the change in per cli-
ent profits for an MFI which offers both a regular contract (at APR of 17.5 percent) and a grace
period contract as it varies interest rates. The graph is constructed using data from Survey 3.
First, consider the case where higher interest rates introduce no additional adverse
selection or moral hazard. That is, we hold fixed the repayment rates as the interest
rate rises, and alter the fraction of clients who prefer the grace period contract at that
interest rate using their stated willingness to pay. Hence, in equation (9) only
the term Pr[ u gp
i
(r) ≥ ui reg ≥ 0 ] varies. The solid line in Figure 4 shows that higher
default by grace period clients (as documented in the experiment at 17.5 percent)
along with changing client selection across contracts at higher interest rates implies
that the interest rate needs to rise to at least 25 percent for the MFI to recover its
losses.
Next, we allow for either adverse or advantageous selection as the interest rate
rises. Specifically, at every higher interest rate we adjust both the fraction of cli-
ents choosing each contract and the fraction of present value repaid by these clients
using data on their default behavior during the experiment.43 The long dash line
in Figure 4 shows that selection, on average, is adverse such that the MFI must
increase the grace period contract interest rate to 38 percent. In other words, a sepa-
rating equilibrium exists, though at an interest rate that is substantially higher than
current rates.
Finally, we allow for moral hazard. In particular, we parameterize moral hazard
using the “elasticity of the repayment rate with respect to the interest rate” (moral
hazard elasticity for short) defined as the percentage fall in the repayment rate due to
43
Note that f igp does not change with r as we are assuming no moral hazard with respect to the interest rate—that
is, a higher interest rate does not change repayment behavior. Under this assumption, we can use the repayment data
for clients who received the grace period contract and report preferring the grace period contract to calculate the
fraction of the present value of the loan that they repaid.
2226
microfinance REVIEW
october 2226
2013
i
a one-percentage-point rise in the interest rate and allow f gp (r) to vary with the inter-
est rate of the grace period contract r. No separating equilibrium exists beyond an
elasticity of 0.46, which is significantly lower than what has been estimated experi-
mentally in other settings (Karlan and Zinman 2009).44
We conclude that a separating equilibrium in which both contracts are offered is
unlikely to exist in many environments, both because moral hazard is likely to sur-
pass 0.46 and because adverse selection may be exacerbated by entry of new clients
who were not willing to take a regular contract but are willing to take up the grace
period contract at interest rates at or above 38 percent. To gauge the response on
this margin, we surveyed 13 potential clients in an existing VFS neighborhood who
said that they would not take out a regular VFS loan and measured their willingness
to pay for a grace period contract.45 Four of the clients stated a willingness to pay
above the MFI’s break-even interest rate of 38 percent for a grace period loan. This
suggests that this source of selection is likely to be important (though we cannot
know ex ante whether it will be adverse or advantageous).
Finally, even if moral hazard and selection effects are moderate enough that
a grace period contract is feasible, MFIs, particularly in South Asia, may find it
hard to overcome regulatory and political pressures (and possibly public outcry)
in order to charge an interest rate as high as 38 percent. Thus, in a wide variety of
environments only offering a regular contract may be constrained efficient for an
MFI.
V. Conclusion
Mirroring findings in the literature (de Mel, McKenzie, and Woodruff 2008;
McKenzie and Woodruff 2008; Dupas and Robinson 2013), we identify high returns
to capital for small entrepreneurs in a developing-country setting. Our evidence sup-
ports the view that liquidity constraints limit small entrepreneurs from exploiting
these high returns. While much of the literature has focused on access to credit
(Banerjee et al. 2009; Karlan and Zinman 2011), to our knowledge, we are the first
to document experimentally the interaction between the nature of high-return invest-
ments available to the poor and microfinance contract flexibility.
44
As a first benchmark we compute the elasticity required to make the grace period contract unprofitable when
there is moral hazard but no adverse selection. At this elasticity, the additional interest payments from raising the
interest rate are exactly offset by the fall in revenue from moral hazard, which occurs at an elasticity of 0.80. This
elasticity provides a natural upper bound for moral hazard in existing microfinance contracts. This exercise also
demonstrates the importance of adverse selection in our sample. With adverse selection, the highest the moral haz-
ard elasticity can be without making the grace period contract unsustainable is 0.46, while without adverse selec-
tion, it is 0.80. The difference of 0.34 provides a measure of the extent of adverse selection in our sample. A second
