Topic: The Effect of AI-Enabled Credit Scoring on Financial Inclusion: Evidence from One Million Underserved Population
Speaker: Professor Bin Gu, Boston University, USA
Time and Date: 9:00, Oct. 28th, 2022
Platform: Tencent Meeting ID: 570-233-071
Speaker Profile:
Bin Gu is Everett W. Lord Distinguished Faculty Scholar, Professor and Department Chair of Information Systems at the Questrom School of Business, Boston University. His research interests include the future of work, fintech, online social media and social network, digital platforms, the sharing economy, and the societal/business value of data analytics and artificial intelligence. His research has been published in Management Science, MIS Quarterly, Information Systems Research, Journal of Management Information Systems, among others. Professor Gu serves or had served as a senior editor on the editorial boards of Information Systems Research and MIS Quarterly and as a conference co-chair, track chair, or associate editor for major IS conferences. Professor Gu obtained his PhD degree from the Wharton School of Business at University of Pennsylvania.
Abstract:
Financial inclusion has been a global challenge for financial institutions due to the lack of credit history and other financial data of the underserved population. The advances in artificial intelligence (AI) and machine learning, combining with the richness of alternative data, enable financial institutions to use AI models to evaluate underserved borrowers. We investigate whether and how AI models influence financial inclusion and lending performance, as measured by approval rate, default rate, and utilization level. We cooperate with a large regional bank in China with more than 50 million customers which previously use traditional underwriting process (i.e. rules designed by experts). We take advantage of a nationwide financial policy change in China which encourages banks to use AI models in underwriting to complement their current approach. The focal bank developed an AI model for its credit line product A as a pilot and we identify a similar credit line product B which didn’t use the AI model. We apply a difference-in-differences approach to investigate the treatment effect of the AI model. We find a 1.5% decrease in the approval rate for the entire population, which was driven by a 3.5% decrease for the regular population and a 16.5% increase in the underserved population. In the meantime, the default rate for both groups decreases significantly and utilization level increases significantly. Taken together, the introduction of the AI model increases financial inclusion for the underserved population by enhancing approval rate and reducing default rate simultaneously. Further analysis attributes the improvement in financial inclusion to two factors, the inclusion of weak signals that were ignored by the human experts, and the development of more sophisticated decision models. The inclusion of weak signals contributes slightly less than one half of the improvement and the development of more sophisticated decision models contributes slightly more than one half. The inclusion of weak signals is especially helpful to the underserved population while the regular population benefits more from the development of more complicated decision models. Our study documents the impacts of AI models on financial inclusion and provide the first piece of evidence regarding the underlying mechanisms, helping financial institutions and regulators develop a better understanding on how AI models can complement traditional models in improving financial inclusion.
