Predicting probability of default with the help of macroeconomic indicators in ifrs 9 provision calculations

Abstract

IFRS 9 process has a very important issue for banks. IFRS 9 process assists banks in calculating and managing their required provision. Probability of Default is one of the important parameters in IFRS 9 process. There are two different Probability of Default (Through-The-Cycle and Point-In Time) in this process. However, Point-In-Time Probability of Default where only macroeconomic effects are reflected is used in required provision calculation. Through-The-Cycle Probability of Default cannot be used directly for provision calculation. Point-In-Time Probability of Default will be obtained by reflecting macroeconomic effects. In this study, it has been provided to convert Through-The-Cycle Probability of Default to Point-In-Time Probability of Default. Relief Approach and Genetic Algorithm (Evolutionary Search) were used in feature selection stage. k-Nearest Neighbors, Multi-Layer Perceptron and Extreme Gradient Boosting were used during creation of model. In this study, contemporary feature selection and modeling techniques were applied to the data set and the results were compared. In this study, contemporary regression models used seem to be quite successful.