ReinforcementPortfolio

Set of codes for the paper Deep Reinforcement Learning for Portfolio Management: A Simulation Study

View documentation (currently in progress)

TODO List

High priority

• Documentation
• Example usage
• Pre-stored SimulatorEnv parameters.
• Generalize SimulatorEnv to have the same components: a simulator which generates prices, returns, features. => this can then be used to unify state(), reset!(), and env(action). Note: must keep track of returns for rewards such as SR.

Medium priority

• Make agents work for any environment type, not only GPEnv -> Solved throught the generalization of SimulatorEnv
• Add RNG everywhere instead of resetting seeds within functions? -> Unsure
• Gradient clipping for PPO
• CPU/GPU agnosticism
• Recurrent policies

Low priority

• State space for the environment
• GAIL