Scheduling method of national macroeconomic mixed frequency data based on internet of things technology

Abstract The MEM-MIDAS model proposed in this paper realizes the paradigm shift of macroeconomic monitoring from low-frequency lag to high-frequency real-time through a cross-frequency domain fusion architecture. The model has made significant breakthroughs in prediction accuracy. For example, out-of-sample tests show that the MSE (1.83 × 10 −3 ) and QLIKE (0.092) of the model are 38.4% and 35.2% lower than the optimal baseline LSTM-MIDAS, respectively. This is mainly due to the decomposition modeling of long-term and short-term fluctuations by the product structure and the enhanced capture of the persistence of bear market fluctuations by the asymmetric leverage effect mechanism ( γ parameter). In terms of warning timeliness, the model achieves a lead time of 14.7 days, and the precision (0.89) and recall (0.85) of the model are better than the existing solutions. The system adopts an edge-cloud collaborative framework, compresses the computing delay to within 500 ms through GPU acceleration, and supports minute-level decision-making. Robustness tests show that the model has strong adaptability to data noise and 20% data missing. In addition, parameter sensitivity analysis reveals the key role of weight decay coefficient ω 2 and macroeconomic parameter θ . Among them, ignoring macro factors ( θ = 0) will lead to a 23% decrease in forecasting performance. This technology can reduce the loss of policy errors by 0.5%–1.2% for every 1 percentage point increase in economic forecast accuracy, providing real-time and accurate technical support for national economic governance.