| With the increasing trend of economic globalization and trade integration,economic and trade activities among countries are expanding constantly.As the link of global financial exchanges,exchange rate plays a vital role in the economic exchanges between countries.In recent years,in order to raise the marketization level of RMB exchange rate,we have implemented a series of reforms of the exchange rate system based on market supply and demand,and the flexibility of RMB exchange rate has increased obviously.In2020,as the COVID-19 epidemic spread around the world,the economic environment of various countries was hit many times,so the exchange rate of the RMB fluctuated greatly.To master the fluctuation rules of exchange rate,the government can extract relevant economic and financial information in time and formulate or optimize its own financial and monetary policies,which is conducive to the healthy and sustainable development of the national economy.For large multinational corporations and individual investors,timely and effective foreign exchange risk management measures can be adopted to reduce losses caused by exchange rate fluctuations.Therefore,a reasonable and accurate analysis of the trend of RMB exchange rate is related to various economic decisions of the government,enterprises and individuals to a certain extent.In the study of RMB exchange rate forecasting,the prediction effect of a single time series model and artificial intelligence model often cannot fully explore the characteristics and laws of RMB exchange rate series,and the decomposition and integration algorithm is a more effective forecasting method.In this paper,the ensemble empirical mode decomposition algorithm(EEMD)is used to decompose the signal of the central price of the Sino-US exchange rate,and multiple connotative modal components(IMF)with different fluctuation frequencies are obtained.In order to simplify the subsequent calculation process,the average level and complexity of each component are measured according to the average value and fuzzy entropy of the IMF components,and then these IMF components are reorganized into high-frequency series,low-frequency sequences and trend sequences,and the high-frequency series and low-frequency sequences obtained by the recombination are modeled and predicted by differentially integrated moving average autoregressive model(ARIMA),long-short-term memory neural network algorithm(LSTM)and support vector regression algorithm(SVR).Since the restructured trend series is a quadratic function,the trend series is not predicted in this article.The predicted values of high-frequency series and low-frequency series are summed according to the corresponding time correspondence method,and the actual value of the trend series is added to obtain the individual prediction values based on the EEMD decomposition algorithm.Next,a measure of similarity between two vectors,Jaccard distance,is introduced as the optimization criterion for constructing a combinatorial prediction model,and an optimal combinatorial prediction model with weighted coefficients based on Jaccard distance is constructed,and the properties of the optimal combinatorial prediction model are studied and proved from the theoretical level.On this basis,considering that the prediction effects of each single prediction method at different points in time have different advantages and disadvantages,a generalized induced ordered weighted average(GIOWA)operator is introduced,and an optimal combination prediction model of variable weight coefficient based on the GIOWA operator and Jaccard distance is constructed,and multiple prediction error indicators based on the model under three representative parameters and the Jaccard distance between the predicted value and the actual value of the model combination are calculated,and compared with the corresponding indicators of each single prediction method.It can be judged that the decomposition ensemble predictive model constructed in this paper can effectively improve the prediction accuracy on the basis of a single prediction model.Considering that the different values of the parameters in the GIOWA operator may lead to changes in the combined prediction results,it is also necessary to analyze the sensitivity of the parameters in the GIOWA operator,and finally select a combinatorial prediction model with higher accuracy to predict the mid-price of the USD/RMB exchange rate in January 2023.From the forecast results of this paper,the prediction effect of the weighted coefficient combination forecasting model and the variable weight coefficient combination forecasting model are better than the single forecasting method,which can also provide a reference for the prediction of other time series data. |
