A New Model to Expand Credit Access Globally

Access to credit can provide a pathway out of poverty — and thus better health, education, and a more satisfying life. The problem is that there isn’t enough of it. In developing countries, more than 130 million businesses and 1.4 billion individuals lack access to credit. Even when it is available, the rules can be onerous and interest rates high — from 25% to over 100% for microfinance institutions, with research suggesting that women tend to pay higher rates. The informal market can be even more costly. There is reason for this. Poor consumers in credit-scarce economies are not easy to serve. These borrowers can be vulnerable to shocks, such as natural disasters or illness, and may not be able to provide documentation, like salary records, bank statements, or credit histories. All this makes it difficult to distinguish low- from high-risk borrowers. The simple fact is: Because it is expensive and complicated to provide credit in these markets, it isn’t being done. The corollary is that if the costs could be cut, access would improve. One opportunity is in underwriting. There have been efforts in this direction. One approach is to adapt technology and automation to transform traditional “expert-driven underwriting.” Rather than an in-person investigation of each potential borrower’s assets and liabilities, the idea is that digital tools, video interviews, and remote psychometric tests can provide enough information, at lower cost. In advanced economies, this approach is similar to that used by private banks when customers have bespoke financial situations, for instance in the case of high-net-worth individuals. In emerging markets, some microfinance institutions use this technique for small- and medium-sized business lending. Another approach is to digitize borrower data to expand the reach of “default model underwriting.” This technique uses large amounts of data to correlate borrower characteristics with known loan defaults, and then uses those correlations to make credit decisions for each person. Traditionally used by retail lenders in data-rich environments like the United States, fintechs have started digitizing borrower data in emerging markets, including airtime purchase behavior of cell phone owners, to assess income regularity. Both approaches have run into limits for the poorest potential borrowers. Automation can only go so far to replace human expertise. In addition, gathering the information required to create a high-quality default model can be expensive — and the effort can still miss the poorest borrowers, whose digital footprint is generally light. Even when data can be collected, it’s generally only enough to underwrite small, short-term loans and not the productive credit products that enable people to boost their incomes. What is needed, then, is a low-cost, reliable way for lenders to assess individuals’ ability to repay, even when there is little data to work with. We think we may have found it: predictive cash flow modeling (PCM), or the use of statistical models to forecast income and expenses for defined population segments. The principle is that this knowledge can tell lenders what size loan suits each segment, and it can also inform decisions on timing and risk. Here is how it could work. First, the modelers collect the data to understand the income and costs that are typical to a specific segment — cattle farmers in southern Nigeria, for example — through interviews and research. With the data in place, it is possible to estimate how much cash people in that segment are likely to be able to direct toward repaying a loan. Such calculations go a long way to defining how much credit to extend, and on what terms, for each segment. Let’s consider a hypothetical example of a family of five in Kenya. The family own half a hectare of land (the average size of smallholding in Kenya) and live on less than $2.19 per day. The family members grow potatoes, which are harvested in July and August, and maize, which is harvested in October and November. They also sell eggs from a few chickens and do casual labor. By collecting and evaluating data on the size of landholdings, farm prices, and regular costs — everything from seeds to school fees — it is possible to get a good picture of the cash flow dynamics of a cohort of similar families, and thus how much credit to offer. Such a model may show there is a 75% likelihood that a family that generates $100 a year in discretionary cash can repay a $50 loan — but only a 2% likelihood for a $200 loan. Lenders can then use this information to extend the right amount of credit given their desired risk profile. The model can also be designed to price in one-off events, such as a natural disaster or illness. Compiling data by population segment creates economies of scale, allowing lenders to serve borrowers who were previously too expensive to underwrite on an individual basis. Indeed, once the model is built, marginal underwriting costs are next to zero because no additional data collection is required. And unlike credit bureaus, who keep their models close, the PCM and data are accessible to lenders, allowing them to continuously improve product design and targeting. The basic concept of using probabilistic cash flow-based modeling is not entirely new. Financial institutions have long used probabilistic cash flow-based modeling for commercial real estate portfolios, where default rates are lower and so associated data is sparse. Even some microfinance institutions have experimented with a precursor to the idea we propose, for example, using a deterministic model of cash flow to estimate income and expenses for dairy farmers. The difference here is in the sophistication and the scale of the approach. Sophistication is in statistically projecting both income and expenses, based on limited information. Scale comes from the large unbanked population with little data, for whom this approach offers potential new opportunity. Building these cash flow models will likely be complex and expensive — and each market will need multiple models. The needs of, say, micro-entrepreneurs in Kolkata are different than those of their rural counterparts in the rest of West Bengal. Moreover, smaller microfinance institutions may lack the technical talent and geographic breadth required to build, maintain, and gradually improve cash flow models, or the geographic reach to build diversified loan portfolios. This is where philanthropic capital could step in, helping with data collection and providing technical support. In India, for example, the Bill and Melinda Gates Foundation, where one of us (Michael) works, will be testing this sort of statistical modeling and working with market participants to devise pilots. Putting PCM into practice will probably not be straightforward. Different markets could require additional testing to fulfill risk and regulatory requirements. Lenders may face challenges integrating PCM into their existing systems. Finally, while PCM sheds light onto borrowers’ ability to repay loans, it says nothing about their willingness to do so. Of course, PCM is no magic elixir that will solve all the challenges around providing financial services. Even so, we believe PCM may solve a complex problem — underwriting for the very poorest consumers — and as such, significantly expand credit access, at reasonable cost, to those who are now shut out. Note: The authors would like to thank Lukas Freiheyt for his contributions to this article.

A New Model to Expand Credit Access Globally

Summary.

Partner Center