Description:

Speaker: Bruno Fava

Title: Algorithmic Bias in Microcredit: The Unseen Costs of Data-Driven Lending Practices

Abstract: New machine learning methods may allow lenders to increase profits by targeting their loans to the most lucrative borrowers. But how will this affect which borrowers can access credit and the impacts of the loans on borrower outcomes? Using data from four RCTs on microcredit we find that profit targeting is possible and can increase returns for lenders. But profit targeting leads lenders to sideline poorer and less educated clients. Targeting loans to the borrowers who would see a large increase in income is unsuccessful given existing data and algorithms. This suggests caution is necessary when lenders begin optimizing for profits especially for "social'' lenders who claim to care about a double bottom line.