Long-Range Dependence in Financial Markets: Empirical Evidence and Generative Modeling Challenges

Battle Count: 67

Highly relevant for quantitative trading in several ways: (1) LRD in volatility affects risk models used in position sizing and portfolio construction; (2) the finding that Quant GANs fail to capture LRD is critical for practitioners using synthetic data for backtesting and stress testing; (3) the ARFIMA-FIGARCH framework provides actionable models for volatility forecasting; (4) the COVID-19 regime shift analysis informs adaptive trading strategies; (5) however, the paper is primarily diagnostic/empirical rather than proposing a new trading strategy directly.

Moderate-to-high complexity: (1) R/S and DFA are relatively straightforward to implement; (2) ARFIMA-FIGARCH with Student's t requires specialized econometric software (e.g., rugarch in R); (3) Quant GAN training with TCN architecture requires deep learning infrastructure (~1h per asset); (4) full-ensemble validation of 10,000 paths per asset with 180,000 Hurst estimations adds computational burden; (5) segmented MF-DFA with Legendre transforms adds mathematical complexity; (6) benchmark comparisons require multiple model implementations.

The paper uses publicly available data from Yahoo Finance, provides a GitHub repository link for the Quant GAN implementation (https://github.com/KseniaKingsep/quantgan), specifies default hyperparameters, and reports detailed estimation results with confidence intervals and p-values. The preprocessing steps (synchronization, normalization) are clearly described. However, exact training seeds and hardware specifications are not fully detailed, and the approximate training time (~1h per asset) depends on environment.