Fast simulation of Volterra processes using random Fourier features with application to the log-stationary fractional Brownian motion

Battle Count: 57

The paper is highly relevant to quantitative finance, particularly rough volatility modeling which has become a central paradigm since the rough Bergomi (2016) and rough Heston (2019) models. The S-fBM kernel and Log S-fBM model directly address the roughness of realized volatility (H≈0.1 for indices, H≈0.01 for single stocks). The fast simulation scheme enables efficient Monte Carlo pricing and calibration of rough volatility models, which is critical for derivatives pricing, risk management, and algorithmic trading strategies. However, the paper focuses on the simulation methodology rather than direct trading applications, and does not include option pricing experiments or hedging strategies. The computational speedup (up to 200x) is practically significant for real-time Monte Carlo applications.

The RFF-Euler scheme (Algorithm 3) is relatively straightforward to implement with O(N×M) complexity. However, the full pipeline requires: (1) computing the spectral density via hypergeometric functions (1F2) or Bessel functions, which needs specialized numerical libraries; (2) implementing HMC with leapfrog integration, requiring gradient computation of the log spectral density; (3) handling the positive definiteness condition (0 < H ≤ (3-d)/4); (4) managing truncation of the hypergeometric series with error control; (5) the GMM parameter estimation framework. The theoretical error bounds are complex but the actual simulation algorithm is moderate in complexity. The HMC tuning (step size, number of leapfrog steps, mass matrix) requires careful calibration.

The paper provides detailed algorithms (Algorithm 1: Euler scheme, Algorithm 2: HMC, Algorithm 3: RFF scheme), explicit mathematical formulas for the spectral density (Theorem 3, Eq. 57), positive definiteness conditions (Theorem 4), and error bounds (Theorems 1-7). Numerical parameters are specified (M=8000, ν²=50, H=0.1, T values, time lag sequences). However, no code repository is mentioned, and the HMC implementation details (step size ε, number of leapfrog steps L, mass matrix M) are not fully specified for reproducibility. The hypergeometric function evaluation and Bessel function computations would require specialized numerical libraries.