/ForexRL

A Deep Reinforcement Learning model for high volume and frequency Forex Portfolio Management

Primary LanguagePythonApache License 2.0Apache-2.0

ForexRL

A Deep Reinforcement Learning model for high volume Forex Portfolio Management

GitHub GitHub code size in bytes GitHub repo size GitHub last commit

Attributes

Constraints Description
Features FX market is represented via 512 features in X_train and X_test.
Summary 512 features summarizes the price-actions of 10+1 assets in past 10 days.
Return Hourly log returns of assets during train & test periods are in y_train and y_test.
Risk Calmar, Sortino, Omega ratio(s), etc. (Included but limited)

GPU Results

Constraints Result
Max. Drawdown 6.24%
Sortino Ratio 10.10x
Sharpe Ratio 3.15x
Stability 91.31%
Tail Ratio 3.57x
Value at Risk -0.84%

CPU is not tested

Install and Run

Note: Don't forget to unzip the Xtrain tar file

pip install -r requirements.txt

Pytorch implementation with GPU and CPU version seprately with shared auto encoder

python gpu.py

Run autoencoder for plots and results

python autoencoder.py

Results

Daily Portfolio Balances DPB Annual Cumulative Return Weekly Portfolio Log Return Autoencoder Interpretation

Caution: Don't use it for live forex market.