A Compound Generative Adversarial Network Designed for Stock Price Prediction Based on WGAN

In the stock market, the combination of historical stock data and machine learning methods has gradually replaced the investment method that relies solely on human experience. We have implemented a composite Generative Adversarial Network based on a pre-training model, which deeply analyzes the characteristics of stock trends and is more suitable for processing stock data than traditional methods. The models introduced in this manuscript including pre-training model and deep training model. The deep training model also includes the Generative Adversarial Network, ARIMA-Lasso units, and selectors. When dealing with the historical data set of American stock market in recent 3 years, we find that the optimal accuracy of our model is more than 84% in all experiments. In addition, we also compared this model with other excellent models, which also proved that this model is outstanding.