crude benchmark comes from Karlan and Zinman (2009) who find that a 1 percentage point decrease in the future
interest rate decreases default by 4 percent. They do not compute repayment rates. Under the strong assumption that
our estimates that defaulting decreases the rate of repayment in the treatment group by 40 percent can be extrapo-
lated, it follows that 1 percentage point increase in the interest rate will lead to a moral hazard induced decrease
in repayment of 0.40 × 4 percent = 1.6 percent. With the caveats on estimate extrapolation across contexts, our
estimate of the moral hazard necessary to make the grace period contract unprofitable (0.34 − 0.80 depending on
the level of selection) is within the bounds of the moral hazard found in Karlan and Zinman (2009).
45
Consistent with VFS lending criteria, potential clients were screened by loan officers on age, marital status,
and household income. Roughly 92 percent of respondents reported a household business. The demographic and
preference results are available from authors.
2227
microfinance REVIEW
october 2227
2013
Our calibrations suggest that MFIs may not offer a higher interest rate grace
period contract because of the associated adverse selection and moral hazard con-
cerns. An open question, thus, is whether subsidizing MFIs who offer grace period
contracts can encourage higher return but riskier investments among the poor. A
complete answer will require knowledge of both the welfare gain to clients and
the total subsidy cost to the government. Our simulation exercise provides one
estimate of the second component. A subsidy of Rs 150 per client would make
the MFI indifferent between only offering the regular contract and offering both the
regular and grace period contracts at the baseline 17.5 percent interest rate.
Calculating client welfare gains is more difficult. Ideally, we need the differential
between consumption in the treatment and control groups from the beginning of
the loan and a client utility function. Instead, we only observe the differential in
profits at one point three years after the introduction of the loan. Under the heroic
assumptions that (i) clients are risk neutral, (ii) from three years past disburse-
ment onwards the difference in consumption is equal to the difference in prof-
its between grace period and regular contract clients, and that (iii) prior to three
years the difference in consumption is zero, the implied social rate is a return of
178 percent per year.46
While much more work is needed to obtain robust estimates, this back of the
envelope calculation suggests that carefully targeted subsidies may well allow
microfinance to alleviate credit constraints among the poor in a manner that encour-
ages profitable investments among the poor. Related to this, we note that the typical
small-business loan contract in rich countries is often significantly more flexible
than a typical MFI contract and often subsidized. One such example are Small
Business Administration loans in the United States which are subsidized, have rela-
tively flexible contract terms, and default rates between 13–15 percent (compared to
2–5 percent on typical MFI loans) (Glennon and Nigro 2005).47
More broadly, the results in this paper suggest that evaluating the economic impact
of debt contract design can provide valuable insights on entrepreneurial behavior and
help identify alternative methods of reducing liquidity constraints.
46
From Figure 4, at the baseline interest rate of 17.5 percent APR, the calibration implies that MFI profits
per client drop by Rs 150 after introducing the grace period. Using the top coded client profit estimates, we per-
form an exercise with client profits analogous to the calibration in Section IVB comparing clients who want the
grace period contract and received the grace period contract with clients who wanted the gp contract but received
the regular clients. We find a difference in client profits of Rs 287. The monthly interest rate r that sets 150
= 287 _ ( (1 + r)
1 _
36 +
1
37 + ⋯
(1 + r) ) _
1= 287
36 is 8.9 percent which is equivalent to a return of 178 percent
r(1 + r)
per year.
47
For instance, flexible repayment options are available on Small Business Administration (SBA) loans in the
United States, and typically negotiated on a loan-by-loan basis. Payments are typically via monthly installments
of principal and interest. There are no balloon payments, and borrowers may delay their first payment up to three
months with prior arrangement. For details, see for instance https://www.key.com/business/loans/sba-bank-loans.jsp.
One reason that US lending institutions may be willing to withstand higher default rates is that SBA loans are guar-
anteed by the SBA. In addition, the SBA sets interest rate caps for SBA loans.
2228
microfinance REVIEW
october 2228
2013
Appendix
In this Appendix, we derive equation (9). By definition of πgp, reg(rgp, rreg | u ireg ≥ 0)
and πreg(rreg), we have
(A1) πgp, reg(rgp, rreg | u ireg ≥ 0) − πreg(rreg)
= ( E[ f gp
i
(r)PVgp(r) | u gp
i
(r) ≥ ui reg ≥ 0 ] × Pr[i u gp(r) ≥i u reg ≥ 0 ]
+ E[ fi regPVreg | ui reg ≥ ui gp(r), 0 ] × Pr[i u reg ≥ i u gp(r), 0 ] )
− E[ fi reg(rreg)PVreg | ui reg(rreg) ≥ 0 ].
Decomposing the second term, we have
(A2) ( E[ f igp(r)PVgp(r) | u gpi (r) ≥ ui reg ≥ 0 ] × Pr[i u gp(r) ≥i u reg ≥ 0 ]

