Fraud Vulnerability & Fraud Prevention
Digital Financial Inclusion: Building Resilience Against Scams
Lisa Spantig, Laura Barasa, Elif Kubilay, & Eva Raiber
The recent expansion of digital financial services (DFS) has been beneficial for financial inclusion, especially in developing countries. However, this expansion also leads to severe consumer protection issues such as fraud and scams. As scams can contribute to the erosion of trust in digital services, avoiding victimization has become an important policy objective. In a cluster randomized controlled trial in Kenya, we will evaluate the effects of digital financial literacy training on DFS usage and take-up among rural women. We will test two approaches to build resilience against scams that will be incorporated into the training in two additional arms. The first approach involves educating groups about scams, similar to the predominant policy response to such consumer protection issues. The second approach is novel and encourages participants to share their experiences with scams. Sharing experiences creates a dynamic response to scams and ensures information is tailored to the recipient. Therefore, we expect that the sharing arm will be more effective in mitigating adverse effects of scams than the education arm.
A Bridge over Troubled Digital Water: An Experiment on Non-Institutional Fraud and Building Trust in Digital Financial Services In Nigeria
Michael King, Shane Byrne, Chaning Jang, & Daniel Putman
Non-institutional fraud targeted at micro, small, and medium enterprises (MSMEs) is pervasive across low- and middle-income countries. Fraud leads to monetary loss and psychological impacts, and in the long term can lower trust in, and willingness to access, digital financial services (DFS). In a lab-in-the-field experiment in Northern Nigeria with 750 Micro, Small, and Medium Enterprises (MSMEs), we assessed the impact of a short anti-fraud educational intervention and a pre-specified authentication code on capacity to distinguish between fraud and legitimate communications. With the experiment 70 percent complete, preliminary results show that the learning interventions do not significantly improve performance in the successful identification of fraudulent versus genuine communications, but do significantly impact the confidence with which participants report their verdicts by between 9% and 13%. Notably, we find that participants armed with the knowledge of the educational treatments report being significantly more likely to trust (by up to 42%) and to use (by up to 33%) DFS in future. We find tentative evidence that participants are less successful distinguishing between fraud and legitimate communications under time pressure. Results from the UCC technical intervention are pending.
Pricing & Consumer Choice
The Impact of Price Comparison Tools in Consumer Credit Markets
Sheisha Kulkarni, Erik Berwart, Sean Higgins, & Santiago Truffa
Consumer credit markets feature large amounts of price dispersion in loan costs, even conditional on loan and borrower characteristics. If consumers are unaware of the extent of this price dispersion, they may shop less and take out loans at higher interest rates than they would otherwise. We conduct a randomized controlled trial in Chile where we provide just-in-time, personalized information about the distribution of interest rates or the benefits of search to people searching for loans on Google. The first treatment arm is a price comparison tool that shows prospective borrowers a conditional distribution of interest rates based on administrative data on originated loans for borrowers and loans with similar observable characteristics. The second treatment is a simplified message that shows prospective borrowers an estimate of the monthly and total amount they could save by shopping at more banks. We find that consumers indeed underestimate price dispersion, and that the price comparison tool causes them to double their estimate of how much dispersion there is in interest rates. Consumers also underestimate the interest rate they will obtain, and the price comparison tool causes them to increase their prior about the interest rate they will obtain by 17 percentage points over a control mean of 23%. The simplified message treatment does not appear to affect priors about dispersion or the rate people expect to obtain. Our trial is still ongoing; after finishing our data collection, we will analyze results on loan characteristics using administrative data from Chile’s financial regulator. We will also conduct a follow-up phone survey that will be used to measure effects on search behavior and better understand how people form and update priors in consumer credit markets.
New Data Sources & Methods of Analysis
Social Media for Consumer Protection Supervision: Nigeria
Rik Ray
Monitoring and judicious use of digital media technologies and platforms are increasingly becoming salient aspects of digital financial services. Although such platforms and technologies have facilitated swifter communications between financial institutions and consumers, they are also frequently misused for defrauding consumers. The growing popularity of cryptocurrencies coupled with the challenges of regulating them further illustrates the need for institutions to develop and adopt robust strategies for monitoring digital media content, especially in the domain of consumer protection. We are working with the Central Bank of Nigeria (CBN) to evaluate their current social monitoring solution and developing alternative tools to improve their social monitoring and listening capabilities. Our initial findings indicate that the third-party tool currently in use does not offer any significant advantage over free and open-source solutions, either in terms of data gathering or analysis. Although we encountered some difficulties in training supervised machine learning text classifiers, a hybrid approach combining both unsupervised and supervised techniques showed promising results. Currently, our goal is to integrate the data collection, analysis, and visualization dashboard developed by a CBN team into a seamless package. In addition, when analyzed and interpreted cautiously, digital trace data can also provide broad insights into public opinion on financial policies and engagement with services that go beyond sentiment analysis and text classification. As such, we are also aiming to evaluate the impact of COVID-19 on consumers’ engagement with financial institutions in Nigeria on digital platforms using the pandemic as a natural experimental condition.
FinRegLab's Research on the Explainability and Fairness of Machine Learning in Consumer Lending
Natalia Bailey
In April 2022, FinRegLab and Professors Laura Blattner and Jann Spiess of the Stanford Graduate School of Business released new research on “Machine Learning Explainability and Fairness: Insights from Consumer Lending.”
This presentation will discuss the main findings of the research, which focused on the implications of using machine learning for extending consumer credit by analyzing tools for diagnosing and managing questions about the transparency and fairness of lending algorithms. The research empirically evaluates model diagnostic tools to help lenders address these transparency challenges and manage machine learning underwriting models as required by law.
The analyses are potentially useful to other sectors where machine learning predictive models are being used to make important decisions, such as medicine, criminal justice, and employment.