Impact of Artificial intelligence through credit scoring methods on the non-performing loan rate of top U.S. banks

Abstract

The analysis of the non-performing loan ratio has seen a rise in interest in the literature since the great financial crisis of 2007. In this thesis, we explored multiple facets of the non-performing loan issue in the banking system. We introduced the evolution of credit scoring methodologies for loan acceptance and management alongside the evolving role of artificial intelligence in the banking sector. We then investigated the intersection of these two fields by identifying the latest developments in credit scoring through machine learning algorithms, as well as their known issues for broader acceptance in this field. We reviewed the literature related to the determinants of the non-performing loan rate and identified proxies of technological innovation related to artificial intelligence through the use of patent information. Finally, we tested our hypotheses through a multiple regression panel model on a sample of 43 U.S. banks during the 2010-2023 period to determine whether banks that are pioneers in artificial intelligence demonstrate a better distribution of their non-performing loan rate. Our results do not allow us to reject the null hypothesis that banks pioneering in AI development do not demonstrate a better NPL rate among their loan portfolio.