Skip to content

New study brings evidence on how big data can predict loan repayment

June 10, 2015

SustainabilityA recent paper adds evidence to the common knowledge that phone data hold valuable information on the payment profile of the holder. Researchers at Brown University used the telephone data of a Caribbean company and built a model to predict loan repayment. The approach compares the predictability (scoring) from the telephone data to a classic credit scoring. What caught our attention is that the model did not use payment information of phone bills, but other behavioral data such as the number of calls made. Overall the model using behavior showed a good predictability of loan repayment – a positive news for financial inclusion.

Data is not a bottleneck anymore. There was a time when predicting repayment capacity was limited by the amount of available information. Emphasis was placed on improving and interconnecting credit bureaus, and on group lending as a way to go around the lack of credit history of borrowers. Recent research has shown that other sources can provide valuable information, from payment data (utilities bills), to cash flow management (simply analyzing deposits’ movements), to social and behavorial information (using data available on social networks, for example).

The Brown study uses phone information, not so much for what it says about payment patterns but more for its behavioural clues. The goal is to predict the likelihood of default based on behavioral features derived from mobile usage. As widely documented, households are much more likely to have a phone than a bank account, especially in lower income countries. Behavioral information include a set of features, such as when the holder of the phone makes a top-up payment (is she prudent and makes the top-up before the credit expires?), call durations, intensity of phone usage, mobility, diversity of social connections.

This paper adds to the literature on how big data can serve financial inclusion. The challenge is moving away from data availability on payers to building solid predictive models.

No comments yet

What is on you mind after reading this post? Share your thoughts with us!

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s