Crashing Together, Rallying Apart: Dynamic Conditional Tail Dependence in Cryptocurrency Markets

Battle Count: 68

Highly relevant for quantitative trading in cryptocurrency markets. The findings directly challenge the diversification premise of crypto portfolios, showing that intra-crypto diversification fails on the downside. The dynamic extremal graph provides a real-time systemic risk monitoring tool. The eightfold understatement of joint crash probability by covariance-based models has direct implications for position sizing, portfolio construction, and risk management. The Bitcoin-Ethereum core hierarchy informs asset selection. However, the paper is descriptive rather than predictive, and does not directly propose trading strategies.

High complexity. Requires expertise in extreme value theory, graphical models, multivariate statistics, and time series econometrics. The pipeline involves: (1) AR(1)-GJR-GARCH(1,1)-t marginal filtering, (2) empirical marginal transformation to Pareto scales, (3) penalised HR extremal log-likelihood estimation with L1 sparsity, (4) Bayesian GGM estimation, (5) dynamic sliding window re-estimation across 89 windows, (6) network statistics computation, (7) moving-block bootstrap inference, and (8) Monte Carlo simulation for economic quantification. The HR graphical model estimation is non-standard and requires specialized software (R packages for graphical extremes). Cross-validation for penalty selection adds computational burden.

All code is publicly available on GitHub (https://github.com/manueleleonelli/HR_crypto). Data is sourced from Yahoo Finance (publicly available). The paper provides detailed methodology, parameter choices, and robustness checks in appendices. Software (R) is specified. Marginal filtering diagnostics are fully reported. Robustness to threshold, window length, and estimator choice is examined.