Multi-step-ahead time series forecasting based on CEEMDAN decomposition and temporal convolutional networks

Recently, multi-step-ahead forecasting models are considered as robust and reliable methods to predict the value of time serier, especially in financial sector. This paper proposes a multi-step-ahead forecasting model based on CEEMDAN decomposition and temporal convolutional networks. The proposed strategy can be divided into three stages: (1) decompositing stage by using CEEMDAN decomposition, (2) predicting stage by using temporal convolutional networks, and (3) reconstructing stage by using the formula of CEEMDAN decomposition again. We also provide an evaluation study for VN-Index data from Ho Chi Minh Stock Exchange in order to compare the performance of our results to other benmark models.