Abstract: The uncertainty of the global economy has largely intensified the international trade risk. An early prediction and warning system for international trade risk is designed based on the trade competitiveness index. Considering the non-linearity, non-stationarity, strong volatility and relatively small sample size of international trade data, a hybrid prediction model with autoregressive integrated moving average model (ARIMA) and gated recurrent unit (GRU) after wavelet decomposition is proposed. Specifically, the trade time series data is decomposed into high-frequency sequence data and low-frequency sequence data through wavelet transform. According to the characteristics of data, the ARIMA-GRU hybrid model is constructed, and the prediction results of each frequency data are ensembled together to get the final prediction result of the trade competitiveness index. In addition, an early warning mechanism is proposed to facilitate the practical management of trade risks. To verify the effectiveness, applicability and practicability of the early prediction and warning system, electromechanical products are taken as an example to conduct the empirical analysis. Comparing with other commonly used single models and hybrid models such as LSTM, GRU and RNN, the results indicate that the proposed method has higher prediction accuracy, demonstrating satisfactory performance in long-term and short-term trade risk prediction.

Key words: the prediction of international trade risk; wavelet decomposition; autoregressive integrated moving average model; gated recurrent unit