KANHedge: Efficient Hedging of High-Dimensional Options Using Kolmogorov-Arnold Network-Based BSDE Solver

Battle Count: 70

Highly relevant for quantitative finance practitioners dealing with high-dimensional derivative pricing and hedging. The paper directly addresses the practical challenge of delta hedging in multi-asset portfolios, which is central to market-making and risk management desks. The CVaR-based hedging cost evaluation aligns with institutional risk management practices. However, the method is primarily for derivative desks rather than directional trading strategies.

The BSDE framework with Euler-Maruyama discretization is well-established. The main complexity lies in implementing KANs with B-spline activation functions, managing spline grid parameters, and ensuring numerical stability during training. The training algorithm is straightforward (Algorithm 1), but tuning KAN-specific hyperparameters (spline order, number of control points, regularization) adds complexity compared to standard MLP implementations. The unified European/American framework via penalty method simplifies architecture management.

The paper provides detailed mathematical formulations, algorithm pseudocode (Algorithm 1), and comprehensive experimental configurations with specific parameter values. However, no code repository is mentioned. The use of standard BSDE discretization and publicly available KAN implementations aids reproducibility, but specific hyperparameters (learning rate schedules, KAN spline grid sizes, number of layers) are not fully detailed.