When Bitcoin Sneezes, Does Cardano Catch a Cold? Spillovers Between Brown and Green Cryptos: A Graph Neural Networks approach using intraday data

This thesis investigates volatility spillover effects in the cryptocurrency market by applying graph-based volatility forecasting models to a portfolio of Proof-of-Work (PoW) and Proof-ofStake (PoS) cryptocurrencies. Four models are developed and compared, HAR, GHAR, GNNHAR-1L, and GNNHAR-2L to address three core research questions: (1) Do volatility spillovers exist among cryptocurrencies? (2) Are these spillovers nonlinear? and (3) Do indirect (multi-hop) neighbors contribute meaningfully to volatility transmission? High-frequency price data sampled at 5-minute intervals over the 2023–2024 period is used to construct realized volatility measures, HAR-style lagged features, and dynamic graph structures estimated via rolling-window Graphical LASSO. Model performance is evaluated using the Model Confidence Set (MCS) procedure to ensure statistically robust comparisons. The empirical results show that GNNHAR-1L consistently outperforms both GHAR and GNNHAR-2L across all assets, providing strong evidence for the presence of nonlinear spillover effects. By contrast, incorporating second-hop neighbors in GNNHAR-2L does not yield significant forecasting improvements, indicating limited indirect spillover dynamics in this dense volatility network. Furthermore, no meaningful differences are observed between PoW (“brown”) and PoS (“green”) cryptocurrencies in terms of volatility transmission behavior. Instead, the analysis uncovers clear volatility-based clustering: assets with similar volatility levels exhibit stronger connections and more pronounced spillover linkages. These findings contribute to the growing literature on graph-based financial modeling and provide new insights into the structural and nonlinear nature of volatility interdependence in cryptocurrency markets. The study concludes with several directions for future research, including alternative graph-construction methods, multi-horizon volatility modeling, cross-market spillover analysis, and the integration of investor heterogeneity into network-based volatility frameworks.