Exchange Rate Forecasting Using A Hybrid ARMA And Sparse Bayesian Model

The exchange rate is an important national foreign economic variables，along with the improvement of the openness of China’s economic degree, moreand more important role it plays in the exchange rate. From July21,2005, theexchange rate volatility has increased since our country’s RMB exchange ratepeg to the U.S. legal system from the original exchange rate system into amanaged floating exchange rate system. Therefore, how to accurately describeand forecast the RMB exchange rate for the entire financial sector has importanttheoretical research significance and practical application value.On the basis of previous studies, this paper according to the linear andnonlinear time series data characteristics of the exchange rate, and ARMA-basedsparse Bayesian model is proposed according to ARMA model and sparseBayesian model. First, the research background and significance of the paper,the research overview abroad of the exchange rate forecasting model wereintroduced, and we presented the main contents of this paper. Then, the basicprinciples and forms of the ARMA model, as well as Sparse Bayesian modelwere introduced, followed by the ARMA-Sparse Bayesian model. The modeltook full advantage of the unique linear and nonlinear modeling character ofARMA and Sparse Bayesian models. Namely, the yield sequence wasdecomposed into the linear component and nonlinear residual component, thenwe used ARMA model to forecast the linear component, Sparse Bayesian modelto forecast the nonlinear component, lastly the relevant result was predictedfrom the synthesis of the RMB against the U.S. dollar rate sequence. Theexperimental results showed that using ARMA-Sparse Bayesian model ofexchange rate forecasting can obtain better prediction effect. Finally the modelwas applied in the closing price of SSE Composite Index forecast, and achieveda good prediction effect.