A Research On Forecasting Of RMB Exchange Rate Based On Hilbert-Huang Transform And Recurrent Neural Network

The study on exchange rate has been an important and hot issue in national finance theory and policy research. Exchange rate fluctuation is the decisive factor to generate exchange rate risk. Predicting the direction and the range of exchange rate volatility are of priority to determine the size and impact. It is of great significance to gain a deep understanding of the exchange rates behavior, to reveal its innate running mechanism and ultimately improve forecasting accuracy for the countries to develop relevant ex-change rate policies and adjust foreign exchange reserves, international business deci-sion making of banks, and international trade activities of enterprises.This paper reviews relevant methods and models about exchange rate forecasting, and evaluates the applation of neural network, time-frequency analysis method, espe-cially the Hilbert-Huang transform in exchange rate forecasting field to determine the research base. An exchange rate forecasting model based on Hilbert-Huang transform and recurrent neural networks is designed. Hilbert-Huang Transform can take full use of the raw sequence. The recurrent neural network has one or more feedback layers, so it can r truly realize dynamic modeling of the nonlinear system.This paper selects weekly exchange rate of RMB against the U.S. Dollar from January 1, 1994 to June 30, 2009 as research sample. Firstly, the author tests the nonlinearity of the RMB exchange rate sequence. Then the author conducts empirical mode decomposition on it, and reconstructs the results of decomposition to prove the validity of the decomposition. And then the author conducts Hilbert transform on intrin-sic mode function obtained by on the empirical mode decomposition to find time-frequency characteristics of the RMB exchange rate sequence. On the basis of de-termining the key parameters of neural network, this paper compares the forecasting effect between recurrent neural network and the model based on Hilbert-Huang trans-form and recurrent neural network.The empirical results show that the model presented in this paper is valid.