Moody's Machine Learning Case Study

How has Moody’s used data science?

Moody’s is a credit rating agency who compared their internal parametric model against several machine learning models to assess the credit risk of small and medium-sized corporate borrowers. Moody’s proprietary model uses a traditional parametric general additive framework (GAM). The alternative machine learning models used were random forest, boosting, and neural networks. Two different datasets were used for the comparison study. The first dataset utilises only firm information and financial ratios, while the second dataset adds additional data to the first such as behavioural information including credit line usage, loan payment behaviour, and other alternative data.

Generalisation results were impressive with machine learning models outperforming Moody’s proprietary model from 2 to 3% using both datasets. There is also an 8-10% jump in accuracy when adding loan behavioural information.