全球经济政策不确定性的演变趋势：基于EMD-ARIMA模型的预测分析

作为经济政策变动的风向标，全球经济政策不确定性（Global Economic Policy Uncertainty，GEPU）指数的动态走势对于经济政策的制定和调整具有重要的参考价值。然而GEPU指数动态路径的影响因素复杂多变，其数据生成过程难以在一个时间序列模型中得到准确的体现。基于“先分解后集成”的建模思路，首先采用经验模态分解（Empirical mode decomposition，EMD）方法将全球经济政策不确定性指数分解为若干相互独立、频率不同的可读信号，其次运用非平稳时间序列ARIMA模型对可读信号分别进行建模预测，最后集成各类可读信号的预测结果。在此基础上，进一步应用VAR模型考察了全球贸易、新冠疫情等因素对GEPU指数的动态影响。研究发现：(1)通过对训练组和测试组数据的预测值与真实值的对比，发现EMD-ARIMA模型对训练组和测试组数据的拟合精度均优于ARIMA模型；(2)与ARIMA模型相比，EMD-ARIMA 模型能够解决由原始数据不确定性、非线性以及不稳定性所导致预测偏差问题，得到精度较高的预测结果；(3)全球贸易、新冠肺炎疫情等因素对全球经济政策不确定性均产生了显著的影响，EMD-ARIMA模型的样本外预测结果显示，GEPU指数在2021年7月之前呈增加趋势，2021年7月至12月逐渐趋于稳定。

Evolutionary Trend of Global Economic Policy Uncertainty：A Forecasting Analysis with EMD-ARIMA Model

As the vane of economic policy change, global economic policy uncertainty index has important reference value for scientific formulation of economic policy. However, the factors affecting the dynamic path of GEPU index are complex and changeable, and the process of data generation is difficult to be accurately reflected in a single time series model. Based on the idea of “decomposing before integrating”, this paper firstly decomposes the GEPU index into some readable signals with the empirical mode decomposition method, and then simulates and predicts the readable signals by using the mainstream non-stationary time series model (ARIMA). Finally, the prediction results of all kinds of readable signals are integrated. The results show that: (1)EMD-ARIMA model is better than ARIMA in fitting the data of training group and test group by comparing the predicted value with the real value; (2)Compared with ARIMA model, EMD-ARIMA model can solve the problem of prediction deviation caused by uncertainty, nonlinearity and instability of original data, and obtain higher precision prediction results. (3)Factors such as global trade and the epidemic of NCP have a significant impact on the uncertainty of global economic policy. The out-of-sample prediction results of EMD-ARIMA model showed that GEPU index increased before July 2021 and gradually stabilized from July to December 2021.