Multi-period Learning for Financial Time Series Forecasting

The Multi-period Learning Framework (MLF) has been introduced to address the critical need for accurate financial time series forecasting, which is influenced by both short-term public opinions and long-term market trends. Traditional forecasting models often rely on single-period inputs, limiting their effectiveness. MLF enhances performance by integrating multi-period inputs through three key modules: Inter-period Redundancy Filtering, which eliminates redundant information for better self-attention modeling; Learnable Weighted-average Integration, which effectively combines forecasts from different periods; and Multi-period self-Adaptive Patching, which ensures balanced representation across periods. Additionally, the Patch Squeeze module is designed to streamline self-attention modeling. The development of MLF is a significant step forward in financial forecasting, promising improved accuracy and efficiency, and the associated codes and datasets are made available for further resea…