Reinforcement Learning for Option Hedging: Static Implied-Volatility Fit versus Shortfall-Aware Performance

Battle Count: 95

Highly relevant for quantitative trading desks involved in options market-making and hedging. The paper directly addresses the practical disconnect between static IV calibration and realized hedging P&L under transaction costs. The shortfall-aware framework is particularly relevant for risk-managed trading desks. The bidirectional architecture (QLBS for cost-aware stability, RLOP for capital-constrained desks) provides actionable guidance for practitioners. Empirical validation on SPY and XOP during both crisis and calm periods strengthens practical relevance.

Moderate to high complexity. Requires implementing two distinct RL frameworks (backward value-based QLBS and forward replication-based RLOP), neural network policy parametrization with ResNet architecture, REINFORCE training with baseline, Monte Carlo simulation of GBM paths, transaction cost modeling, and comprehensive backtesting infrastructure. The reward shaping with stacked maturities in RLOP adds additional complexity. However, the algorithms are well-specified mathematically and use standard RL components.

The paper provides detailed mathematical formulations, algorithm descriptions, and hyperparameters (learning rate 10^-4, Adam optimizer, REINFORCE with baseline). Data is publicly available from DoltHub. However, no code repository is explicitly linked, and some implementation details (e.g., exact network architecture dimensions, training epochs, number of Monte Carlo paths) are not fully specified. The proof for Proposition 1 is referenced to supplementary materials.