deep_call_hedger

A Python package which implements the deep hedging algorithm. Target is learning optimal hedges through a deep feed forward neural network (DFNN), in order to minimize terminal loss when hedging a written European call option.

FEATURES

• Simulates stock prices following a geometric Brownian motion
• Calculates call prices under the Black Scholes model
• Trains a DFNN at each timestep on stocks, calls & strike to minimize terminal error
• Runs predictions using the trained model
• Benchmarks model performance against Black Scholes visible in charts, KPIs & KRIs

REQUIREMENTS

• Python 3.11+ (tested with 3.11.9)
• Required dependencies (see requirements.txt for details):
  numpy==2.1.3
  tensorflow==2.19.0
  scipy==1.16.0
  matplotlib==3.10.5
  seaborn==0.13.2
  

INSTALLATION

1. Clone the repository:
   git clone https://github.com/hb84ffm/deep-hedging.git
   cd deep_call_hedger
   

2. Create & activate your virtual environment:
   python3 -m venv venv
   source venv/bin/activate # On Mac/Linux
   venv\Scripts\activate # On Windows

3. Install dependencies:
   pip install -r requirements.txt

USAGE

1. Train the deep hedging model (or use the pretrained model deep_hedging_64.keras):
   python main_training.py
   -> Saves a trained ".keras" model to the models/ folder.

2. Run prediction & analysis:
   python main_prediction.py
   -> Uses the trained model to run simulations & generate plots.

PACKAGE STRUCTURE

deep_call_hedger/
├─── __init__.py
├─── main_prediction.py             # Orchestrates the prediction
├─── main_training.py               # Orchestrates the training
├─── dh_model/
     ├─── __init__.py
     ├─── dh_model.py               # Designs the model (computational graph) by Keras functional API
├─── models/                        
     ├─── deep_hedging_64.keras     # Pretrained deep_hedging_64.keras model trained across 64 timesteps
├── options/
     ├─── __init__.py
     ├─── bs.py                     # Black Scholes calculator for European calls
├── prediction/
     ├─── analysis.py               # Derives analytics (charts, KPIs & KRIs) on predicted data
     ├─── prediction.py             # Generates prediction data & runs prediction using the trained model 
├── stocks/                  
     ├── __init__.py
     ├─── stocks.py                 # Stock simulation
├── training/                   
     ├─── __init__.py
     ├─── training.py               # Compiles the model & starts the training

EXAMPLE WORKFLOW

See provided Jupyter notebook example for explanation.

CREDITS

Josef Teichmann's implementation

AUTHOR

For questions or feedback reach out to me via: GitHub.