Skip to content

Deep reinforcement learning for time series: playing idealized trading games

License

Notifications You must be signed in to change notification settings

suyinw/deep-RL-time-series

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

deep-RL-time-series

Deep reinforcement learning for time series: playing idealized trading games

https://arxiv.org/abs/1803.03916

In this paper I explored deep reinforcement learing as a method to find the optimal strategies for trading. I compared several neural networks: Stacked Gated Recurrent Unit (GRU), stacked Long Short-Term Memory (LSTM), stacked Convolutional Neural Network (CNN), and multi-layer perception (MLP). I designed two simple trading games aiming to test if the trained agent can 1) capture the underlying dynamics (to be used in mometum trading), and 2) utilize the hidden relation among the inputs (to be used in arbitrage trading). It turns out that GRU performs best. However as these are just simplified worlds for the agent to play with, more further investigation is deserved.

About

Deep reinforcement learning for time series: playing idealized trading games

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%