Can Digital Savings Reduce Risks in Digital Credit?

10 May 2017
3 comments

When the financial inclusion community talks about innovative digital credit and savings products these days, it seems like we focus almost exclusively on the credit side.

This makes some sense. Credit products generate revenue for lenders, whereas small-balance savings accounts are often money-losing products, and overall we have seen relatively small balances kept on these accounts. But recent data from an experiment in Tanzania, which used interactive SMS to teach M-Pawa customers how to save and borrow responsibly, suggests that savings should be a bigger part of the conversation. The findings indicate that we may be overlooking how savings accounts can support positive borrowing outcomes when they are offered on the same mobile wallets that customers use to obtain loans.

M-Pawa is a fully digital mobile money product of Vodacom and Commercial Bank of Africa that offers both savings and credit features. Its new SMS platform, run by Arifu, provides customized learning content based on consumers’ preferences and responses. The platform was rolled out to farmers in rural Tanzania as part TechnoServe’s Connected Farmers Alliance. CGAP and the Busara Centre for Behavioral Economics used the account activity of more than 21,000 farmers, from June 2014 to May 2016, to measure how Arifu’s interactive learning content impacted the farmers’ behavior.

Digital savings word cloud


One of the most interesting findings was that interactive SMS increased savings on M-Pawa. On average, Arifu users more than doubled their savings balances from Tsh2,673 ($1.20) to Tsh7,120 ($3.20) after interacting with Arifu’s learning content. Looking at the loan behavior of Arifu users, we find something more interesting still. Overall, there were improvements in Arifu users’ borrowing behaviors both in comparison to other users and to their own activity before engaging with learning content:

  • Arifu users took out loans that were Tsh1,666 ($0.75) larger than loans taken by non-Arifu users.

  • Arifu users had Tsh2,654 ($1.19) less in outstanding amounts and made payments almost three-and-a-half days earlier than non-Arifu users.

  • Arifu users took out loans that were Tsh1,017 ($0.46) larger on average than the loans they took out before interacting with the learning content.

  • Arifu users repaid loans five-and-a-half days sooner and made first payments that were on average Tsh1,730 ($0.77) larger than the payments they made before interacting with the learning content.

These behaviors — taking out larger loans, maintaining lower outstanding amounts, and repaying faster — mean that customers are more likely to borrow responsibly, improving the sustainability of lending.

Taking a bigger picture view, these findings offer two key lessons for digital credit:

  • There are business and consumer welfare benefits to using tools like interactive SMS to better understand and engage with consumers beyond just getting them to take out a loan with a few clicks on a USSD menu or app.

  • Digital credit models that offer a single channel for both credit and savings can leverage the complementary nature of saving and borrowing for increased consumer activity and responsible borrowing.

Financial services providers and mobile network operators have proved the concept of digital nano-credit. Now it is time to take things a step further, get to know our digital consumers better, and offer them the tools — such as linked savings accounts — to become better borrowers and manage their money across diverse products.

 

Comments

Submitted by Jeremiah L Grossman on
Hi Rafe, this is really interesting. A few questions: 1) How were farmers selected to participate or not participate in Arifu? Was it self-selected (in which case it could be correlation but not causation) or were control and experimental groups established? 2) For the figures cited above (e.g., "Arifu users took out loans that were Tsh1,666 ($0.75) larger than loans taken by non-Arifu users..."), was there an analysis of the extent to which these differences were statistically significant? Thanks and keep up the interesting work!

Submitted by Julian Dyer on
Hi Jeremiah, To answer your questions, the farmers did select into interaction with Arifu, which is why we focus on the difference in behaviour of Arifu users before and after their interaction with Arifu. We use the difference between Arifu users and non-Arifu users prior to any interaction with Arifu as a measure of this selection effect. The main points we stress in this article are those that compare pre-interaction behaviour with post-interaction behaviour. The point that "On average, Arifu users more than doubled their savings balances from Tsh2,673 ($1.20) to Tsh7,120 ($3.20) after interacting with Arifu’s learning content" is one we interpret as being closer to a trreatment effect. The first two points Rafe listed ("Arifu users took out loans that were Tsh1,666 ($0.75) larger" and "Arifu users had Tsh2,654 ($1.19) less in outstanding amounts and made payments almost three-and-a-half days earlier than non-Arifu users.") are based on these pre-interaction differences and should be interpreted as selection effects. The second two points ("Arifu users took out loans that were Tsh1,017 ($0.46) larger on average than the loans they took out before interacting with the learning content" and "Arifu users repaid loans five-and-a-half days sooner and made first payments that were on average Tsh1,730 ($0.77) larger than the payments they made before interacting with the learning content.") So in short, we found that there were some selection effects, but that controlling for these selection effects by looking at pre-post interaction comparisons, and bounding these selection effects by looking at the pre-treatment differences between users and non-users, and controlling non-parametrically for time trends, we still find plausibly causal effects of Arifu interaction on financial behaviour. All the results Rafe listed in the article are statistically significant at the ten percent level, and in the specification where we deal with outliers, the main results are significant at the five percent level. I hope this is helpful - if not, I'm more than happy to answer any further questions! Cheers, Julian

Submitted by joseph somba on
after going through the attincle ,am interested to know the following 1.mobile loan is function of what in user's mobile wallet? 2.how the eligible loan amount is reached for the mobile loan application 3.what are the predictive features used to determine the risk level of the customer applying for mobile loan? 4.what are the data used to come up with predictive model? 5.how is integration done btw the mobile wallets accounts and the bank accounts in core banking system? 6.what are the recovery measures for default customers? 7.how do banks and mobile operators( MNO) fetch profit from this digital loan ans savings products?

Add new comment