When a bank is lending money to a company, detecting which companies are reliable and which aren’t is crucial. In 2008, we created predictive models that were able to predict with 98% accuracy if a company will go bankrupt in the next 6 months.
The dataset was composed of the annual financial status of each company. We had over 3 years of data with 500 raw variables per year, so about 1500 variables in total. The raw variables were not very interesting themselves, so we added about 22500 “evolution” variables with Anatella, using the “Calculator Vectorized” box. The final dataset used to create the predictive model was composed of more than 24000 variables. The final models were able to predict bankruptcy within 6 month with 98% accuracy.