Selective Forgetting in Option Calibration: An Operator-Theoretic Gauss–Newton Framework

Battle Count: 74

Highly relevant for quantitative trading desks and risk management systems that rely on continuously recalibrated option pricing models. The framework enables efficient removal of corrupted/stale quotes without full recalibration, which is critical for maintaining model accuracy in fast-moving markets. Directly applicable to options trading desks, volatility surface maintenance, and regulatory compliance in financial institutions. However, it is more of an infrastructure/model-management tool than a direct trading strategy generator. The computational speedup (4-5 orders of magnitude) is significant for systems processing thousands of option books daily.

Moderate complexity. The core algorithm is algebraically simple (subtracting per-quote statistics from cached aggregates), but requires: (1) a working Heston (or other) pricing engine with Jacobian computation, (2) proper caching infrastructure for sufficient statistics, (3) shard management for the recompute variant, (4) numerical linear algebra (Cholesky factorization, condition number monitoring), and (5) careful handling of relinearization points. The theoretical framework is well-developed but practical deployment requires integration with existing calibration pipelines. The fast refactor operator is simpler to implement than the sharded recompute variant.

The paper provides detailed algorithmic descriptions, mathematical formulations, and synthetic data generation parameters (Heston parameters, grid specifications, noise levels, shard sizes). Scripts are mentioned as Python-based. However, no GitHub repository or code link is provided. The synthetic nature of experiments aids reproducibility, but the lack of published code limits full replication. All parameters for the Heston model, Fourier-Simpson integration settings, and experimental configurations are explicitly stated.