The New Morality of Debt – IMF F&D
Increased datafication of debt raises ethical questions and calls for a new approach to regulating lending
Throughout history, society has debated the morality of debt. In ancient
times, debt—borrowing from another on the promise of repayment—was viewed
in many cultures as sinful, with lending at interest especially repugnant.
The concern that borrowers would become overindebted and enslaved to
lenders meant that debts were routinely forgiven. These concerns continue
to influence perceptions of lending and the regulation of credit markets
today. Consider the prohibition against charging interest in Islamic
finance and interest rate caps on payday lenders—companies that offer
high-cost, online loans. Likewise, proponents of debt forgiveness
appeal in part to morality when they advocate relieving hard-up debtors of
the burden of unsustainable debt.
In much of this debate, the principal moral value at play is fairness;
specifically, distributional fairness. Debt is deemed to be unfair and thus
immoral because of the inequality of knowledge, wealth, and power between
borrowers and lenders, which lenders can and often do exploit. Recent
technological advances in lending have added new dimensions to debt’s
morality. Notably, the datafication of consumer lending has
amplified moral concerns about harm to individual privacy, autonomy,
identity, and dignity. Datafication in this context describes the rapidly
growing use of personal data for consumer credit decision-making—particularly “alternative” social and behavioral data, such as a person’s
social media activity and mobile phone data—together with more
sophisticated data-driven machine learning algorithms to analyze those data
(Hurley and Adebayo 2017).
These techniques enable lenders to predict the behavior of consumers and
shape their financial identities in much more granular ways than in the
past. For example, it has been shown that borrowers who use iOS devices,
have larger and more stable social networks, or spend more time scrolling
through a lender’s terms and conditions are more likely to be creditworthy
and repay debt on time (of course, many of these variables proxy for
fundamental credit life-cycle variables, such as income). Innovation in
datafied lending has been driven largely by fintech start-ups, particularly
peer-to-peer lending platforms such as LendingClub and Zopa and Big Tech
companies like Alibaba/Ant Group. However, alternative data and
machine-learning techniques are increasingly being adopted by traditional
bank lenders, as highlighted by recent surveys from the Bank of England and
the Cambridge Centre for Alternative Finance.
These practices diminish consumers’ ability to craft their own identity as
they become increasingly chained to their “data self,” or algorithmic
identity. Moreover, the ubiquitous collection of data and surveillance that
fuels datafied lending constrains consumers from acting freely lest their
actions negatively affect their creditworthiness. And the commodification
of certain types of personal data for lending decisions raises moral
concern about harm to individual dignity. Is it moral for lenders to use
highly intimate health and relationship data—for example, captured from
social media and dating apps—to determine consumer creditworthiness?
Consumers may willingly share their data in specific contexts and for
specific purposes, such as to facilitate online dating and social
interaction. However, this does not imply that they consent to the use of
that information in new contexts and for different purposes, particularly
commercial purposes such as credit scoring and marketing.
Datafication also amplifies existing concerns about fairness and inequality
in consumer lending. Lenders are prone to abuse data-driven insights, for
example, to target the most vulnerable consumers with unfavorable credit
offers. Data-driven profiling of borrowers also facilitates more aggressive
and intrusive debt-collection practices against the poor. And more accurate
screening and price discrimination using alternative data and machine
learning increase the cost of borrowing for consumers previously subsidized
by hidden information (Fuster and others 2020).
In addition, increasingly data-driven, algorithmic lending could amplify
unfairness as a result of racial and gender-based discrimination, as
highlighted by the recent Apple Card debacle, when women were offered
smaller lines of credit than men. In particular, biases and proxy variables
in the data used to train machine-learning models could exacerbate indirect
discrimination in lending against minority groups—particularly where the
data reflect long-standing structural discrimination. Alternative data,
such as social media data, are typically more feature-rich than financial
credit data and thus embed more proxy variables for protected
characteristics, such as race and gender. The limited interpretability of
certain machine-learning methods (such as deep neural networks) could
impede efforts to detect discrimination by proxy. Deploying these
machine-learning models without rigorously testing their results, and
without meaningful human oversight, therefore risks reinforcing social
biases and historical patterns of unlawful discrimination, perpetuating the
exclusion of less-advantaged and minority groups from consumer lending
Yet the datafication of consumer lending could also uphold the
morality of debt, by improving other dimensions of distributional fairness
in consumer credit markets. Notably, more accurate credit assessment thanks
to machine learning and alternative data in algorithmic credit scoring will
improve access to credit, particularly for (creditworthy) “thin-file” and
“no-file” consumers previously locked out of mainstream credit markets
because of insufficient credit data, such as a credit history (Aggarwal
2019). Estimates from Experian and the US Consumer Financial Protection
Bureau suggest, respectively, that nearly 10 percent of the UK population,
and nearly 15 percent of the US population, have thin files or no files
(also described as “credit invisibles”) and lack access to affordable
credit. In developing economies, this figure is several times greater.
According to the World Bank Global Financial Inclusion Index, more than 90
percent of people living in south Asia and sub-Saharan Africa lack access
to formal credit.
