Financial time series prediction is challenging due to the uncertainty of financial markets, especially in China’s stock market. This paper innovates to integrate sparse Gaussian Graph Model (GGM) information into Generative Adversarial Network (GAN) to forecast stock price for the Chinese A-share market. In the GAN, we use the Long Short-Term Memory (LSTM) as generate network and Convolution Neural Network (CNN) as discriminate network. As for the input of LSTM, the features of predicted stock are combined with correlated stocks which are selected by Sparse GGM. Experimental evidence shows that our model achieves better performance in the predition error and direction accuracy than the other models, such as ARIMAGARCHA, SVM, LSTM, GAN and GGM-LSTM.
Paper
Full text
Financial Time Series Prediction Based on GGM-GAN
Semantic Scholar · Computer Science · 2021
Abstract
Financial time series prediction is challenging due to the uncertainty of financial markets, especially in China’s stock market. This paper innovates to integrate sparse Gaussian Graph Model (GGM) information into Generative Adversarial Network (GAN) to forecast stock price for the Chinese A-share market. In the GAN, we use the Long Short-Term Memory (LSTM) as generate network and Convolution Neural Network (CNN) as discriminate network. As for the input of LSTM, the features of predicted stock are combined with correlated stocks which are selected by Sparse GGM. Experimental evidence shows that our model achieves better performance in the predition error and direction accuracy than the other models, such as ARIMAGARCHA, SVM, LSTM, GAN and GGM-LSTM.