Aplikace metod umělé inteligence při obchodování na akciových trzích

This master’s thesis focuses on the application of artificial intelligence methods in stock market trading and aims to empirically assess whether LSTM, Random Forest, XGBoost, and logistic regression models can construct trading strategies that outperform the passive „Buy and Hold“ approach based on predictions derived from historical data (daily prices of 30 stocks from various sectors over the period 2006–2025). The models are trained and tested using a rolling window methodology, which helps to reduce the risk of look-ahead bias and enables a more realistic simulation of the predictive process. Model predictions are transformed into trading signals and evaluated through a backtest of an equally weighted portfolio strategy that accounts for transaction costs. The results indicate that predictive accuracy remains close to the level of random classification over the long term and that the performance of model-driven strategies is strongly influenced by trading activity, highlighting the practical limitations of applying artificial intelligence methods in active trading while also suggesting potential directions for future research.