Given that these consumers are often the least-advantaged members of
society, typically from ethnic minority and lower-income groups, improving
their access to credit supports financial inclusion and enhances
fairness—as well as efficiency—in consumer lending markets. Datafied,
algorithmic lending also stands to support fairness by reducing more
visceral forms of direct discrimination in lending—for example, stemming
from sexist or racist preferences of a (human) loan officer (Bartlett and
others 2017). Moreover, better access to credit and the accompanying
opportunities can enhance the autonomy and dignity of consumers.
More broadly, the digitalization and automation of lending stand to
increase financial inclusion by reducing transaction costs and making it
more feasible for lenders to extend small-value loans and reach consumers
traditionally excluded from borrowing by their remote physical location
(for example, a lack of bank branches in “banking deserts”). Data-driven
technology also can support financial inclusion by improving consumer
financial literacy and personal debt management. For example, automated
saving and debt pay-down features of many fintech credit apps can help
overcome some of the more common behavioral biases that undermine sound
personal financial management.
The rise of machine learning and datafied lending renders the morality of
debt much more nuanced. The Goldilocks challenge for regulators is to find
the right balance between the benefits and harms of datafied lending. They
must protect consumers from its greatest harms—in terms of privacy, unfair
discrimination, and exploitation—while still capturing the key benefits,
particularly improved access to credit and financial inclusion. However,
existing regulatory frameworks governing consumer credit markets and
datafied lending in places such as the United Kingdom, United States, and
European Union do not strike the right balance. In particular, they do not
sufficiently alleviate the privacy, autonomy, and dignity harms of datafied
The Goldilocks challenge for regulators is to find the right balance between the benefits and harms of datafied lending.
The prevailing approach to regulating consumer privacy in these
jurisdictions is distinctly individualistic. It relies on consumers to
consent to all aspects of data processing and to self-manage their
privacy—for example, by exercising their right to access, correct, and
erase their own data. However, this approach cannot protect consumers in
ever-more-datafied consumer credit markets. These markets display steep
asymmetries of information and power between borrowers and lenders,
negative externalities related to data processing, and biases that impede
consumer decision-making, such that individuals cannot on their own
safeguard their privacy and autonomy.
In a new article in the Cambridge Law Journal, I recommend ways to
address these inadequacies and close the privacy gap in consumer credit
markets through substantive and institutional regulatory reforms (Aggarwal
2021). To begin with, a more top-down regulatory approach is needed. Firms
should be subject to more rigorous obligations to justify the processing of
personal data under the paradigm of datafied lending. This should include
stricter ex ante restrictions on the types and granularity of (personal)
data that can be used for credit decision-making. For example, the use of
intimate, feature-rich data, such as social media data, should be
explicitly prohibited, and anonymization of personal data should be the
Firms should, moreover, bear a higher burden of proof regarding the
necessity and proportionality of processing personal data and thus their
encroachment on consumer privacy. This should include stricter, ongoing
model validation and data quality verification obligations, particularly
for nonbank fintech lenders. For example, in the context of algorithmic
credit scoring, lenders should be required to demonstrate that the
processing of alternative data yields a sufficiently significant
improvement in the accuracy of creditworthiness assessment.
These reforms should be accompanied by changes to the regulatory
architecture to improve the enforcement of consumer privacy protection in
consumer credit markets. In particular, regulatory agencies responsible for
consumer financial protection, such as the UK Financial Conduct Authority,
should have expanded authority to enforce privacy and data protection in
consumer credit markets. I argue that data protection is consumer
financial protection. Given their expertise and experience working with
consumer credit firms, sectoral agencies are in many ways better positioned
than cross-sectoral data protection and consumer protection agencies to
enforce data protection in consumer financial markets. However, they should
continue to collaborate with cross-sectoral regulators, such as the UK
Information Commissioner’s Office, that have expertise in data protection
Of course, these reforms are not needed only for datafied consumer lending
and its regulation. To truly safeguard the privacy of (credit) consumers,
stricter limits on the processing of personal data are called for in all
contexts, not only consumer credit markets, and on all actors in the
development life cycle of consumer-facing information systems. Likewise, in
an increasingly datafied economy, the optimal institutional arrangement for
data protection regulation entails a greater role for sectoral regulators
and deeper collaboration between sectoral and cross-sectoral regulators
everywhere—not just in consumer credit markets.
Aggarwal, Nikita. 2019. “Machine Learning, Big Data and the Regulation of
Consumer Credit Markets: The Case of Algorithmic Credit Scoring.” In Autonomous Systems and the Law, edited by N. Aggarwal, H.
Eidenmüller, L. Enriques, J. Payne, and K. van Zwieten. Munich: C. H. Beck.
———. 2021. “The Norms of Algorithmic Credit Scoring.” Cambridge Law Journal.
Bartlett, Robert, Adair Morse, Richard Stanton, and Nancy Wallace. 2017.
“Consumer-Lending Discrimination in the FinTech Era.” University of
California, Berkeley, Public Law Research Paper.
Fuster, Andreas, Paul Goldsmith-Pinkham, Tarun Ramadorai, and Ansgar
Walther. 2020. “Predictably Unequal? The Effects of Machine Learning on
Hurley, Mikella, and Julius Adebayo. 2017. “Credit Scoring in the Era of
Big Data.” Yale Journal of Law and Technology 18 (1): 147–216.
Opinions expressed in articles and other materials are those of the authors; they do not necessarily represent the views of the IMF and its Executive Board, or IMF policy.