+ E[ fi reg PVreg | ui reg ≥ ui gp(r), 0 ] × Pr[i u reg ≥ i u gp(r), 0 ] )
− ( E[ f ireg PVreg | u gpi (r) ≥ ui reg ≥ 0 ] × Pr[i u gp(r) ≥i u reg ≥ 0 ]

+ E[ fi reg PVreg | ui reg ≥ ui gp(r), 0 ] × Pr[i u reg ≥ i u gp(r), 0 ] ).
And finally, canceling terms yields
(A3) Pr[ u igp(r) ≥ ui reg ≥ 0 ]

× ( E[ f gp(r)PVgp(r) | u gp(r) ≥ u reg ≥ 0 ] − E[ f reg PVreg | u gp(r) ≥ u reg ≥ 0 ] ).
i i i i i i
REFERENCES
Armendáriz, Beatriz, and Jonathan Morduch. 2005. The Economics of Microfinance. Cambridge,
MA: MIT Press.
Banerjee, Abhijit V., and Esther Duflo. 2012. “Do Firms Want to Borrow More? Testing Credit Con-
straints Using a Directed Lending Program.” Unpublished.
Banerjee, Abhijit, Esther Duflo, Rachel Glennerster, and Cynthia Kinnan. 2009. “The Miracle of
Microfinance? Evidence from a Randomized Evaluation.” Unpublished.
Barclay, Michael J., and Clifford W. Smith Jr. 1995. “The Maturity Structure of Corporate Debt.” Jour-
nal of Finance 50 (2): 609–31.
Barnea, Amir, Robert A. Haugen, and Lemma W. Senbet. 1980. “A Rationale for Debt Maturity Struc-
ture and Call Provisions in the Agency Theoretic Framework.” Journal of Finance 35 (5): 1223–34.
Brockman, Paul, Xiumin Martin, and Emre Unlu. 2010. “Executive Compensation and the Maturity
Structure of Corporate Debt.” Journal of Finance 65 (3): 1123–61.
Daley-Harris, Sam. 2006. State of the Microcredit Summit Campaign Report 2006. Washington, DC:
Microcredit Summit Campaign.
de Mel, Suresh, David McKenzie, and Christopher Woodruff. 2008. “Returns to Capital in Microen-
terprises: Evidence from a Field Experiment.” Quarterly Journal of Economics 123 (4): 1329–72.
2229
microfinance REVIEW
october 2229
2013
de Mel, Suresh, David J. McKenzie, and Christopher Woodruff. 2009. “Measuring Microenterprise
Profits: Must We Ask How the Sausage Is Made?” Journal of Development Economics 88 (1):
19–31.
de Mel, Suresh, David McKenzie, and Christopher Woodruff. 2012. “One-Time Transfers of Cash or
Capital Have Long-Lasting Effects on Microenterprises in Sri Lanka.” Science 335 (6071): 962–66.
Dupas, Pascaline, and Jonathan Robinson. 2013. “Savings Constraints and Microenterprise Develop-
ment: Evidence from a Field Experiment in Kenya.” American Economic Journal: Applied Econom-
ics 5 (1): 163–92.
Evans, David S., and Boyan Jovanovic. 1989. “An Estimated Model of Entrepreneurial Choice under
Liquidity Constraints.” Journal of Political Economy 97 (4): 808–27.
Feigenberg, Benjamin, Erica Field, and Rohini Pande. 2012. “The Economic Returns to Social Inter-
action: Experimental Evidence from Microfinance.” Unpublished.
Field, Erica, and Rohini Pande. 2008. “Repayment Frequency and Default in Microfinance: Evidence
from India.” Journal of the European Economic Association 6 (2-3): 501–09.
Field, Erica, Rohini Pande, John Papp, and Jeanette Park. 2012. “Repayment Flexibility Can Reduce
Financial Stress: A Randomized Control Trial with Microfinance Clients in India.” PloS ONE 7 (9):
e45679. doi:10.1371/journal.pone.0045679.
Field, Erica, Rohini Pande, John Papp, and Natalia Rigol. 2013. “Does the Classic Microfinance Model
Discourage Entrepreneurship Among the Poor? Experimental Evidence from India: Data- set.”
American Economic Review. http://dx.doi.org/10.1257/aer.103.6.2196.
Fischer, Greg. 2013. “Contract Structure, Risk-Sharing, and Investment Choice.” Econometrica 81 (3):
883–939.
Fischer, Greg, and Maitreesh Ghatak. 2010. “Repayment Frequency in Microfinance Contracts with
