Distilling a Disruptive Disintermediary’s Data: Interpretable Machine-Learning Explanations for LendingClub Customers

Machine learning may assist peer-to-peer lenders in exploiting their informational advantage through distilling large volumes of data into an evaluation of borrower credit quality. In this chapter, we use explainable artificial intelligence to pare back the opacity associated with machine learning. Using LIME (Local Interpretable Model-Agnostic Explanation) and Shapley Values, we provide a visual representation of the factors found to influence credit risk for the LendingClub peer-to-peer platform. Empirical findings indicate that FICO scores are still relevant, that experienced borrowers are less risky, that loans for credit card repayments are charged more, and that administration burdens such as verifying income leads to a higher cost of credit. Our work links to ongoing regulatory initiatives by providing a mechanism to provide meaningful interpretations from machine learning models to customers, regulators, and investors.