Прогнозирование банковских продаж на примере ПАО «Сбербанк»

Abstract

Introduction. This scientific study highlights the relevance of modeling and forecasting sales of Sberbank in terms of effective business management. Sales forecast is an important tool for predicting the demand for goods and services, as well as determining the adequate strategies and tactics to achieve the company’s goals. The research is distinguished by its reference to artificial intelligence methods in the field of marketing. Forecasting methods applied to a proprietary data sample of Sberbank’s daily sales give novel results, which reliably supports the development of adequate strategies and tactics for successful business management. The key hypothesis of the study is to check the prognostic potential of machine learning methods against the traditional econometric approaches to modeling Sberbank’s sales. The purpose of the study is to develop sales forecasting models for multifunctional products and their practical instruments for Sberbank’s Sales Network Block. Materials and Methods. The study relies on the methods of system-oriented analysis, statistical and economic mathematical methods of data analysis and their processing. Collected and pre-processed sales data for Sberbank’s phantom products reflecting the dynamics of bank sales were used for computational experiments to build a few forecasting models and justify the choice of the best model among those built. Results. Random Forest and Gradient Busting (XGBRegressor) Models used training and test samples to give the forecasts with the accuracy significantly higher than the accuracy of forecasts by ARIMA-model and linear regression. Conclusions. The results of the analysis reliably confirm that machine learning methods are currently promising methods for forecasting bank sales and can be the subject of further research in this area. Machine learning techniques introduced into banking practices have the potential to significantly improve the effectiveness of existing sales and risk management.