Financial Time Series Prediction Based on Adversarial Network Generated by Attention Mechanism

Abstract

Financial time series forecasting is a technology to make reasonable prediction on the development of future data according to historical laws. It is of great significance to government departments, investment institutions and investors. However, with the continuous improvement of the development of the financial market, the amount of time series data in the financial market is getting larger and larger, and the data generation and accumulation speed is fast. The traditional measurement model cannot meet the processing requirements of big data for nonlinear and high-noise data. In this paper, the financial time series prediction based on dual attention mechanism generating admittedly network is proposed. Firstly, the input attention mechanism is introduced into the generator to adaptively select the input features because the input features of the financial time series are too many and difficult to be selected adaptively. Secondly, the time attention mechanism is introduced into the generator to capture the long time dependence of financial time series, which is difficult to capture. We use the CSI 300 index to verify the prediction performance of the model, and the mean square error is 0.0012. The experimental results show that the model can adaptively select the input features, capture the long-term dependence of financial time series, reduce the prediction error of the model, and improve the prediction accuracy of the model.