Present-Biased Borrowers.” London School of Economics and Political Science, Suntory and
Toyota International Centres for Economics and Related Disciplines, Discussion Paper EOPP 021.
Glennon, Dennis, and Peter Nigro. 2005. “Measuring the Default Risk of Small Business Loans: A
Survival Analysis Approach.” Journal of Money, Credit, and Banking 37 (5): 923–47.
Gonzalez, Adrian. 2010. “Is Microfinance Growing Too Fast?” Microfinance Information Exchange,
MIX Data Brief No.5, June 2010, http://www.themix.org/publications/mix-microfinance-
world/2010/06/microfinance-growing-too-fast.
Hurst, Erik, and Annamaria Lusardi. 2004. “Liquidity Constraints, Household Wealth, and Entrepre-
neurship.” Journal of Political Economy 112 (2): 319–47.
Jensen, Michael C., and William H. Meckling. 1976. “Theory of the Firm: Managerial Behavior,
Agency Costs and Ownership Structure.” Journal of Financial Economics 3 (4): 305–60.
Kaboski, Joseph P., and Robert M. Townsend. 2011. “A Structural Evaluation of Large-Scale Quasi-
experimental Microfinance Initiative.” Econometrica 79 (5): 1357–1406.
Karlan, Dean, and Jonathan Zinman. 2009. “Observing Unobservables: Identifying Information
Asymmetries with a Consumer Credit Field Experiment.” Econometrica 77 (6): 1993–2008.
Karlan, Dean, and Jonathan Zinman. 2011. “Microcredit in Theory and Practice: Using Randomized
Credit Scoring for Impact Evaluation.” Science 332 (6035): 1278–84.
Leland, Hayne E., and Klaus Bjerre Toft. 1996. “Optimal Capital Structure, Endogenous Bankruptcy,
and the Term Structure of Credit Spreads.” Journal of Finance 51 (3): 987–1019.
Madajewicz, Malgosia. 2011. “Joint Liability versus Individual Liability in Credit Contracts.” Journal
of Economic Behavior and Organization 77 (2): 107–23.
McIntosh, Craig. 2008. “Estimating Treatment Effects from Spatial Policy Experiments: An Applica-
tion to Ugandan Microfinance.” Review of Economics and Statistics 90 (1): 15–28.
McKenzie, David, and Christopher Woodruff. 2008. “Symposium on Access to Finance: Experimental
Evidence on Returns to Capital and Access to Finance in Mexico.” World Bank Economic Review
22 (3): 457–82.
Myers, Stewart C. 1977. “Determinants of Corporate Borrowing.” Journal of Financial Economics 5
(2): 147–75.
National Sample Survey Office. 2011. Employment and Unemployment Situation in India 2009–10.
NSS Report No. 537(66/10/1).
Tirole, Jean. 2005. The Theory of Corporate Finance. Princeton: Princeton University Press.
Udry, Christopher, and Santosh Anagol. 2006. “The Return to Capital in Ghana.” American Economic
Review 96 (2): 388–93.
This article has been cited by:
1. Erica Field, Seema Jayachandran, Rohini Pande, Natalia Rigol. 2016. Friendship at Work: Can Peer
Effects Catalyze Female Entrepreneurship?. American Economic Journal: Economic Policy 8:2, 125-153.
[Abstract] [View PDF article] [PDF with links]
2. Benjamin A. Olken. 2015. Promises and Perils of Pre-Analysis Plans. Journal of Economic Perspectives
29:3, 61-80. [Abstract] [View PDF article] [PDF with links]
3. Manuela Angelucci, Dean Karlan, Jonathan Zinman. 2015. Microcredit Impacts: Evidence from
a Randomized Microcredit Program Placement Experiment by Compartamos Banco. American
Economic Journal: Applied Economics 7:1, 151-182. [Abstract] [View PDF article] [PDF with links]
4. Abhijit Banerjee, Dean Karlan, Jonathan Zinman. 2015. Six Randomized Evaluations of Microcredit:
Introduction and Further Steps. American Economic Journal: Applied Economics 7:1, 1-21. [Abstract]
[View PDF article] [PDF with links]

