A hitchhikers guide to FinRL: A Deep Reinforcement Learning Framework for Quantitative Finance
Published in
9 min readSep 8, 2021
- FinRL is a deep RL library that aims to provide a framework to implement Quantitative finance with RL. So in this beginner's guide, we will start with Reinforcement learning components, and the directory structure of the library, then go through a code to understand the implementation and discuss other tutorials you can look into given your use case. This blog post assumes that you know Reinforcement learning basics, Policy gradient, and Actor-Critic methods like DDPG, PPO, SAC, TD3, and A2C. Also, check out this blog post.
REINFORCEMENT LEARNING FOR FINANCE SETTING
- There are three main components in Reinforcement learning, (i)State, (ii)Action and (iii) Reward. An RL agent observes the state space and then comes up with an action so that it can maximize the total reward going forward
STATE-SPACE
- State-space corresponds to observation space and the internal state of the agent. Observation space consists of market information like Open, Close, High, Low prices. It can consist of other information, like technical indicators: MACD (Moving Average Convergence Divergence), RSI (Relative Strength Index), CCI (Commodity channel index) and DMI (Directional Movement Index) [defaults…
