Borrower Identification vs. Product Design: Does "Who" or "What" Matter More?

Given the mixed results of recent randomized evaluations of microfinance, an open question is whether there are broad limits to the benefits of microloans or whether programs can be tailored in specific ways to maximize impact.  Two features of microfinance programs that may matter are targeting and product design.  A recent working paper by Pushkar Maitra, Sandip Mitra, Dilip Mookherjee, Alberto Motta and Sujata Visaria investigates the role of these features by studying a microfinance program they term TRAIL, or Trader Agent Intermediated Lending.

The paper compares the impacts of a traditional group-based lending microfinance model to a more innovative and targeted model in the context of smallholder farming in West Bengal.  The TRAIL model targets loans by incentivizing local traders to identify high potential borrowers for unsecured individual loans. The loans also have some innovative terms.  First the loan cycle matches the agricultural production cycle for potatoes.  Potatoes are the major cash crop in West Bengal.  They generate significantly higher incomes in the districts studied, but are more investment-intensive, requiring substantially higher working capital.  Another product innovation is that the repayment amounts for TRAIL loans are indexed to account for yield risk and price risk in potatoes.  That means farmers are at least partially insured against both the possibility of crop failure and the possibility of low prices for potatoes (if there is a bumper harvest, for instance).

These combined innovations in borrower identification and loan terms lead to measurable and significant gains in farm incomes and investment in potato cultivation.  The rate of return on potato cultivation implied by the study is quite high, between 70 and 115 percent, but potentially credit-constrained households do not invest more in potato cultivation due to the cost of inputs.  The analysis finds that TRAIL loans increased acreage devoted to potato cultivation by 20 percent and farm incomes by 18 percent. 

The improvements look substantial but we might ask whether it is borrower identification or the design of the lending product, or both, that matter for outcomes. The paper also examines an alternative lending model, the Group Based Lending or GBL model, which more closely follows traditional microfinance, packaged with the loan product innovations described above.  Potential borrowers chose to become members of groups that met twice a month and were required to meet a savings target before becoming eligible for group loans.  The loans offered in the GBL scheme were also index-insured and matched the agricultural farm season, features often absent from traditional microfinance lending.

In contrast to the TRAIL loans, GBL loans had no significant impact. Even with index insurance and lending terms that matched agricultural seasons, there was little apparent benefit to the group-based lending schemes in this context.  This suggests that the impact of  TRAIL may have been driven by improved borrower identification rather than improved product design. This is particularly surprising since there is good reason to believe that agricultural loans should be “better” for borrowers if they match agricultural income cycles. While innovation in product design was present in both models, only the TRAIL loans, given to high-potential borrowers as identified by traders, appeared to encourage cash crop cultivation and raise farmer incomes.

Product design and borrower selection may both have profound influences on the impacts of microcredit loans on borrower incomes and well-being.  This study shows evidence for the influence of borrower identification, but not product design, on the impacts of microcredit.  Borrower identification may be an important margin on which microfinance programs can be improved, as in Banerjee et al (2014), it is only top borrowers who see significant gains from the microcredit programs in the context they study.  To the extent that we see large gains with the TRAIL model, there may be hope for overcoming information asymmetries in the market for microloans by using well-informed local agents to identify potential borrowers.  However, as the paper notes, agent-led models may be subject to corruption and they point to the randomization of the loans following borrower identification – a feature of the experiment that would not necessarily be preserved in a full implementation – and objective criteria imposed by the program including maximum landholdings as important in limiting the amount of favoritism allowed for by the agent-intermediated lending program.

The jury is still out on product design, but as far as we can see in this paper, there is little evidence that two important features – indexing and loan terms – have meaningful influences on program impact.  This may come as a surprise to microfinance practitioners as these are two potentially very important product features that could have significant benefits depending on the extent of insurance offered.  Just as with other studies we have to be careful about assuming external validity of these findings.  What the study does remind us is that “improved products” don’t necessarily mean improved outcomes and therefore they deserve careful testing.