Breaking the doom loop: the danger of self-fulfilling prophecies in modern credit risk

Breaking the doom loop: the danger of self-fulfilling prophecies in modern credit risk

The ability to distribute trustworthy credit is a societal cornerstone. But what happens when traditional credit scoring methodologies aren’t available? Will new 'advanced' credit models in emerging markets be self-fulfilling prophecies?

Philip Mackenzie ([email protected])

Credit Prophecy

Right in the wrong ways?

In the previous PoV in this series, I considered scoring mechanisms like the Chinese social credit score, and asked about the risks that might arise if these types of scoring models are wrong: if loading social engineering into a credit model might distort its outcomes. I proposed that this result was fairly likely given the necessary time it takes to tune credit scores.

A more subtle question is ‘what if credit scoring models are right, but in the wrong ways?’. Arguably this is a risk posed by the new breed of models, such as those mentioned in the previous article. Primarily, what if predicting someone to be at risk of default causes them to default? What if a credit scoring mechanism predicts a risk of default and this then causes the default because the individual cannot gain access to other ways of earning money?

Unpredictable predictions

The concept of gathering and monetizing customer data predates the internet. Credit scores have been around since the 1950s, and from the 1980s onward firms have used loyalty programs and internal company records to keep track of their customers. Since the arrival of the internet the amount of available data has become truly vast, and large ecosystems of data firms that collect, clean and trade consumer data among themselves have emerged.

Typically this data is used in one of two ways: to influence (as in advertising, the core of the business models of firms such as Facebook and Google), or to predict customer outcomes. This article is concerned primarily with predictions – the results of which can be unexpected, as illustrated by one notable example. In the retail space, one particularly profitable type of customer is newly expectant mothers: they have to buy a lot of things for a new baby, and are time-constrained enough that they will likely do so in one place. Back in 2012, US retailer Target sent catalogs of maternity items to a customer that its system, based on her buying patterns, had identified as probably expecting a child. As it turned out, the expectant mother was a teenage girl, whose father was somewhat taken aback by the revelation that not only was his daughter pregnant, but that Target had picked up this information before he did.

This ability to predict outcomes is at the core of much of credit risk analytics. Reliable credit (‘if I lend to this person, am I likely to get it back?’) is broadly considered one of the foundations of a functioning capitalist society. Institutions lend mortgages so people can own houses, and loans so they can start businesses. Typically an institution will use external data (whether observed or inferred) to predict how likely someone is to be a credit risk – fundamentally, how likely they are to default on loans.

Ideally, the best indicators for credit risk scoring are transaction data and credit history. If an individual or firm has a history of failing to repay its debts, then there is a good chance it will fail to do so again in future.

The problem arises when access to transaction and credit data is limited, or when credit data doesn’t exist at all. According to the World Bank’s 2017 Global Findex report, about 1.7 billion adults globally remain unbanked (and hence lack a transaction history). To compensate for the lack of traditional scoring inputs, firms must employ proxies, which use other factors about how people behave to figure out how likely they are to default.

Proxies, prophecies and loops

A number of scoring companies are currently using proxies. This includes using mobile data services to build a model of someone’s lifecycle and corresponding credit risk profile, or calculating credit scores based on people’s online behavior (such as how they use mobile apps).

New ‘proxy’ credit models are being targeted at users in several emerging marketplaces (mainly South America and Africa but also parts of Asia), where people have access to mobile or internet data but are otherwise unbanked, and where they may be using services such as Kenyan mobile banking application M-Pesa.

The concept of ‘self-fulfilling prophecies’ in credit risk has been noted in the corporate space with respect to ratings agencies, but also with respect to activities such as job applications. In the latter case, people were analyzed for employment opportunities with their credit score and ultimately turned down, which then caused their credit score to deteriorate.

Basing credit scoring on even more tenuous factors could cause more of these ‘doom loops’ to arise. It could justifiably be pointed out that an opaque or inaccurate credit-score proxy is better than nothing at all (institutions need to build a case to lend money, after all, otherwise they simply won’t lend). But this means little to someone who becomes excluded from the financial system for, say, not charging their phone enough.

Many financial institutions will be looking at emerging markets as key places to invest, because there the middle class is rising and/or emerging. While there are significant opportunities for firms to invest, however, it is worth examining some of the underlying assumptions behind financial growth and what it means, and how potential winners and losers can be created by ‘new’ methods just as they can by old ones. Emerging credit-scoring methods aren’t a protection against either default risks or potential reputational damage, and for investors, pulling the trigger on these new firms may be riskier than they expect.

Further reading

Why more is not always better in the world of credit

(July 2019, Chartis Research)

Technology Solutions for Credit Risk 2.0: Vendor Landscape

(May 2018, Chartis Research)

Financial Crime Risk Management Systems: Enterprise Fraud: Market Update and Vendor Landscape

(October 2019, Chartis Research)

Points of View are short articles in which members of the Chartis team express their opinions on relevant topics in the risk technology marketplace. Chartis is a trading name of Infopro Digital Services Limited, whose branded publications consist of the opinions of its research analysts and should not be construed as advice.

If you have any comments or queries on Chartis Points of View, you can email the individual author, or email Chartis at [email protected].