Read the original article on EuroMoney.

‍

This summer, Fair Isaac Corporation, the Silicon Valley-based data science and predictive analytics company, will launch the latest version of its famous credit score, FICO Score 10.

‍

The FICO score, which is used in 90% of all consumer credit decisions in the US and by all the large credit card and auto loan companies, will in future include trended credit bureau data, rather than just an assessment of individuals’ income, indebtedness and previous credit event history at a given moment in time.

‍

FICO Score 10 will consider an historical view of data such as account balances for the previous 24 months and payment patterns looking across credit cards, mortgages, auto loans and personal loans, giving lenders more insight into how individuals are managing their credit. The company claims that by adopting FICO Score 10, a lender could reduce the number of defaults in their portfolio by as much as 10% among newly originated bank cards and 9% among newly originated auto loans, compared with using FICO Score 9.

‍

Ethan Dornhelm, vice-president of FICO scores and predictive analytics, adds: “Beyond our regular releases of new flagship FICO Score models, we continue to develop new, innovative models using alternative data to help consumers who fall outside the traditional credit reporting process and have difficulty getting access to credit.” These might include incorporating consumers’ telco and utility payments in their FICO scores and allowing consumers to permission access to their checking, savings or money market accounts to demonstrate sound financial behaviour.

‍

Changes in consumer credit behaviour, such as the increasing use of personal loans for debt consolidation, and advances in data science exploring the information content in various data points relative to credit risk are prompting new evolutions of credit underwriting. However, are utility bills and trended credit card balances alternative data or just more of the same?

‍

“Technology evolves – culture does not,” says Michele Tucci, chief product officer of Singapore-based CredoLab, an artificial-intelligence-based credit scoring platform that crunches smartphone metadata to discover predictive delinquent behaviour patterns.

‍

Trended v alternative data

‍

Tucci contends that what works in Anglo-Saxon countries does not export very well. Tucci tells Euromoney: “The way credit scoring works in the US and the UK is very different from how it works in countries in southeast Asia, for example, which lack credit bureaus and have very high levels of cash-based transactions.

‍

“FICO and Experian may well add more transaction data to their algorithms, especially if borrowers allow them to connect to their accounts through open banking APIs. But that trended data is not really alternative data, which rather comes from digital footprints online, e-commerce and smartphones.”

‍

While bank account and credit bureau penetration may be low in emerging markets compared with the US, smartphone penetration is high. CredoLab markets its credit scoring to lenders including banks, such as Malaysia-headquartered CIMB Group, consumer credit companies, insurers and retailers.

‍

When a prospective customer applies for credit, the bank or other lender forwards a link to CredoLab’s credit scoring app and a password to activate it. The app then crunches anonymized metadata from the phone in a matter of seconds and delivers a score to the prospective lender. The app can then be deleted. This is a one-off scoring process. It does not collect the applicant’s identity.

‍

CredoLab does not peer into banking apps on the phone. It cannot access these in any case without a log-in and password. Rather, it looks for other indicators that might fit patterns of potential loan delinquency.

‍

Someone with a lot of apps dedicated to health, wellbeing and playing sports might have a different potential delinquency score than someone with lots of gambling and gaming apps   - Michele Tucci, CredoLab

‍

While traditional credit bureaus concentrate on ability to repay, CredoLab concentrates on willingness to repay, using machine learning to identify patterns of behaviour on mobile phones that correlate to delinquency. “For example, the model will look at the categories of apps on a loan applicant’s phone,” says Tucci. “Someone with a lot of apps dedicated to health, wellbeing and playing sports might have a different potential delinquency score than someone with lots of gambling and gaming apps.

‍

“It will also look at contacts data. How many contacts does a person have? How frequently or recently were they uploaded? How many have a first name, second name, additional phone number and email? The model is looking for signs that the phone holder is well organized.” Similarly, the model takes patterns from the phone’s calendar.

‍

Tucci continues: “Does the phone owner schedule a lot of meetings? What is the frequency and distribution of meetings? Again, the model is looking for signs of being well organized.” Credit scoring models are themselves measured by their Gini coefficient: a score from 0, which indicates a model no better at predicting a good credit than tossing a coin, up to 1, which is perfection. Tucci says: “In emerging markets, where only from 20% to 60% of potential loan applicants might have data tracked by credit bureaus, a bank might hope to achieve a Gini coefficient of 0.25 without credit bureau data, up to 0.4 with credit bureau data.

‍

“Due to the low correlation of the different sources of data, if you combine that 0.4 with CredoLab, it can then bring the Gini coefficient up to 0.6, which is closing in on developed market scores, which might be around 0.7.”

‍

Predictive power

‍

Risk managers at banks and other lenders – CredoLab’s customers also include a large ride-hailing company that lends to its drivers and an online travel company that offers a fly-now-pay-later service – want as much data as possible with high predictive power on as many applicants as possible.

‍

Operating in markets with high mobile-phone penetration, CredoLab offers the hope of expanding the pool of potential customers on which lenders and other businesses can do some credit scoring. In the US, the initial read on the impact of FICO 10 is that those consumers with already strong credit scores will now have slightly better ones, while those already with low scores will look a little worse. Most won’t be much affected.

‍

Could new credit scoring methods come to the US?

‍

In January, Maxine Waters, chairwoman of the House Financial Services Committee, wrote to the Government Accountability Office requesting more information on the potential use of alternative data in credit scoring for Americans. Waters pointed to findings from the Consumer Financial Protection Bureau that 45 million Americans cannot be scored by the traditional credit rating agencies because they have no credit history or limited or outdated history. Only one third of Americans aged between 18 and 29 have a credit card, but nearly all millennials have a mobile phone.

‍

“Alternative data can help lenders identify creditworthy potential borrowers that lenders would otherwise miss,” writes Waters. Sounds like an emerging market.

‍