cgurkan / pairstrade-fyp-2019

Trying to replicate the results of Pair Trading

Yuri Updates

The authors added helpful instructions how to run the code here. I ran the rl_train code by using these parameters:

--job_name train_0_1_2_test_3 --run_mode "train" --train_indices 0 1 2 --test_indices 3

I added datasets used by the authors here. However, when I proccessed the raw data into the transformed data via the process_data.py script, the resulting log prices were different than the data used by the authors. After training and testing the reinforcement code with my processed raw data, my results were marginally positive. I believe the authors' dataset may have had look ahead bias when they originally transformed the data. My results are here (slightly positive but not large enough to warrant trading), while the results I was able to achieve using the authors' dataset is located here (40%+ as mentioned in their paper).

I made one adjustment to the code, and believe another is warranted.

1. The incur_commission script added 10%+ cost over the course of a year. Since the strategy generally executed few trades, I expected this to be lower. Based on my knowledge, this should be 0.25% - 0.50%, so I divided the cost by 50.

2. I believe the compute reward should include the shorted value to self.port_val_minus_com[i] when the shorted value goes below zero by adding spv_nex. In a real-world environment, shorted values can go "below zero" (i.e. you owe money to cover the short). I did not make this change to the code though.

pairstrade-fyp-2019

Final year project at HKUST. We tested 3 main approaches for performing Pairs Trading:

• distance method
• cointegration method (rolling OLS, Kalman Filter)
• reinforcement learning agent (proposed)

Final report can be found here. Presentation slides can be found here.

FYP members: myself, Gordon, Brendan

How to get started?

• Run ./setup.sh to install all dependencies

Note

• In our experiments, we used financial data taken from the Interactive Brokers platform, which is not free. Due to their regulations, we cannot release the financial data used in our experiments to the public. Feel free to use your own price data to perform experiments.

Disclaimer

• The strategies we implemented have not been proven to be profitable in a live trading account
• The reported returns are purely from backtesting procedures, and they may be susceptible to lookahead bias that we are not aware of