Forecasting the profitability of insurance companies based on neural network models

Abstract

The relevance of the study is determined by the growing need to improve analytical tools for forecasting the profitability of insurance companies under conditions of war risks, structural changes, and financial market instability. Modern statistical analysis methods lose effectiveness in environments where indicator relationships are nonlinear. Under these conditions, the use of artificial neural networks, capable of modelling complex financial patterns and improving forecast accuracy, becomes particularly significant. The purpose of the article is to conduct a comparative analysis of the effectiveness of the main neural network architectures for forecasting insurance companies' profitability, identify their analytical characteristics, assess their adaptability to a volatile market environment, and determine their suitable areas of application. The study uses systematic, comparative, and analytical methods, which enable it to generalize scientific approaches to the use of artificial intelligence in financial forecasting and to identify the advantages and limitations of various neural network models. The constructed comparative tables enabled comparison of the architectures based on resilience to environmental changes, data type, forecasting horizon, and the level of interpretability of results. Results. The analysis shows that basic MLP models are effective for short-term forecasts under stable conditions. Recurrent networks (RNNs, LSTMs, GRUs) are effective for modelling temporal patterns and cyclical changes. Hybrid neural networks (Hybrid NNs) provide the highest forecast accuracy and flexibility in crises. Autoencoders and neuro-fuzzy systems perform auxiliary functions, improving data quality and the reliability of assessments. Conclusions. The practical significance of the results lies in the possibility of using the comparative characteristics of models as a methodological guideline for selecting the optimal architecture in the development of intelligent systems to manage the financial stability of insurance companies. Further research should integrate macroeconomic factors, seasonality, and war risks into the architecture of neural network forecasting systems.