Portfolio Management

This is an implementation of the portfolio management solution described in the following paper: A Deep Reinforcement Learning Framework for the Financial Portfolio Management Problem. The purpose of this project was educational & to make the code more flexible than the original implementation for further usages in different applications.

Requirements

• Code tested with Python 3.7.9, pytorch 1.8.0, cuda 11.1 & cudnn 8.0 on Windows 10
• Requirements are installed from the Pipfile.lock ( all commands are run at the Pipfile folder level).

To install your dependencies run:

pipenv sync

To activate the environment:

pipenv shell

Finally install pytorch ( you should be in the pipenv environment while doing this step, pipenv does not support --find links / -f):

pip install torch==1.8.0+cu111 torchvision==0.9.0+cu111 torchaudio==0.8.0 -f https://download.pytorch.org/whl/torch_stable.html

Train

python train.py
--config_file config/crypto.yaml
--storage_path
/path/to/data/logging

Test

python backtest.py
--config_file config\crypto.yaml
--storage_path
/path/to/data/logging

Concrete Backtest

• These are the results after running the following command:

python backtest.py
--config_file
config\crypto_test_2017.yaml
--storage_path
\path\to\aiie_crypto_storage

Prices

Relative results

• After a six month period of trading the agent made a +40% profit. But we can see that the portfolio value was not stable at all. Therefore, there is place for a lot of improvement.