Optimizing Derivative Hedging Strategies: RL Approach and DNN Approach

---

🎓 Capstone Spotlight: Reinforcement Learning and Deep Neural Networks for Hedging

📄 Optimizing Derivative Hedging Strategies: RL Approach and Deep Neural Networks Approach

This capstone research investigates how Reinforcement Learning (RL) and Deep Neural Networks (DNNs) can improve derivative hedging under realistic market frictions.

📌 Instead of relying on traditional delta hedging, we train RL agents to dynamically adjust hedge ratios by minimizing both expected cost and risk (standard deviation), accounting for transaction costs and volatility shifts.

🧪 We compared:

• RL-based methods (e.g. Q-learning)
• Deep learning models (RNN, TCN, Transformer)
• Classical delta hedging

📊 Results:

• RL and DNN strategies outperformed traditional hedging in volatile regimes
• Transformer-based models offered high adaptability under complex dynamics
• Learned policies adapted to cost tradeoffs and gamma exposure

---