Mixture-VAE: A Generalized Framework for Financial Regime Identification

Mixture-VAE is a versatile framework for classification tasks, specifically financial regime identification. This repository provides a python implementation including usage examples and experiment reproductions.

Overview

The Mixture-VAE framework is designed for the classification of financial regimes by leveraging a mixture-model in the latent space of variational autoencoders. This approach offers flexible and learnable distance measure for the classification models in a data-driven way.

---

Features

• Financial Regime Identification:
  Apply advanced VAE models to identify and classify financial market regimes.
• Synthetic Experiment Reproduction:
  Detailed notebooks enable the reproduction of synthetic experiment results with hyper-parameter search.
• Baseline Comparisons:
  Compare Mixture-VAE performance against baseline models such as HMM, K-Means, and jump models.
• Modular Implementation:
  Clear separation of core components and wrappers for easy extension and maintenance.

---

Installation

To install the required dependencies, run:

pip install -r requirements.txt

---

Usage

• Example Notebook:
  Explore an end-to-end usage example with the example_usage.ipynb notebook.
• Model Implementation:
  The key implementation is provided in model/mixture_vae.py. For batch fitting and inference, refer to the wrapper in model/vae_module.py.

---

Experiments

To reproduce the experimental results, follow these steps:

• Synthetic Experiments:
  Run the notebooks located in exp_notebooks/synthetic_notebooks/mixture-vae for synthetic experiment results with hyper-parameter searching.

• Baseline Models:
  Reproduce baseline results by executing the notebooks in exp_notebooks/synthetic_notebooks/baselines.

• Synthetic Data Generation:
  Learn how the synthetic data is generated by checking out exp_notebooks/synthetic_generation.ipynb.

---

Project Structure

• Core Model Implementation:

  • model/mixture_vae.py – Primary implementation of the Mixture-VAE model.
  • model/vae_module.py – Wrapper module for batch fitting and inference.

• Baseline Models (Wrapped):

  • model/hmm_module.py
  • model/kmeans_module.py
  • model/jump_module.py

• Experiments:

  • Notebooks for synthetic experiments and baseline comparisons are located under exp_notebooks/.

---

Citation

If you use this repository in your research or projects, please cite it as follows:

@misc{mixturevae,
  author = {Yuqi Nie and John M. Mulvey and H. Vincent Poor and Chenyu Yu and Hao Huang},
  title = {Mixture-VAE: A Generalized Framework for Financial Regime Identification},
  year = {2025},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/yuqinie98/Mixture-VAE}},
}

---

License

This project is licensed under the MIT License.