Rethinking Synthetic Realism: Compatibility, Not Fidelity, Drives Hedging Performance

This paper proposes a decision-oriented, compatibility-centric evaluation perspective for synthetic price paths used in deep hedging strategy training, a metric that captures differences in hedging risk between synthetic and real market environments. Theoretically, an upper bound for hedging performance degradation is established, pointing out that statistical realism and compatibility may be structurally divergent. Empirical evidence based on real market data shows that statistical realism cannot stably predict hedging performance, and the performance ranking of generators changes with the category of hedging strategy and task structure.