Signature-based identification of volatility models from path geometry

Battle Count: 78

The paper addresses model identification for stochastic volatility, which is fundamental to option pricing, risk management, and volatility forecasting in quantitative trading. However, it does not directly propose trading strategies or portfolio optimization. The ability to identify volatility model classes from path geometry could inform model selection for derivatives pricing, hedging, and risk assessment. The rough volatility identification is particularly relevant for short-maturity options trading. The method is more of a diagnostic/identification tool than a direct trading application.

Requires understanding of rough path theory, iterated integrals, and path signatures. Implementation involves: (1) simulating paths from multiple stochastic volatility models (Heston, rough Bergomi, OU), (2) computing truncated signatures via vectorized GPU-accelerated code, (3) training XGBoost classifier. The signature computation is the most technically demanding part. GPU acceleration is needed for practical runtimes. The authors use standard Python packages but require custom implementations for path simulation and signature computation. Total pipeline runs in minutes on consumer hardware (RTX 3080 Ti, 128GB RAM).

The authors state that full code will be made available upon publication. They use standard Python packages (xgboost, NumPy, CuPy) and reference existing implementations for rough Bergomi simulation (Bennedsen et al. 2017, McCrickerd and Pakkanen 2018) and signature computation (Peter Foster's implementation). GPU acceleration is used. However, code is not yet publicly available at the time of the paper, and the paper relies on simulated data with specific parameter ranges that must be replicated exactly.