Table 4 Parameters and criteria adopted to classify borrowers according to banks’ experience
From: To supervise or to self-supervise: a machine learning based comparison on credit supervision
Parameters | Description |
|---|---|
Banks | 20 biggest loan portfolios in financial system at the starting date (June/2019) |
Extraction dates | December/ 2016; December, 2017; December 2018 |
Good borrowers | Rated as AA, A or B (LLP inferior to 3%) in each one of the last seven months previously to the dates chosen for extractions |
Bad borrowers | Rated as F, G or H (LLP equal or higher than 50%) in each one of the last three months previously to the dates chosen for extractions. A bad borrower was excluded from the dataset if, in the 6-month period after extraction dates: (1) the debt was paid; (2) the debt was reduced; (3) the rating assigned by the bank improved |
Materiality | Loans in excess of R$ 200,000a |