Experimental Design
We use a randomized controlled trial (RCT) to evaluate our research questions. Starting in 2006, we conducted three field experiments to understand repayment flexibility. In the first (2006 - 2007), clients were randomly assigned to either a weekly or monthly repayment schedule. In the second (2007 - 2008), clients were randomly assigned to either begin loan repayment 2 weeks or 2 months after loan disbursal. In the third (2008 - 2009), the first experiment was repeated with a more heterogeneous sample to measure the interaction between repayment flexibility and client characteristics. The random treatment assignment in the second and third round was strati fied by previous treatment assignments to ensure interpretability of the results. Here, we are interested in the long-term impact of the second intervention, the grace period intervention.
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 fi ve-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 (USD 90) to Rs 10,000 (USD 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. 85 groups were assigned to the regular VFS debt contract with repayment in fixed installments starting 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 meeting 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 e ffective interest rate on the loan. Treatment status was assigned within batches of 20 groups, determined by timing of group formation (the final batch was smaller with nine groups). No clients dropped out between randomization and loan disbursement.
After the 2008 second endline data collection, the 845 clients we re-surveyed in 2010 and 2012 to build a panel data set. Between March and August 2018, we conducted the last round of follow-up, collecting data for 729 of the original 845 clients. As part of the 2018 survey, we also collected biomarkers, health and business outcomes data for all children in the household and detailed education data for all children the respondent ever had. Seventy percent of households in the sample had children ages 0-16 at the time of the intervention and, in 2018, 73% of households still have children living in their household.