Novel Cash-Flow Based Underwriting Model using Gradient Boosted Decision Trees

Credit underwriting is a critical process that lenders engage in to predict the likelihood of a borrower being diligent or delinquent with their payment. Traditionally, the FICO credit score has been used to conduct this evaluation, only considering clients who have at least 6 months of payment history. Widespread usage of this system of evaluation led to a phenomenon called credit invisibility: a state where people are unable to obtain credit scores due to a lack of a viable form of payment history. As lenders can’t evaluate over 47 million credit invisible income-earning people in America, they do not lend them lines of credit and miss out on a significant amount of potential market share. Recently, cash-flow based underwriting has emerged as a potential replacement/improvement to the traditional credit score underwriting system. Using cash flow metrics in primary underwriting can surpass the minimum 6 month time barrier that prevents lenders from expanding. We evaluate the effectiveness of adding cash-flow metrics to machine learning models (Gradient Boosted Decision Trees, Support Vector Machines, and Linear Classification) that are trained on credit metric based data. Predictions from each model are used to simulate sample populations of borrowers. The results show that model accuracy, risk profile, profit percentage, and additional market share for lenders improve significantly when incorporating cash flow metrics.