Ekonometrická analýza ceny Bitcoinu na globálním trhu

Abstract

Bitcoin is the world's largest cryptocurrency. It represents a specific asset class characterised by high price volatility and growing integration with traditional financial markets. Understanding the factors driving its price dynamics is crucial for investors and policymakers. Yet the existing empirical literature offers no consensus on the relative contribution of crypto-specific, macroeconomic, and on-chain predictors. The aim of this thesis is the econometric analysis and prediction of weekly Bitcoin price returns using macro-financial and on-chain predictors. It also seeks to identify specifications with the highest practical utility on an independent hold-out sample covering the year 2025. An automated selection pipeline searching thousands of candidate specifications across five structural branches was employed. Specifications are evaluated by a composite score combining goodness-of-fit and residual diagnostics. OLS models, GARCH-class conditional volatility models, and machine learning methods were estimated. The results indicate a dominant influence of the synchronous Ethereum price on weekly Bitcoin returns. Its exclusion leads to a substantial decline in explanatory and predictive power across all methods. Among supplementary approaches, Random Forest achieved the lowest prediction error, suggesting the presence of nonlinear relationships in the data. GARCH models confirmed strong conditional variance persistence and proved valuable for constructing probabilistic prediction intervals. The developed pipeline provides a unified framework for selection, diagnostics, and out-of-sample validation across three model classes, contributing to a deeper understanding of the factors driving short-term Bitcoin price dynamics.