CUSTOMER STORY | BANKING
Bankruptcy prediction: BNP Paribas reaches 98% accuracy on at-risk business clients
In 2008, BNP Paribas wanted to predict which business clients would go bankrupt within 6 months. TIMi built 24,000 variables from public balance sheets, mostly year-over-year evolutions.
98%
Prediction accuracy
24,000
Variables added to the dataset
6 months
Prediction horizon
OVERVIEW
The challenge.
In 2008, BNP Paribas wanted to improve its predictions of which clients would go bankrupt within the next 6 months. In such a case, the bank will probably not recover its money, which means a net loss for the lender. The dataset is made of public balance sheets, with about 500 variables per year. The raw variables are not very interesting by themselves.
SOLUTION
What was done.
What matters is the evolution of these variables: how much revenue, for example, increases or decreases from one year to the next. With TIMi, the final analytical dataset was built with 24,000 variables, including many evolution variables, that is the ratio or difference of many raw variables from one year to the next. TIMi computes these evolution variables easily.
RESULTS
What changed.
98% prediction accuracy.
USE CASES BEHIND THIS STORY
Read the method.
SEE FOR YOURSELF
Test it on your own data.
Run TIMi on your own data with a free trial, or talk to us about your case.
