Encoder Decoder Generative Adversarial Network Model for Stock Market Prediction
Bahadur Yadav, Sanjay Kumar Mohanty
TL;DR
The paper introduces EDGAN, a GRU-based encoder–decoder GAN for long-horizon stock price forecasting that conditions generation on static and dynamic covariates and uses a temporal decoder with residual connections and windowing to capture temporal dynamics. The generator encodes a compact latent representation from historical prices and covariates, then decodes with a temporal refinement, while a CNN+MLP discriminator judges realism of forecast sequences. Empirical results on real-world stock data show EDGAN achieving higher forecasting accuracy and training stability than standard GANs, WGAN-GP, and DRAGAN across multiple assets. This approach provides a scalable, robust framework for GAN-based time-series forecasting in volatile financial markets.
Abstract
Forecasting stock prices remains challenging due to the volatile and non-linear nature of financial markets. Despite the promise of deep learning, issues such as mode collapse, unstable training, and difficulty in capturing temporal and feature level correlations have limited the applications of GANs in this domain. We propose a GRU-based Encoder-Decoder GAN (EDGAN) model that strikes a balance between expressive power and simplicity. The model introduces key innovations such as a temporal decoder with residual connections for precise reconstruction, conditioning on static and dynamic covariates for contextual learning, and a windowing mechanism to capture temporal dynamics. Here, the generator uses a dense encoder-decoder framework with residual GRU blocks. Extensive experiments on diverse stock datasets demonstrate that EDGAN achieves superior forecasting accuracy and training stability, even in volatile markets. It consistently outperforms traditional GAN variants in forecasting accuracy and convergence stability under market conditions.