NBER Paper: ML Boosts Lending Margins 7-9% but Excludes Women and the Poor

lihua_lei_stat · x · 2026-08-01

A new NBER working paper by Susan Athey et al., analyzing randomized microcredit approvals in South Africa, the Philippines, and Bosnia, reveals that machine learning can identify highly profitable borrowers, boosting lending margins by 7–9 percentage points.

However, this ML-based targeting comes at a social cost, as it systematically excludes women and lower-income households from credit access. The researchers suggest that if lenders fine-tuned targeting within baseline income quintiles rather than broadly, they could balance goals better, though it would cut profit gains by about half.

Original post →

More from AGI Musings

AGI Musings channel →