ASSESSING THE PROFITABILITY OF THE BANKING GROUP

Abstract

The article is devoted to the current problem of improving the reliability of profitability assessment using the example of a banking group. It has been shown that most previously developed profitability level prediction models are based on linear regression, without comparing them with non-linear models. In the work, the nature of the existing dependence of the level of profitability of the banking group on the performance of two participants in this group was analyzed. The construction of the correlation matrix and the inverse to it made it possible to identify a high level of dependence between the profitability levels of the members of the banking group. Therefore, a set of one-factor regression models of the profitability level of the banking group was built using the indicators of each group member separately from the other. Based on the information of each group, a linear, logarithmic and polynomial model was built. The best models for each member of the banking group were selected based on the determination factor, the Fisher criterion, and the average absolute error percentage. It was revealed that for one member of the banking group, the best model for predicting profitability was the polynomial model, for another - linear. Based on the selected models, a model was built that gives an average estimate from these models. In order to improve the quality of the models obtained, it was decided to use machine learning methods. Bagging was chosen as this method. A flowchart of the use of bagging to improve the quality of previously developed models was proposed. To implement the proposed algorithm, a Python program was developed using the Scikit-learn machine learning library. Using this method of machine learning made it possible: to find out the optimal number of samples for each of the selected models, to improve the quality of the developed models and to offer a two-factor model for assessing the profitability level of the banking group. The developed set of models can be used to assess the level of profitability and predict the impact of management decisions on the results of banking groups. It is proposed to use other methods of machine learning, such as a genetic algorithm and a random forest, when building future models.