Forecasting Foreign Trade Exports Using a Nadaraya-Watson and LSTM Composite Model

The uncertainty of the global economy has posed a severe challenge to China's foreign trade exports. Improving the reliability of predicting foreign trade export trends can provide decision-making basis for strengthening foreign trade risk management, thereby better preventing trade risks. This article constructs a combined model (NW-LSTM) based on non parametric method (Nadaraya Watson) and long short-term memory neural network model, and applies it to predict China's export volume. Based on the monthly export data of China from 2007 to 2023, this article compares the prediction results of the combination model with autoregressive moving average (ARMA), cubic exponential smoothing model, and other machine learning models. The results show that the combination model NW-LSTM can better fit historical data of foreign trade exports, and the model has good generalization ability. When the prediction period is 12 months, its RMSE, MAE, MAPA, SMAPE and other indicators are 0.1518, 0.1049, 0.1099 and 9.3704, respectively, which are better than other models, and the prediction effect is more robust; By comparing the correlation dimension and Lyapunov exponent of the exported original sequence with the fitted and predicted sequences, this paper further confirms that the model can effectively learn the dynamic characteristics of nonlinear systems with complex, non proportional time evolution behavior exhibited by the system generating the original sequence under different conditions.