Regime‐Dependent Value‐at‐Risk and Risk Spillovers in Multivariate Cryptocurrency Markets

ABSTRACT

This paper develops a regime‐dependent framework for measuring and analyzing downside risk in cryptocurrency markets. We first adapt a t‐SNE‐based multivariate structural‐break detection method to identify common regime shifts in a correlated system of five cryptocurrencies from 2016 to 2021. The detected breaks correspond to the cryptocurrency boom of 2017 and the COVID‐19 period, suggesting marked changes in the joint dynamics of the market. We then incorporate these breaks into the estimation of value‐at‐risk (VaR) using EGARCH and Generalized Autoregressive Score (GAS) models under alternative residual distributions. The results show that model performance depends strongly on the market regime. In the relatively calm pre‐pandemic period, EGARCH models with Johnson's and Pearson Type IV residual distributions provide better VaR forecasts, whereas during the high‐volatility period, GAS models with skewed Student's and skewed generalized error distributions perform better. Finally, we use group transfer entropy to examine nonlinear risk transmission among cryptocurrencies. The evidence indicates substantial risk spillovers and shows that Bitcoin is not only a transmitter of risk but also a receiver of risk from other cryptocurrencies. Overall, the findings highlight the importance of accounting for common structural breaks, heavy‐tailed residual distributions, and nonlinear dependence when assessing cryptocurrency risk across changing market regimes.