Evaluating economic policy uncertainty forecasts via TimesNet time series modeling of IoT and digital circular economy historical indicators

The development of the Internet of Things (IoT) and the digital circular economy generates vast amounts of high-frequency time series data with multi-scale cycle characteristics. Economic policy uncertainty (EPU) is a key indicator for assessing market confidence and financial risks. Traditional forecasting methods often overlook the dynamic signals in IoT and circular economy data. This paper proposes a multi-scale time series modeling approach based on TimesNet, which integrates 1D→2D time series mapping, Inception convolution, and temporal self-attention (TSA) mechanisms to extract key features for multi-step EPU prediction. We use the EPU indices of the United States and China as the target sequence and incorporate OPSD energy system IoT data as input features. The results show that IoT data enhances forecasting by capturing intricate, dynamic patterns, outperforming LSTM, Autoformer, and DLinear models in terms of RMSE, MAE, and MAPE. On the EPU dataset, MAE is reduced to 25.20, and MAPE is reduced to 8.12%, demonstrating superior accuracy. However, the method requires significant computational resources, especially for long sequences or high-frequency data, which may impact real-time forecasting deployment. This study advances EPU prediction by integrating IoT data and multi-scale modeling, offering a more proactive tool for policy decision-making.