README.md

Time-Causal VAE ⏱

The official code repository for paper Time-Causal VAE: Robust Financial Time Series Generator.

Environments 🔩

• Python 3.11 and PyTorch 2.5

• Run the following commands to install python libraries:

  • python -m pip install -r requirements/development.txt

• You can also use pip-tools to regenerate the requirements/development.txt from requirements/development.in by the following command:

  • python -m piptools compile requirements/train.in --output-file requirements/train.txt
  • and then run python -m pip install -r requirements/development.txt

Training 🧗🏻

You can find a example notebook notebooks/example.ipynb for the training pipeline.

Datasets and Trained Weight 📦

You can find the evaluation of trained models in the notebooks:

• Black-Scholes
• Heston
• Path dependent volatility model
• S&P500 and VIX
• Toy 2d examples

The trained models weights and training configurations are save in trained_models

Reference

• NeuralHedge for log-utility maximization and mean-variance portfolio problem.
• https://github.com/HeKrRuTe/OptStopRandNN for optimal stopping evaluation.
• https://github.com/stephaneckstein/aotnumerics for adapted Wasserstein distance estimation
• https://github.com/eitanrich/gans-n-gmms for NDB analysis
• https://github.com/luchungi/Generative-Model-Signature-MMD for estimating Signature MMD
• pythae and transformers for the trainer and callbacks.

Contact

If you have any questions, please feel free to reach me out!