Neural networks show fewer false positives on bad loans – study

Machine learning method edges regression techniques in linking nonlinearities among delinquent borrowers

Credit risk models that use neural networks are the most accurate method of forecasting repayment of non-performing loans, a new study by academics suggests – one that could have implications for lenders and governments planning loan moratoria programmes.

The study compares five approaches to modelling recovery rates on consumer loans that are overdue but not yet in default. Two of the models rely on classical regression techniques, and three are based on machine learning. The findings were