This repo is the code for this paper. Deep reinforcement learing is used to find optimal strategies in these two scenarios:

• Momentum trading: capture the underlying dynamics
• Arbitrage trading: utilize the hidden relation among the inputs

Several neural networks are compared:

• Recurrent Neural Networks (GRU/LSTM)
• Convolutional Neural Network (CNN)
• Multi-Layer Perception (MLP)

More about deep reinforcement learning (deep Q-learning):

• https://keon.io/deep-q-learning/ and its code on GitHub
• Google’s DeepMind published its famous paper Playing Atari with Deep Reinforcement Learning

Python version: python 3.6

pip install -r requirements.txt

You can get all dependencies via the Anaconda environment file, env.yml:

conda env create -f env.yml

• before start at the 1st time, generate database: python sampler.py
• call the main function: python main.py

You can play with model parameters (specified in main.py), if you get good results or any trouble, please contact me at [email protected]