Prediction Of RMB Exchange Rate By Combination Model And Study Of Equilibrium Exchange Rate

Exchange rate research is not only a hot topic but also a difficult one.Exchange rate forecasting and equilibrium exchange rate are the core issues in the exchange rate issue,and have been always concerned by many scholars.In the research of exchange rate forecasting because the exchange rate itself is a complex and changeable dynamic system,which has both linear and non-linear mixed characteristics,a single linear model or a single non-linear model can not be perfectly competent for the job of exchange rate forecasting.This paper considers the combination of ARIMA model and artificial neural network model,that is,ARIMA model is used to describe the linear characteristics of exchange rate,and neural network is used to mine the non-linear characteristics of exchange rate.The delayed feedback ofthe dynamic neural network model makes it have the function of memory and storage,so it can describe the dynamic behavior characteristics of financial time series and predict it accurately.Therefore,Elman neural network and NARX neural network are selected as the neural networks to build the combined model.The former is the delayed feedback of the hidden layer neurons,and the latter is the delayed feedback of the output layer neurons.In order to learn the advantages of the two networks,the HIF-NARX neural network is also constructed,that is,the delayed feedback from the hidden layer to the input layer is added to the NARX neural network.This makes HIF-NARX neural network have two memory modes.Through empirical research and prediction of RMB exchange rate,it is found that ARIMA-ANN model has higher accuracy than single model In 10 days,the precision of combination model(RMSE)is about 0.03,that of single neural network model is about 0.05,and that of ARIMA model is about 0.15.Moreover,the predictive ability of the combined model is far greater than that of the single model in the longer term(20,30 days).In the study of equilibrium exchange rate,the traditional BEER cointegration equation is first established,then the state space theory and the semi-parametric variable coefficient theory are introduced to construct the time-varying coefficient,and the latter is estimated by the local linear regression method and the cubic uniform B-spline method,respectively.Then the improved BEER model is used to measure the equilibrium exchange rate and the degree of exchange rate imbalance.The conclusion shows that the results of the improved BEER model based on the semi-parametric variable coefficient theory are closer to the real economic significance;the degree of exchange rate imbalance of the real effective exchange rate of RMB has been declining from 2010 onwards,and the ability of the RMB exchange rate to return to the equilibrium exchange rate has been improving,and it is getting closer to the equilibrium exchange rate,showing a two-way fluctuation trend.