A Deep Learning Framework for High-Frequency Signal Forecasting Based on Graph and Temporal-Macro Fusion

With the increase in trading frequency and the growing complexity of data structures, traditional quantitative strategies have gradually encountered bottlenecks in modeling capacity, real-time responsiveness, and multi-dimensional information integration. To address these limitations, a high-frequency signal generation framework is proposed, which integrates graph neural networks, cross-scale Transformer architectures, and macro factor modeling. This framework enables unified modeling of structural dependencies, temporal fluctuations, and macroeconomic disturbances. In predictive validation experiments, the framework achieved a precision of 92.4%, a recall of 91.6%, and an F1-score of 92.0% on classification tasks. For regression tasks, the mean squared error (MSE) and mean absolute error (MAE) were reduced to 1.76×10−4 and 0.96×10−2, respectively. These results significantly outperformed several mainstream models, including LSTM, FinBERT, and StockGCN, demonstrating superior stability and practical applicability.