Abstract:  The accurate forecasting of financial time series data holds a significant position in the operation and management of financial markets. Grounded in the notions of neural networks and ensemble learning, the convolutional neural network ( CNN), long short-term memory ( LSTM) network, and autoregressive moving average ( ARMA) model are integrated within an ensemble framework to put forward a novel ARMA-CNN-LSTM model for forecasting financial time series data. The spatiotemporal characteristics within the data are modeled by the CNN-LSTM model, while the autocorrelation features of the data are simultaneously modeled by the ARMA model, achieving hybrid modeling of linear and nonlinear traits in financial time series data. The experimental results show that, compared with the baseline individual models, the proposed model demonstrates excellent performance in both the accuracy and robustness on predicting financial time series data.

Key words: financial time series, convolutional neural network, long short-term memory, autoregressive integrated moving average, ensemble forecasting models