Classic Microfinance Model

Uploaded by

Copyright:

Available Formats

Classic Microfinance Model

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Classic Microfinance Model

Uploaded by

Copyright:

Available Formats

American Economic Review 2013, 103(6): 2196–2226

Does the Classic Microfinance Model Discourage

Do the repayment requirements of the classic microfinance contract

Lending to entrepreneurs is a risky proposition in the best of cases. In developing

B. Should a Grace Period Influence Investment Behavior?

II. Data and Empirical Strategy

three-quarters of households reported some kind of microenterprise at the time

Online Appendix Table 1 reports a randomization balance check using Survey 1

III. The Effect of a Grace Period on Business Activity

Randomization of contract-type implies that differences in average outcomes

(1) y ig = βGg + Bg + δ Xig + ϵig ,

raw materials* costs repairs capital relending durable and rent

Loan use category

Figure 1. Category Wise Loan Use (in Rs) (Survey 2)

B. Effects on Short-Run Investment

period clients. There is more concentration of spending by grace period clients in

Table 1—Impact of Grace Period on Loan Use and Business Formation

Coefficient on grace period dummy (SE)

Panel B. Total nonbusiness spending 1,149.1 −356.1** −371.6**

Panel C. New business 0.02 0.0268** 0.0258*

Distribution of profits for grace and nongrace period clients

Figure 2. Distribution of Long-Run Profits Divided into Rs 500 Bins

Microenterprise Growth.—Does this change in investment influence microenter-

Table 2—Impact of Grace Period on Long-Run Profit, Income, and Capital

Average weekly profits log of monthly HH income Capital

Panel A. Full sample

Panel B. Top coded sample

Panel C. Top coded sample and trimmed at 1 percent

Panel D. Top coded sample and trimmed at 5 percent

in profit outcomes in a regression framework. To address the concern of noise in sur-

Entrepreneurial Risk-Taking.—Three years after entering the experiment, grace

Days from rst meeting to nished repaying

Table 3 examines the statistical significance of these patterns. We consider client

Table 3—Impact of Grace Period on Default

Full loan not repaid Repayment history

Panel A. (No controls)

Panel B. (With controls)

a business closure.24 They also experience higher within-business variance of profits

Table 4—Impact of Grace Period on Business Size and Business Behavior

Average differ- Sold goods

Panel B. (With controls)

Heterogeneity Analysis.—In Section I, we identified testable hypotheses regard-

Table 5—Heterogenous Profits

Monthly profit (in Rs)

Panel A. Full sample; OLS (no controls)

Panel C. Top coded sample; OLS (no controls)

Panel D. Top coded sample; OLS (with controls)

What do these results imply about returns to capital? Following de Mel,

(2) PROFITSi = βCAPITALi + αHOURSi + ϵi .

worked as a covariate.34 Online Appendix Table 3 reports the IV estimates. The

IV. The Viability of Grace Period Contract

Consider a regular loan contract with principal b paid to a borrower at t = 0.

πc(rc) ≡ E[ f ic(rc)PVc − B | ui c(rc) ≥ 0

where the expectation is taken over the distribution of clients.

(7) πgp, reg(rgp, rreg)

≡ E[ f igp(rgp)PVgp − B | ui gp(rgp) ≥ 0 ≥ ui reg(rreg) ] × Pri[ u gp(rgp) ≥ 0 ≥i u

+ E[ fi gp(rgp)PVgp − B | ui gp(rgp) ≥ iu reg(rreg) ≥ 0 ]

+ E[ fi reg(rreg)PVreg − B |iu reg(rreg) ≥ iu gp(rgp), 0 ] × Pri [ u reg(rreg) ≥i u gp(rgp) ]

( E[ f gp(r)PVgp(r) | u gp(r) ≥ u reg ≥ 0 ] − E[ f regPVreg | u gp(r) ≥ u reg ≥ 0

Panel B. Total nonbusiness spending 1,149.1 −356.1 −371.6