This website uses cookies.

By clicking the "Accept" button or continuing to browse our site, you agree to first-party and session-only cookies being stored on your device to enhance site navigation and analyze site performance and traffic. For more information on our use of cookies, please see our Privacy Policy.

  1. Home
  2. ASSA Annual Meeting
  3. 2021
  4. Preliminary Program
  5. Digital Credit in Developing Countries
« Back to Results

Digital Credit in Developing Countries

Paper Session

Monday, Jan. 4, 2021 3:45 PM - 5:45 PM (EST)

Hosted By: American Economic Association
  • Chair: Jonathan Robinson, University of California-Santa Cruz

Manipulation-Proof Machine Learning

Daniel Bjorkegren
,
Brown University
Joshua Blumenstock
,
University of California-Berkeley
Samsun Knight
,
Brown University

Abstract

An increasing number of decisions are guided by machine learning algorithms. In many settings, from consumer credit to criminal justice, those decisions are made by applying an estimator to data on an individual's observed behavior. But when consequential decisions are encoded in rules, individuals may strategically alter their behavior to achieve desired outcomes. This paper develops a new class of estimator that is stable under manipulation, even when the decision rule is fully transparent. We explicitly model the costs of manipulating different behaviors, and identify decision rules that are stable in equilibrium. Through a large field experiment in Kenya, we show that decision rules estimated with our strategy-robust method outperform those based on standard supervised learning approaches.

Gender-Differentiated Digital Credit Algorithms Using Machine Learning

Paul Gertler
,
University of California-Berkeley
Sean Higgins
,
Northwestern University

Abstract

Low-income women disproportionately lack access to credit, often because they lack credit histories, property rights, and formal earnings. This, in turn, leads to a cycle of exclusion from formal credit markets, as a lack of data to assess low-income women's creditworthiness prohibits them from building a credit history. Partnering with a commercial bank in the Dominican Republic, we add an innovation to machine learning credit scoring algorithms used to disburse digital credit by gender-differentiating them. Specifically, our model uses machine learning algorithms to sift through and transform a broad range of characteristics from existing low-income clients' mobile phone data to determine the best predictors of creditworthiness separately for men and for women. We find that even when a pooled model with men and women considers gender as a predictor variable, 80% of women receive a higher predicted probability of repayment in a gender-differentiated model. We also find that this is achieved with minimal impact to the overall accuracy of the model. Thus, low-income women would be more likely to receive credit and gender equity would be improved if lenders use a gender-differentiated model.

Too Fast, Too Furious? Digital Credit Speed and Repayment Rates

Alfredo Burlando
,
University of Oregon
Michael Kuhn
,
University of Oregon
Silvia Prina
,
Northeastern University

Abstract

Digital loans are a source of fast short-term credit for millions of people. While digital credit broadens market access and reduces frictions, default rates are high. We study the role of speed of delivery of digital loans on repayments. Our study combines unique administrative data from a digital lender in Mexico with a regression-discontinuity design. We show reducing speed by doubling the loan delivery time from ten to twenty hours reduces the likelihood of loan default by 21%. Our finding hints at waiting periods as a potential consumer protection measure for digital credit.

Filling a Hole or Digging a Hole? Evidence from Malawi on the Impacts of Digital Credit

Valentina Brailovskaya
,
IDInsight
Pascaline Dupas
,
Stanford University
Jonathan Robinson
,
University of California-Santa Cruz

Abstract

Consumer digital credit has the potential to expand credit access dramatically; however, as currently offered, loans tend to be for small amounts, with short terms, at high effective interest rates (well over 100% APR annually). What is the effect of access to these loans? Are consumers informed about loan terms? We use two research approaches to answer these questions in Malawi. First, we use a regression discontinuity design around credit score thresholds to estimate impacts. Second, we conduct a financial literacy RCT which informed customers of the sizeable fees and penalties associated with the loan.
JEL Classifications
  • G5 - Household Finance
  • O1 - Economic Development