Research On Forecasting Exchange Rate Based On Combination Model

With the development of the world economy,exchanges between countries are increasing day by day.The exchange rate,as a bridge to maintain economic and trade exchanges between countries,plays a vital role in the economic exchanges between countries.The exchange rate changes directly affect the world.Trade and Economic.Therefore,correctly analyzing and predicting the exchange rate is of great significance for the country to formulate economic policies and avoid risks.At present,the research on the exchange rate prediction problem generally follows two kinds of ideas,one is to use a single model for prediction,and the other is to use multiple model combination prediction.Many scholars often use traditional time series models and linear models to predict the fluctuation of future exchange rate values,but empirical analysis shows that exchange rate fluctuations not only have linear laws,but also exhibit certain nonlinear characteristics,simply using linear or nonlinear models.Predicting the exchange rate will inevitably produce a large error.Based on this,this paper selects the appropriate linear model and nonlinear model to combine,and uses the combined model to fit and predict the spot exchange rate of RMB/USD.Firstly,using a single time series ARIMA model prediction,considering the residual distribution problem in the model,the residuals that do not obey the normal distribution are inversely transformed by interpolation,and the residual pair model after fitting is found.The accuracy of the prediction error has been improved.Secondly,the background value of the GM model is improved and predicted by the combination of the gray GM model and the quadratic interpolation method and the Simpson formula.At the same time,the BP neural network model is used to predict the exchange rate.The simulation results show that the prediction results of BP neural network model are better than those of other models for the prediction of exchange rate.Thirdly,the ARIMA model,the improved ARIMA model,the improved gray GM model and the nonlinear BP neural network model are used to predict the exchange rate.The research shows that the prediction error of the combined model is much smaller than the prediction error of the single model.The combination of the inverse transform ARIMA model and the BP neural network model is superior to other combined models.Finally,through empirical analysis,the exchange rate is predicted and the method proposed in this paper is verified.The research shows that the method of fitting the model residuals by inverse transform can improve the prediction accuracy of the model.The results have certain research significance for predicting and analyzing the trend of exchange rate.