PT - JOURNAL ARTICLE AU - Jiayi Du AU - Muyang Jin AU - Petter N. Kolm AU - Gordon Ritter AU - Yixuan Wang AU - Bofei Zhang TI - Deep Reinforcement Learning for Option Replication and Hedging AID - 10.3905/jfds.2020.1.045 DP - 2020 Oct 31 TA - The Journal of Financial Data Science PG - 44--57 VI - 2 IP - 4 4099 - https://pm-research.com/content/2/4/44.short 4100 - https://pm-research.com/content/2/4/44.full AB - The authors propose models for the solution of the fundamental problem of option replication subject to discrete trading, round lotting, and nonlinear transaction costs using state-of-the-art methods in deep reinforcement learning (DRL), including deep Q-learning, deep Q-learning with Pop-Art, and proximal policy optimization (PPO). Each DRL model is trained to hedge a whole range of strikes, and no retraining is needed when the user changes to another strike within the range. The models are general, allowing the user to plug in any option pricing and simulation library and then train them with no further modifications to hedge arbitrary option portfolios. Through a series of simulations, the authors show that the DRL models learn similar or better strategies as compared to delta hedging. Out of all models, PPO performs the best in terms of profit and loss, training time, and amount of data needed for training.TOPICS: Big data/machine learning, options, risk management, simulationsKey Findings• The authors propose models for the replication of options over a whole range of strikes subject to discrete trading, round lotting, and nonlinear transaction costs based on state-of-the-art methods in deep reinforcement learning including deep Q-learning and proximal policy optimization.• The models allow the user to plug in any option pricing and simulation library and then train them with no further modifications to hedge arbitrary option portfolios.• A series of simulations demonstrates that the deep reinforcement learning models learn similar or better strategies as compared to delta hedging.• Proximal policy optimization outperforms the other models in terms of profit and loss, training time, and amount of data needed for training.