Risk prediction for non-affiliated companies in the banking sector

The challenge

The banking entity has tools to detect the risk of default and establish pre-approved credit limits. The model is accurate for companies where the bank has sufficient information. However, for those companies where there is limited information, the bank applies a restrictive expert criterion and only offers minimum limits, regardless of their economic solvency.

Objectives

Expand the credit portfolio of healthy customers by incorporating non-affiliated clients who, although they do not have detailed financial information, have adequate economic solvency.

How have we done it?

Machine Learning
XGBoost
Feature engineering
Cost and optimization function

Characterization of clients based on their risk of default
Characterization of unknown clients from limited information on known clients
Risk Scoring

Results

The current model only allows for 250,000 pre-approved companies to apply for credit limits. The new models improve this number up to 150,000 non-affiliated companies.