Transformation of financial analytics tools under the influence of machine learning technologies

Abstract

The objective of the study is to develop a methodology and hybrid architecture for forecasting key financial indicators of issuers based on the integration of classical statistical models, ensemble machine learning algorithms and expert assessments of a financial analyst. Key results of the study: theoretical approaches to the evolution of predictive analytics have been systematized; a design for a comparative analysis has been developed on a sample of 96 Russian public companies for the period 2018-2025, including a time validation protocol and data balancing procedures; a statistically significant reduction in revenue and EBITDA forecast errors by the hybrid model compared to the market consensus by 39% and 37.6%, respectively, has been proven, according to the Diebold-Mariano test; the SHAP analysis method revealed that the greatest contribution to the forecast accuracy is made by relative financial ratios (34.2%) and semantic vectors of explanatory notes (22.8% in total). The hybrid architecture demonstrated increased resilience to structural shifts during the 2022-2023 crisis compared to individual algorithms, owing to the compensation for the dynamics of expert interpretation concepts. An updated financial analyst competency matrix was developed, including feature engineering, model validation, explainable analytics, and simulation modeling. Finally, it was concluded that the current paradigm for building financial analytics, based on deterministic financial models and market consensus, requires a shift toward integrating algorithmic scoring with expert validation to achieve a balance between predictive accuracy, interpretability, and regulatory acceptability. It was found that achieving target indicators for the reliability of cash flow forecasts and resilience to macroeconomic shocks is impossible without synchronizing machine learning methods with the company's aggregated financial model. The approach proposed in the study can be used by investment departments and analytical units of credit institutions in developing internal standards for validating forecast models, in higher education programs in the field of training “Finance and Credit” to update professional competencies in the context of the digital transformation of the economy.