Key message 3 – Screening borrowers using local information can lead to better targeting and improved welfare
In addition to the multiple restrictions in the traditional microfinance model, like regular meetings and saving obligations, which might affect the performance of these loans, the traditional model is also unable to screen out unproductive borrowers. Given their greater likelihood of defaulting, unproductive borrowers pay high-interest rates in the informal credit market. As a result, such borrowers have a strong incentive to apply for microcredit loans. Since microfinance loan officers lack fine-grained information about the risk and productivity of poor borrowers, as discussed in the previous section, they cannot screen these unproductive borrowers with sufficient precision.
Against this backdrop, an IGC study by Maitra et al. (2017) designed an alternative mechanism to leverage the information about borrower characteristics that exists within the local community, called agent-intermediated lending. The study was conducted with potato-growing farmers in the Indian state of West Bengal,where the authors tested two types of agent‑intermediated lending:
- Trader-agent intermediated lending (TRAIL): The agent is a private trader/shopkeeper with considerable experience in lending within the community.
- Gram panchayat-agent intermediated lending (GRAIL): The agent is appointed by the local government (village council).
The authors relaxed norms, such as those pertaining to meetings and saving, for both the intermediated lending schemes, and the duration of the loan cycle was increased to match that of the cropping cycle. The TRAIL and GRAIL schemes were tested against each other as well as against a more traditional group‑based lending (GBL)3 scheme.
The results showed that recipients in the TRAIL scheme were particularly successful in increasing potato cultivation and output. Their farm incomes increased significantly, without any off-setting decline in income from other sources (See Table 1). Furthermore, TRAIL increased the amount of land under potato cultivation and net profits of TRAIL borrowers.
Community members need to be incentivised to provide accurate information about borrowers’ creditworthiness
Eliciting reliable information from community members in a high-stake setting is not a straightforward task. In the Maitra et al. (2017) study, while the farm incomes of borrowers identified by TRAIL agents (economic agents) increased significantly, the outcomes of GRAIL (political agents) borrowers did not change. This was despite all features of both the TRAIL and GRAIL schemes being the same and identical financial incentives offered to the agents under both schemes.
In the case of Hussam at el. (2017), the study was designed to investigate whether community members distort their reports if they are provided with different incentives. It found that they did distort their reports when told that their information would influence the distribution of grants.
Both the studies showed that while community information can be quite valuable for targeting, its accuracy is sensitive to the conditions under which it is elicited. Both studies identified a natural tendency for respondents to favour their friends, fellow political affiliates, voters, and family members. The studies also showed that a variety of techniques, in particular small monetary payments for accuracy, eliciting reports in public rather than in private, and accessing information from non-political agents can all improve the accuracy of information.
The outcomes of borrowers in GRAIL and GBL did not change appreciably. This is despite the fact that all loans were provided at below-market-average interest rates, had repayment durations that matched local crop cycles, and included insurance against local yield and price shocks.
Another study by Hussam et al. (2017), similarly uses community information to identify productive entrepreneurs and assess credit risk. In an experiment in the Indian state of Maharashtra, the authors asked entrepreneurs to rank their peers on various metrics of business profitability, growth, and entrepreneur characteristics. The authors then assessed the validity of the information in an experiment where one‑third of these entrepreneurs randomly received a cash grant of about $100 for their businesses.
The study found that community members can identify high-return entrepreneurs with stunning accuracy. While the average marginal return to the grant was about 8% per month, entrepreneurs ranked in the top third of the community earned returns between 17–27%. Had the authors distributed their grants using community reports on predicted marginal returns instead of random assignment, they could have more than tripled the total return on investment. The findings suggest that community information is valuable above and beyond information that can be predicted through machine learning techniques using observable information about clients.
The idea that social networks – friends, family, colleagues, and local leaders – are a rich source of information has deep roots in development economics literature. However, there have been relatively few studies on exacting this information and using it. Both Maitra et al. (2017) and Hussam at el. (2017) show that the knowledge neighbours have about one another is highly predictive of clients’ marginal returns to capital and can be used for improving microcredit products. Both these studies also show that community information can lead to better targeting and increased profits, without affecting defaults.