Microcredit Made to Measure. Experimental Evidence from Rural Morocco

1. Grace period loan: A loan product is offered where borrowers only pay monthly interest during a five-month grace period and both interest and capital thereafter.
2. Tailored loan: A loan product is offered where the repayment schedule is split into three periods of equal length. The borrower and loan officer jointly decide on the monthly amount that the borrower has to repay in each of the three periods in order to more closely match the borrower’s expected cash flows (which may vary significantly over time). The repayment schedule can thus be decomposed into ‘low’ – ‘medium’ – ‘high’ repayment brackets (not necessarily in that order) with potentially substantial variation in the monthly repayment amount across these three brackets. The research team developed an Apache OpenOffice spreadsheet to help loan officers determine the cyclicality of participants’ expected cash flows.

We test the following:
1. Standard microcredit contracts lack a sufficiently long grace period. Entrepreneurs may have difficulties repaying loans immediately after disbursement because it takes time for investments to generate revenues.
2. The rigid and linear repayment schedule of the standard microcredit contract is not (sufficiently) relaxed by a generic grace period that is identical for all borrowers and reduces liquidity demands early in the loan cycle. Instead, a repayment schedule is required that matches the expected revenue flows of individual borrowers more closely throughout the loan cycle as a whole.
We have partnered with a leading Moroccan microfinance institution (Al Amana) to test these two hypotheses. We do so by randomly relaxing either of two constraints in individual-liability microcredit contracts. We test the first hypothesis by offering microcredit with a significantly longer grace period (five months instead of one). The availability of such a longer grace period may make projects with a high expected return, but that are initially illiquid, more attractive to borrowers. While prior evidence (Field, Pande, Papp, and Rigol, 2013) shows that the introduction of grace periods (of two months) may change the type of investments that borrowers select, our contribution is to also provide evidence on the broader impacts on borrowers and their households.
We test the second hypothesis by offering microcredit with a repayment schedule that is designed ex ante to more closely match the expected revenues of individual borrowers throughout the repayment period. The availability of such tailor-made loans may make projects with a high expected return, but with highly seasonal revenues (for instance due to natural or religious cycles), more attractive to borrowers. Unlike Barboni and Agarwal (2018) and Battaglia, Gulesci, and Madestam (2018), who both test the impact of ex-post repayment flexibility, in which borrowers are allowed to waive certain repayments during the loan cycle, our loans do not provide repayment flexibility after disbursement. Instead, we match repayments more closely with future (expected) revenue streams before disbursement. Our contribution is therefore to investigate how customizing microcredit more to the needs of individual borrowers (while maintaining ex post repayment discipline) affects loan uptake, investment decisions, and ultimately the welfare of those who take these loans.
To estimate the impact of relaxing these constraints on borrower welfare, we randomize individuals who willingly demonstrate a desire to take out a standard microcredit loan (by visiting a branch and applying for such a loan) and who are eligible for a standard loan. We randomize these individuals into either of the two new flexible loan offers or into a standard loan offer (the control group). By restricting our experiment to individuals willing to take a standard loan, we can evaluate the impact of the flexible loans on borrower welfare while netting out potential selection effects. If the new loan offers would instead have been made available to everyone, then the characteristics of the client pool would likely vary across the different loan offers.
A second important contribution is our focus on changes in borrower selection and repayment performance once tailor-made microcredit becomes available. The individual randomization will allow us to identify the types of borrowers who prefer the new loans versus the standard ones. In addition to this, we randomize loan-type specific information campaigns to stimulate take up at the village level. This allows us to measure differential selection into loan demand across treatment arms.
Overall, this integrated approach helps us to clarify the causal link from offering new loan contracts to borrowing and repayment behavior and to evaluate the effect of these new contracts on borrower welfare in the population that was initially interested in the standard loan offer.

2018-04-09

2021-06-30

The primary outcomes vary across the two parts of the experiment. In Part A, we focus on the profits and revenues derived from the economic activities of the clients as well as on their consumption and income. In Part B, we focus on the number of loans disbursed, repayment rates, and the portfolio at risk. In the second part of the study, we will rely on administrative data to measure loan take-up, amounts borrowed, reimbursement length and frequency, client retention, and portfolio at risk.

Revenues (sales) and profits: Total revenues (sales) and profits of the existing businesses over the 12 months before the survey. Income: Total income of the household over the 12 months before the survey (transfers, wage labor, salaried contracts, profits of businesses). Subjective income expectations: Expected profits from entrepreneurial activities over the next 12 months. Expected profits in positive and negative scenario as well as likelihood that profits will be above/below average expectation. Consumption: Per capita consumption of the household over the 12 months before survey (Food consumption (short term) and durables). Credit take-up: Take-up of flexible loans or standard loan. Repayment rate: Repayment status of the loan 30 days after the due date.

Our secondary outcomes are variables that may also be affected by the intervention but are not the core variables that determine whether the interventions have been successful. These include loan use, type and level of investments, time allocation (Part A), and reimbursement length and frequency, amount borrowed, characteristics of the pool of clients taking up loans, and client retention (Part B).

