Pair-Trading-with-RL-DDDQN

Description

Optimize pair trading performance with reinforcement learning algorithm (DDQN and DDDQN).

Structure

• data
  • *.csv: daily stock price downloaded from Yahoo Finance
  • get_data.ipynb: download data from Yahoo Finance
  • process.ipynb: explore cointegration relationship of two stocks
  • select_pair.ipynb: use distance approach and cointegration approach to select appropriate stock pair for pair trading
• pre-train
  • main.ipynb: create a base model composed of convolution layers, train the model on specific stock pair to predict the trend of spread. The well-trained base model will later be used in RL models train with DDQN and DDDQN.
• train
  • double_dqn.ipynb: use double dqn algorithm to train RL agent in pair trading
  • dueling_double_dqn.ipynb: use dueling double dqn algorithm to train RL agent in pair trading
  • pair trading environment is defined in these two files
• test
  • baseline_testing.ipynb: test the performance of baseline approach (rule-based)
  • nn_testing.ipynb: test the performance of RL agent trained by DDQN and DDDQN
  • xgboost_testing.ipynb: test the performance of XGBoost algorithm

Run

All scripts are written in jupyter notebook, you can run it on colab or kaggle.

Caution

• Make sure the the path of dataset is correct in scripts.
  
• The version of TensorFlow >= 2.0

Resource

You may use these resources to uderstand this project.

• Presentation (Slide, Video)
• Report