Bayesovský přístup pro predikci HDP ČR

The aim of this bachelor thesis is to model and forecast the year-on-year growth of the Gross Domestic Product (GDP) of the Czech Republic using methods of Bayesian econometrics. The thesis compares two second-order autoregressive models – a standard model with normally distributed errors and a robust model utilizing the Student's t distribution. Model comparison using the Bayes factor suggests that the robust specification may better capture extreme macroeconomic shocks (e.g., the COVID-19 pandemic) due to its heavy tailed property. Parameter estimation is performed using MCMC simulation methods (the NUTS algorithm) using nonlinear reparameterization for direct estimation of the long-term mean of the process. Subsequently, the out-of-sample predictive accuracy of the models is evaluated using the Sliding Window method. Empirical results suggest that while the model with a non-informative prior provides slightly higher point accuracy in the test period, the informative prior contributes to increased estimation efficiency through narrower credibility intervals. The thesis suggests that a Bayesian model, if appropriately specified for the given purpose, represents a flexible tool for macroeconomic analysis in turbulent periods.