The Hard Facts on 'Soft Data': Determining risk for loan applicants without credit scores is more than a numbers game

Wednesday
February 24
2016

The Hard Facts on ‘Soft Data’: Determining risk for loan applicants without credit scores is more than a numbers game

Anyone who has been turned down for a loan based solely on their credit rating knows the sting of being treated as just a number. We all want to be judged on our intents and aspirations, not condemned based on a few late credit card payments after last summer’s vacation.

As it turns out, judging based on a single number can shortchange lenders as well as borrowers. Research is showing that a more comprehensive picture based on “soft information” as well as numbers can serve as a firmer basis for judging risk of default.

Soft information can include anything from a picture of the applicant, to a narrative of their purpose for the loan, to the search history stored on their mobile phone. In the past it was too costly to take these factors into consideration when deciding on micro and small loans, but crowdsourcing, personality testing and mobile technology are now making it possible.

This is good news for small borrowers – those in rich countries with low credit scores and those in poor countries with no credit score. And it represents a promising new direction for those seeking to lend to them.

The Value of Soft Information

A number of studies over the last 20 years have looked at huge numbers of lenders and borrowers, using the physical distance between the two to examine which banks use soft information and how this changes their lending. The underlying idea is that, compared to a bank located on the other side of the country, a loan officer located in the borrower’s hometown is far more likely to base decisions on things like a trustworthy appearance and a convincing business plan. These studies have shown that banks are more successful at identifying quality borrowers when they can use such information – perhaps charging higher interest rates to an appealing borrower with an unappealing credit rating.

Lately, the rise of online peer-to-peer (P2P) lending websites, where private individuals bid on unsecured microloans requested by others, has given researchers the opportunity to look closely at how soft information works among a large sample of non-expert lenders.

Last year on this blog, we at the Harvard Kennedy School reported on a study by Asim Khwaja of Harvard’s Evidence for Policy Design (and coauthors) that looked at Prosper.com, the largest of the online P2P sites. The study found that individual P2P lenders using the information given to them through online loan applications predicted borrower default with 45 percent greater accuracy than the credit score. This was despite some of the information being unverified and voluntarily provided, like the borrower’s photograph, or their own description of why they needed the loan and what made them a good gamble. Furthermore, peers predicted default with a remarkable 87 percent of the accuracy of an expert econometrician who was given all information available on Prosper.com as well as further data on default rates that banks wouldn’t see.

Beyond Just Numbers

Shivani Siroya, founder and CEO of the alternative credit scoring and lending company InVenture, likens applying for a loan to applying to university, and a credit rating to a standardized test score. While college admission is based on a wealth of other information in addition to SAT scores, credit relies heavily on the one number. The microbusiness owner and the poor borrower seeking help in the face of an income shock don’t even have “SAT scores,” so to speak, but they do have “extracurricular activities.” And these days, data on our extracurriculars are automatically gathered by our phones.

Working in Kenya, a country with famously high penetration of mobile money, InVenture has made a smartphone app called Mkopo Rahisi – “easy loan” in Swahili – that offers no-strings-attached microloans in less than a minute based on the data stored on the applicant’s mobile phone.

The data used to build InVenture’s credit scores often involves information not traditionally thought of as financial indicators – including a customer’s communication data, browser history, social networking and map usage. Since launching in Kenya, InVenture has used this information to make 200,000 loans to more than 50,000 people, with repayment rates above 90 percent. Late last year, the company expanded to Tanzania and will next launch in Southeast Asia.

This technology is an exciting new development for the 2.5 billion people in the world who lack any financial identity on paper. It builds a formal financial identity based not on their borrowing history, but on the data they generate while living their daily lives. Siroya calls this “basing decisions on the person, rather than the system.”

But numbers still play a vital role in lending decisions that involve soft data. Sophisticated algorithms are an important part of counteracting the biases soft information can introduce into the loan approval process. For instance, research based on the Prosper.com website has shown that providing pictures and other soft information can lead to discrimination. One paper showed that black applicants were 25 to 35 percent less likely to receive funding than white applicants with similar credit profiles. Another paper showed that applicants whose beauty was rated above average were more likely to get a loan and pay a lower interest rate than average-looking applicants with the same credentials – even though beautiful borrowers showed substantially higher delinquency rates. InVenture’s app aims to overcome such errors and biases by basing its decisions on rich data rather than more subjective human judgments, gathering over 10,000 data points per user.

NextBillion - An Initiative of the William Davidson Institute at the University of Michigan

Articles.