Profit vs. equity: Lenders can raise both with alternative data

New research shows that using alternative data sources, such as social media, can help lenders expand their services to applicants without traditional payment histories and credit data.

One-quarter of the world's adults — including 7% in the U.S. — have no bank or mobile phone account that they can use to send and receive money or pay back small loans to prove they’re creditworthy. Even if lenders want to support these customers, loan companies can’t risk defaults from unproven borrowers. "Profit and equity are contradictory," says Assistant Professor of Information Systems Tian Lu. "This is a classic economic dilemma."

So, how can lenders expand their services to “thin file” applicants who don’t have the traditional payment histories and credit data to demonstrate that they’re low risk? In Asia, some lenders are turning to alternative data, digitized clues that reveal a person’s habits, movements, and interests. "Today, everyone has a certain amount of alternative data," Lu explains. "Most people shop online, use a cell phone, or have at least one social media account, and these are alternative data sources." While such information is sensitive and presents privacy concerns, Lu says lenders only use it when loan applicants give permission.

Is it worth the effort and cost for lenders to collect and use this information in loan decisions? Lu and two colleagues studied the question in Asia and found that alternative data can expand financial access to underserved borrowers. However, only one data source expanded financial access and fairness for customers while raising profits for the lending company.

The more the merrier

There’s a good reason for lenders to open their doors to a wider range of borrowers. Study after study shows that increasing financial inclusion raises economic growth. A year before Lu and his team’s data conducted their experiment with the micro-lender, researchers at the McKinsey Global Institute found that financial technology, which includes microloan websites like the one studied, could increase the GDP (gross domestic product) of emerging economies by as much as 6% (or $3.7 trillion) by 2025.

"Micro-lenders offer short-term products with lower loan-amount sizes," Lu notes. "The credit risk of this kind of customer pool is higher than what traditional financial institutions face because micro-lenders serve a much broader customer pool. This forced the micro-lending service providers to improve their financial risk assessment."

One way micro-lenders in Asia do this is by putting a pop-up window on the application asking would-be borrowers if they’d share their mobile, shopping, and social media data. If applicants say yes, the lenders can access but not store this information.

"From a customer perspective, this somehow damages their privacy and information security," Lu says. He wondered if it was necessary to collect all that data to increase access to loans for thin-file borrowers. He also wondered if the data improved profits for the micro-lender.

To find out, Lu and his colleagues convinced the lending website operator to randomly select 40% of all loan applicants and approve them without using any risk evaluation or selection strategies. Knowing who paid back loans and who didn’t, the researchers then ran simulations looking at a variety of borrower attributes under diverse financial risk assessment strategies.

Thick and thin

Those borrower characteristics included traditional loan application data such as income, age, education level, home ownership, and marital status. They also included alternative data comprised of shopping records from the country’s two largest online shopping platforms, social media statistics, and cellphone usage data, including GPS locational data.

The shopping information tracked a wide range of purchase categories: takeout food, online game cards, durable goods, books, alcohol, caffeine, tobacco, medicine, adult products, and more. It also examined ratios, such as money spent versus income or alcohol buys versus durable goods purchases.

On the social media side, the researchers looked at a borrower's presence on the country’s largest Twitter-lookalike micro-blogging site. Did the borrower have an account? If so, how many messages did the borrower post? How many followers and likes did this person receive? These and other details contributed to the loan applicant’s profile.

Cell phone data rounded out the information examined. The researchers looked at the number and type of apps used, contacts, call statistics, and even GPS data indicating whether the borrower spent more time near socially beneficial sites, like hospitals and schools, versus shopping and entertainment venues, like malls and movie theaters.

All told, the research team evaluated 117 borrower attributes, enough to thicken any thin file. The approach — which started with 100% approval on a random sample of applicants — allowed the scholars to observe the performance of borrowers who might otherwise have been rejected. It also lets them try out several loan-approval ranking methods using alternative data to achieve two goals: more borrower inclusion and higher profits for the loan maker.

The team found that the micro-loan website can achieve its highest profits at a 45% loan approval rate. When all three categories of alternative data were applied to evaluate borrowers’ credit risk and inform loan-making decisions, the lender could raise profit by 28% beyond yield from conventional information only. This is intriguing. However, two categories — social media and shopping activities — introduced and even amplified bias into the decision-making process.

Things like gender, income, and education level correlate with different behavior patterns, Lu says. When it comes to shopping, men and people with lower education levels spend more on virtual products such as entertainment, online games, and electronic books than women. "This feature is quite important to improve financial credit accuracy, which is good. But, because this feature is highly related to gender and education level, that actually will amplify the financial unfairness," he notes.

Loan selection based on mobile activity yielded 22% more economic gains than selection based on conventional information only under the optimal 45% approval rate. Better yet, it was 1.3 times more effective in improving financial inclusion than using social media information to assess risk.

In conducting this research, Lu hoped he’d determine whether lenders could be less invasive on the data privacy front and skip gathering all the alternative data they could get their hands on. It turns out, they can.

"Smartphone usage data is enough," says Lu. "When you combine all the different sources, we see marginal additional improvement in profit, so it's not necessary to collect the shopping data and social media information. The value of cell phone data dominates the others, can help improve financial fairness, and is a good source to balance the trade-off between profit and equity.