Publications Repository - Gdańsk University of Technology

This study examines the impact of battlefield developments in the ongoing Russian–Ukrainian war, which to date has lasted over 1000 days, on the stock prices of defence corporations such as BAE Systems, Booz Allen Hamilton, Huntington Ingalls, and Rheinmetall AG. Stock prices were analysed alongside sentiment data extracted from news articles, and processed using machine learning models leveraging natural language processing (NLP). Although the main hypothesis was not confirmed due to methodological and data limitations, the study demonstrated that neural network-based models, specifically long short-term memory (LSTM) networks, effectively captured hidden temporal patterns. The model's performance was evaluated using root mean squared error (RMSE). Alternative models, including XGBoost, ARIMA, and VAR, were also tested but did not yield accurate forecasts. The findings highlight nonlinear patterns in the data and emphasise the importance of hyperparameter optimisation, such as tuning the number of epochs andLSTM layer sizes. Techniques such as Grid Search and Random Search significantly enhanced forecasting performance, resulting in stock price predictions with low RMSE.

Authors

• dr hab. Oleksandr Melnychenko link open in new tab