Avantage Reply supported a large European bank based in Luxembourg in automating the back-testing of their retail models. The bank had a manual and tedious process, which required a lot of manual manipulations.
Avantage Reply automated the whole process using Python thus saving the bank important resources and improving the overall test procedures and back-testing guidelines.
The annual back-testing of retail models requires a lot of effort, in particular manual manipulations. The customer wanted to automate the whole process thus reducing the human effort and therefore improving the model’s performance if necessary.
To fully test a model, a sufficient amount of test data is required. In this case, the bank lacked an adequate amount of defaults data to test their Loss Given Default model.
Avantage Reply automated the whole back-testing process using Python. The automation process was based on the Bank’s current and historical portfolio composition and Data from Moody’s Default and Recovery Database (DRD). During the automation process, statistical tests such as T-test, Sign Test and Wilcoxon Test were performed on the observed and predicted values of the model using Python and were embedded for future back-tests. This contributed to improving the back-testing guidelines.