AI-DRIVEN DV01 HEDGING OPTIMIZATION VIA NEURAL SURROGATE MODELING AND STRUCTURED LEAST-SQUARES ACCELERATION

An AI-assisted framework for accelerating DV01-based interest-rate hedging is presented. Classical bumpand-revalue DV01 computation is computationally intensive when applied to large portfolios or complex pricing engines. The proposed framework integrates a neural operator surrogate for DV01 prediction and a graph neural network (GNN) for learning sparsity patterns in the DV01 matrix. A trust-region reflective (TRF) least-squares solver then exploits surrogate DV01 evaluations and learned sparsity to reduce computational cost while preserving the original hedging formulation. A controlled numerical experiment is further conducted to compare the baseline TRF solver and the proposed hybrid AI-LS method. The results highlight substantial reductions in runtime while maintaining hedging accuracy.