RESEARCH · GLOBAL
Deep hedging for S&P 500 options learns systematic delta corrections but faces regime fragility
arXiv paper studies reinforcement-learning agents trained to hedge S&P 500 index options (2015–2023). Agents learned sensible delta haircuts vs. Black-Scholes but failed during 2022 regime shift, signaling brittleness to volatility shocks.
WHY IT MATTERS
AI-driven hedging shows promise in placid markets but breaks under stress. Risk teams must stress-test learned hedge policies, not assume they generalize across regimes.
Source: arXiv · 2026-05-22