Aplikace modelů s proměnlivými režimy při předvídání finančních časových řad

This thesis focuses on the application of regime-switching models in forecasting financial time series. The aim of the thesis is to compare the predictive performance of the linear ARMA model with selected nonlinear models, namely SETAR, STAR, and Markov Switching models. The theoretical part summarizes the fundamental principles of time series analysis and describes the construction and estimation of the considered models. The empirical part applies these models to selected commodity markets, with particular emphasis on crude oil prices (WTI). Forecasts are generated using a roll-up approach and evaluated in terms of both statistical accuracy and investment performance. The results of the analysis indicate that regime-switching models do not provide systematic improvements in statistical forecasting accuracy compared to linear approaches; however their application may be beneficial in certain specific cases.