Generative Adversarial Networks (GANs) in finance: an overview

This code is referenced in the paper https://arxiv.org/abs/2106.06364.

It explores the use of Generative Adversarial Networks (GANs) for synthetic financial time series generation.

Overview

The notebook (gans_in_finance.ipynb) implements and compares 10 GAN variants applied to real financial data (loaded via yfinance):

DCGAN — standard deep convolutional GAN
LSGAN — least squares loss variant
SAGAN — self-attention GAN
WGAN — Wasserstein GAN
WGAN-GP — Wasserstein GAN with gradient penalty
DRAGAN — deep regret analytic GAN
RaGAN / RaLSGAN — relativistic loss improvements
YLgan — further loss variant
BigGAN / BigGAN-deep — large-scale GAN architectures

Each model is trained on 1D financial time series windows and evaluated on its ability to generate realistic synthetic data. Originally developed for Google Colab, adapted to run locally with TensorFlow.

The default setting is DCGAN at 30 epochs, for ideal training configuration, refer to the related paper.

Requirements

Python 3.12
A virtual environment (recommended)

Setup

1. Create and activate a virtual environment

python3.12 -m venv .venv
source .venv/bin/activate   # macOS / Linux
# .venv\Scripts\activate    # Windows

2. Install dependencies

pip install \
  "tensorflow==2.16.2" \
  "numpy<2" \
  "yfinance==1.2.0" \
  "scipy==1.17.1" \
  "pandas==3.0.1" \
  "matplotlib==3.10.8" \
  "ipykernel==7.2.0"

Important: NumPy must be pinned to <2. TensorFlow 2.16.2 was compiled against NumPy 1.x and will fail to import with NumPy 2.x (numpy.core was removed in NumPy 2.0).

3. Register the kernel with Jupyter

python -m ipykernel install --user --name=ba_code --display-name "BA Code (.venv)"

4. Launch the notebook

jupyter notebook gans_in_finance.ipynb
# or
jupyter lab gans_in_finance.ipynb

Select the "BA Code (.venv)" kernel when prompted.

Known Issues

NumPy 2.x incompatibility: TensorFlow 2.16.2 requires numpy<2. If you see ImportError: numpy.core.umath failed to import, run pip install "numpy<2" to downgrade.
yfinance API changes: Data download uses yfinance. If the Yahoo Finance API changes, the download_data() function in the notebook may need adjustment.

eckerli/BA_code