A Multivariate Stock Prediction Method Based on CNN-Transformer

Financial time series data are characterized by nonlinearity and high dimensionality, which makes traditional prediction models have certain limitations and makes it difficult to capture the deep inner relationships. Convolutional neural network (CNN) is widely favored for its excellent performance in feature extraction, which is suitable for processing image and time series data; meanwhile, the Transformer model, with its unique self-attention mechanism, can effectively solve the problem of long-distance dependence in sequence data. At present, only a single price feature is usually considered in the stock price prediction problem. In this paper, we propose a CNN-Transformer stock price prediction method combining multichannel parallel convolution and Transformer, which is experimentally explored on the stock dataset of Ping An Bank. The experimental results show that our model significantly improved prediction accuracy compared to the other three deep learning models.