An Improved Stock Price Prediction Model Based on Generative Adversarial Network

As an important platform for corporate financing and investor trading, the stock market plays an important role in economic development and capital allocation. How to predict stock price effectively has become the object of many scholars' research. In recent years, with the swift advancement of Generative Adversarial Networks (GANs), their performance in the field of stock price prediction has attracted wide attention. This study proposes an improved Wasserstein Generative Adversarial Network-Gradient Penalty (WGAN-GP) stock price prediction model, with Gated Recurrent Unit (GRU) as a generator. Compared with traditional GANs, the choice of GRU ensures that improved WGAN-GP has excellent prediction accuracy while being stable in training, and the inclusion of Residual Neural Network in the discriminator significantly improves the predictive performance of the model. In addition, Bull/Bear Ratio (BR) is introduced into the model as an input feature, which can well reflect investor sentiment.