Patterns in time series to image phase and multi-step forecasting of financial indices

• Published in: Journal of King Saud University: Computer and Information Sciences
• Main Authors: Jun Wu, Cheng Wang, Zelin Zhang, Lang He, Yaxiong Chen, Wenming Cheng
• Format: Article
• Language: English
• Published: Springer 2025-06-01
• Subjects: Imaging timeseries; High frequency; Financial timeseries; Attention models
• Online Access: https://doi.org/10.1007/s44443-025-00077-4

Abstract Imaging time series have become an important method for constructing pattern features. The idea that transforming the original 1D time series into a 2D image makes properties from different levels in a spanned space. It realizes to analyze short-term autoregression and long-term relations of points at the same time. In this work, an image-time-series-based attention scheme has been investigated by high-frequency forecasting tasks with S&P500 and CSI300 datasets. By comparing non-parametric models and parametric models, as well as deep learning models, the proposed methods take the least residual. The MSE of the model in predicting the next 15 min’ logarithmic indices is 0.0038, and the MSPE value is 0.0001. Besides, the pattern-based CNN model has the fastest convergence rate when predicting the volatility at the same accuracy level. It verifies that the image representation of time series makes the potential information more efficient in analyzing financial time series.