Type of investments: Type of investments since baseline survey (chosen from a pre-specified set). Level of investments: Total value of new investments since the baseline survey. Sum of all investments undertaken. Time allocation: Allocation of household time across self-employment, wage labor, and domestic activities in last seven days before the survey. Time spent on each activity over the last week for each member of the household. Loan amount: Take-up of flexible loan or standard loan. Repayment rate: Repayment status of the loan 30 days after the due date. Client characteristics: Gender, age, type of activity, repayment history (if any).

We implement the following interventions:
1. Grace period loan: A loan product is offered where borrowers only pay monthly interest during a five-month grace period and both interest and capital thereafter.
2. Tailored loan: A loan product is offered where the repayment schedule is split into three periods of equal length. The borrower and loan officer jointly decide on the monthly amount that the borrower has to repay in each of the three periods in order to more closely match the borrower’s expected cash flows (which may vary significantly over time). The repayment schedule can thus be decomposed into ‘low’ – ‘medium’ – ‘high’ repayment brackets (not necessarily in that order) with potentially substantial variation in the monthly repayment amount across these three brackets. The research team developed an Apache OpenOffice spreadsheet to help loan officers determine the cyclicality of participants’ expected cash flows.

The impact of both new loan products on loan demand, repayment quality, and poverty impacts will be evaluated using a two-pronged RCT in close cooperation with our implementing partner, microfinance institution (MFI) Al Amana. Al Amana operates the largest MFI branch network across Morocco: 560 branches of which 301 branches are based in rural areas. Each rural branch serves on average 800 clients who live in the villages (douars) that surround the branch. A total of 40 shortlisted rural branches participate in the study.
Part A: Impact of grace-period loans and tailored loans on borrower welfare (individual-level randomization). In Part A, we assess how grace-period loans and tailored loans affect borrower welfare. This intervention takes place in 40 of Al Amana’s rural branches. It was rolled out in a staggered fashion in three waves. The intervention is expected to last 12 weeks in each branch. On the basis of Al Amana’s administrative data, we estimate that a total of around 3,600 eligible potential borrowers will visit the participating branches to apply for a loan during this period. Part A is restricted to individuals (either renewing or first-time loan applicants) who demonstrate a desire to take out standard microcredit by visiting one of the participating branches to apply for such a loan. Participating individuals are randomly assigned to receive one of three loan offers: a tailored loan (treatment, I1), a grace period loan (treatment, I2), or the standard loan as currently offered by Al Amana (control, I0). We thus evaluate the effect of the new loans on the welfare of the self-selected population of individuals who are willing to take out a standard loan. This strategy allows us to minimize the differential loan take-up between the different treatment groups if the new loans were offered to the entire population. For example, the take-up could be larger for the flexible loans than for the standard loan making potential differences in end-line outcomes between those treatments hard to interpret (because the group of compliers would not be the same) We will be able to evaluate both the effect of the offer of flexible loans on clients’ welfare (intention to treat, ITT, effect) and the effect of the loans for those who take up the offer (local average treatment effect, LATE). Our assumption, based on the initial focus groups, is that the current microcredit product is sub-optimal for (some) existing and potential borrowers. The fact that borrowers need to start repaying right after disbursement and that repayment installments are fixed across the full loan cycle may push (some) borrowers into choosing specific types of investments. We measure the type of investments undertaken by the households, their profitability, as well as the subsequent effects on households’ living conditions and on their repayment capacity. Account managers at each of the 40 participating branches are obliged to use a pre-programmed randomization tool that indicates the type of loan product (tailored, grace period, or standard) that should be offered to each individual applicant. Validation checks have been included in this tool so that loan officers cannot overrule the randomization allocation. For each loan application the research team can observe the outcome of the randomization tool, the loan type that was offered to the client, and therefore any discrepancies that may exist between both. We have implemented stringent ex post checks to guarantee that credit officers fully respect the randomization rule. Loan officers have been trained and instructed to promote the (randomly) assigned loan type to each potential borrower with the aim of convincing them to take out a loan of that type.
Part B: Impact of grace-period loans and tailored loans on the demand for microcredit (village-level randomization). In Part B of the project, we evaluate the demand for the grace-period loans and tailored loans in a separate set of villages (within the catchment area of the same branches as in Part A). This allows us to measure the relative intensity of each constraint on microcredit take up. The entire population of these villages will be targeted. This intervention takes place in the same 40 branches participating in the individual research design (Part A) and will happen after that initial intervention. In each branch’s catchment area, we have identified – using Al Amana’s administrative data – 8 villages with a significant number of potential borrowers. These villages will randomly receive one of the following loan offers: a tailored loan (treatment, V2), a grace period loan (treatment, V3) or a standard loan as currently offered by Al Amana (control, V1). An Al Amana promotional team will visit each of the participating villages and conduct door-to-door discussions to provide information about the assigned loan type to the village inhabitants. Such a promotional visit will take place every two weeks in each participating village (the total number of visits per village will be four). Because the interventions include both access to a new loan offer and regular promotional campaigns, the remaining two villages will receive the current standard loans as per the usual promotional activities carried out by Al Amana (pure control, V0). This will allow us to disentangle the effect of the new loan offer from that of the associated marketing activities.
This research design will measure the effects of the new loans on microcredit take-up and repayment behavior by capturing the effects of the new loan offer on the existing pool of clients as well as on the new pool of clients the new loans may attract. We expect that the interventions increase take-up at the village level in part by attracting different pools of new clients. For example, the intervention on loans with a grace-period may attract riskier borrowers that have seasonal income flows (but that may generate higher returns).
To understand the process that takes place once the interventions at the village level (V1 and V2) are implemented, we will collect data on the characteristics of the participants in the information sessions and collect qualitative data four months after the village-level interventions start. Our goal is to understand changes in intermediary outcomes such as perceptions about credit, knowledge about Al Amana’s offer and the use of borrowed funds. This will be done through focus groups with randomly selected participants in the village-level interventions. These focus groups will take place about two months after the start of the village-level intervention.
For Part B of the analysis, our main final outcomes will be measured at both the village and the individual level: the number of loans disbursed, the average loan size, the proportion of loans taken by women, the average repayment ratio, portfolio at risk, and client retention. At the individual level, we will also use administrative data from Al Amana to describe the characteristics of borrowers in terms of age, type and size of activities, and repayment history (if any), that select into these new flexible loans. We will also compare them to the characteristics of clients in villages where the standard loan product is offered. To this end, we will carefully match Al Amana’s administrative data to each study participant.
Following this promotional campaign, we will monitor the loan disbursements at the participating branches for a period of 24 weeks (6 months) to determine the effect on demand for each of these three types of microcredit. A total of 320 (40x8) villages participate in Part B.
The pool of loan applicants from treatment villages V2 and V3 that approach Al Amana branches during Part B of the study may differ from those coming from control villages where Al Amana’s regular loan product is on offer. We intend to use Al Amana’s administrative data on key characteristics of loan applicants (age, gender, marital status, income, borrowing history, and type of entrepreneurial activity) to describe this population and compare it to the applicant pool coming from the control villages.

