Universality and Heterogeneity of Stylized Facts in Cryptocurrency and Equity Markets

Battle Count: 76

Highly relevant for quantitative traders operating across crypto and equity markets. The findings suggest that while conventional risk models based on tail slopes, Hurst exponents, and multifractal widths may be applicable across both asset classes, they miss structural differences in temporal organization. The CECP and DHVG diagnostics provide additional model-adequacy checks. The high-frequency nature of the differences (strongest at 1-minute, weakening at 5-10 minutes) is particularly relevant for intraday and high-frequency trading strategies. The paper cautions against assuming interchangeability of pricing models across asset classes despite macroscopic statistical similarity.

The methodology involves multiple sophisticated components: (1) Volume-weighted composite price construction across exchanges; (2) Intraday deseasonalization with Gaussian smoothing; (3) DFA and MF-DFA with polynomial detrending; (4) Bandt-Pompe ordinal pattern analysis and CECP computation; (5) Directed HVG construction with degree-resolved analysis; (6) IAAFT surrogate generation with iterative convergence; (7) Monte Carlo statistical testing with multiple comparison corrections. Each component is well-established in the literature but the integrated pipeline with proper surrogate controls and robustness checks requires significant implementation effort. The DHVG construction on ~3 million data points and 100 surrogate realizations per asset per type adds computational cost.

The paper provides detailed methodology descriptions including specific parameters (D=5, tau=1 for CECP; k>=10 for DHVG high-degree; q in [-4,4] for MF-DFA; 100 surrogate realizations). Data sources are specified (Bitstamp, Coinbase, Kraken for crypto; CME for S&P 500 futures). Preprocessing steps (volume-weighted composite, intraday deseasonalization with 30-min Gaussian smoothing) are clearly documented. Supplementary material includes extensive robustness checks. However, no code repository is explicitly mentioned, and the exact data retrieval scripts are not provided.