Time-dependent weighted directed networks of cryptocurrency interaction from high-frequency returns

Battle Count: 94

The paper provides valuable insights into the dynamic hierarchy of influence among cryptocurrencies, which can inform relative strength strategies, sector rotation, and risk management. The identification of Ethereum as a persistent leader and Bitcoin's declining influence has direct trading implications. The network structure reveals which assets transmit information most effectively, useful for pairs trading and statistical arbitrage. However, the paper is primarily descriptive/analytical rather than prescriptive for trading strategies, and does not provide backtested trading signals or portfolio construction recommendations. The high-frequency (1-minute) data and weekly network construction could be adapted for shorter-horizon trading applications.

The core methodology (pairwise Granger causality via VAR, network construction, BH-FDR correction) is well-established and implementable using standard statistical packages (statsmodels in Python, vars in R). However, the full pipeline involves: (1) high-frequency data aggregation to VWAP, (2) stationarity testing for all assets across 275 windows, (3) pairwise VAR estimation for up to 390*389 ordered pairs per week, (4) BH-FDR correction, (5) network construction and analysis, (6) PC1 regression for robustness. The computational cost of ~150,000+ pairwise tests per week across 275 weeks is non-trivial but manageable with parallelization. The network analysis (CCDF fitting, power-law estimation, ranking) adds moderate complexity.

The paper provides detailed methodology including exact equations for Granger causality, VAR model specifications, BIC-based lag selection, BH-FDR correction procedure, and ADF stationarity testing. Data source (Kraken Crypto Exchange) is publicly available with a URL provided. Parameters such as significance levels (alpha=0.01 for ADF, q=0.05 for FDR), maximum lag (10), and window sizes (10,080 data points per week) are explicitly stated. However, no code repository is provided, and the exact list of 390 cryptocurrencies and preprocessing details for edge cases could be more explicit.