Predikce obchodních signálů na českém trhu pomocí indexu volatility a strojového učení

This diploma thesis deals with the use of machine learning for predicting the development of Czech stocks using the VIX and VSTOXX volatility indices. The triple barrier labeling method is used for the construction of the target variable. The predictor base is created based on stock data, volatility indices and selected macroeconomic indicators. The prediction of trading signals is carried out using ensemble tree-based models, namely the random forest and XGBoost, whose hyperparameters are optimized using the grid search method with cross-validation. The models are applied to the stocks of ČEZ, a. s., Komerční banka, a. s. and Philip Morris ČR, a. s. for the period from 2013 to October 2025. The results show that both models achieve slightly better outcomes than random classification, but after accounting for transaction costs they are usually unable to systematically outperform the benchmark strategies Buy and Hold or the momentum strategy. The thesis thus shows that the prediction of Czech stocks based on volatility indices and tree-based models has limited practical usability, mainly due to the nature of the Czech capital market, the high frequency of trades generated by the models and the related transaction costs.