Part A of the study: randomization done at the branch by a computer. Part B of the study: randomization done in office by a computer.

We will randomize at the individual level for the first part of the study and at the village level for the second part of the study.

Yes

Part B will take place in eight villages in the catchment area of each of the 40 participating branches. This part thus involves 320 villages.

The first two waves of the baseline survey have provided us with preliminary estimates of the daily number of eligible clients that visit the branches and that consent to participate in the study. Since the variation across branches is substantial, we undertook power calculations for three scenarios where we assume total sample sizes of either 3,600; 3,000; or 2,400 clients.

Part A of the study: 1,200 individuals control, 1,200 individuals tailored loans (type 1), 1,200 individuals tailored loans (type 2). Part B will take place in eight villages in the catchment area of each of the 40 participating branches. This part thus involves 320 villages.

Part A: Individual-level randomization. For a total sample of 3,600 clients (1,200 being offered the standard loans; 1,200 the tailored loans; and 1,200 the grace-period loans), we will be able to detect increases of 4.2 percentage points (5 percent) in the share of households with a self-employment activity and of 5.3 pp (17 percent) in the share of households where at least one household member does day work. With imperfect compliance, effects among clients that will take up the new loan offers will need to be much larger in order to detect them. For example, for a total sample size of 3,000, standardized MDEs increase to 0.21 for a take-up rate of 0.6 and to 0.31 for a take-up rate of a 0.4. This compares to 0.13 in the ITT scenario. Given the observed take-up rates during the beginning of the experiment implementation, it is important to generate a large enough sample. Our goal is therefore to achieve a total sample size of 3,600 clients (1,200 in each arm). For continuous outcomes, we will be able to detect increases of 17 percent for profits, 18 percent for sales, and 14 percent for consumption. These magnitudes of the minimum detectable effects are equivalent to 0.11 of a standard deviation. Minimum detectable effects increase as the sample size decreases: standardized effects increase to 0.13 and 0.14 of a standard deviation for sample sizes of 3,000 clients (1,000 in each group) and 2,400 clients (800 in each group), respectively. Standardized detectable effects do not change significantly for this range of sample sizes. Given the large standard deviation of continuous outcomes, a standardized effect of 0.14 translates, however, into larger minimum detectable effects of 21 percent for profits, 22 percent for sales, and 17 percent for consumption. Part B will take place in eight villages in the catchment area of each of the 40 participating branches. This part thus involves 320 villages. A sample size of 80 villages per treatment group will allow for the detection of impacts on village-level take-up of at least 6.3 percentage points. We calculate these MDE using data from a related study in rural Morocco (Crépon, Devoto, Duflo, and Parienté, 2015) where the standard deviation of this outcome is 0.14.