Title: Beyond FICO: Leveraging Cash Flow Data to Improve Credit Access
Abstract: This paper asks whether U.S. entrepreneurs who are traditionally disadvantaged by credit score-based underwriting can benefit from cash flow-based underwriting. We use data from three fintech companies; two are lenders and one is a platform connecting applicants to lenders. Incorporating cash flow variables in machine learning models to predict default improves performance suggesting they add value to underwriting. Across groups, we find the largest differential impact for younger relative to older entrepreneurs, with cash flow models being more informative and predicting lower default rates for younger entrepreneurs. There are smaller but similar patterns for other traditionally disadvantaged groups. Cash flows impact both the intensive and extensive margins in underwriting decisions. Within-applicant analysis of data from the platform indicates that cash flow-intensive lenders are more likely to approve traditionally constrained groups, though this appears correlated with lender automation. We then test whether the value of cash flow variables in underwriting improves access to credit. Using random assignment to loan officers at one lender, we show that more cash flow-intensive underwriting causally increases approval rates among younger entrepreneurs who have fewer tangible assets.