The RMB Exchange Rate Prediction Resrach

2005, China began to implement a basket of currencies based on a floating exchange rate regime. China’s exchange rate became more market-oriented. At the same time, the volatility of the exchange rate has increased the risk to China than in the past. Therefore, it was necessary to study the prediction of RMB exchange rate and to find a scientific forecasting method and model to analyze and forecast the fluctuation of exchange rate.Exchange rate forecasting methods mainly included exchange rate decision theory and technical analysis model. The exchange rate decision theory was based on the basic factors that affect the exchange rate. In the long term, it can get a certain effect, the short-term effect was not good. The technical analysis model was a prediction model based on the exchange rate time series. The existing technical model included linear model and nonlinear model. For linear model, the model did not describe the characteristics of exchange rate data because of the complex nonlinear structure. The nonlinear model, especially the non parametric model, can describe the nonlinear structure of the exchange rate, but the model was not the time variant. We know the value of different periods in the past affect the strength of the future was not the same, the extent of the estimated amount of the impact of every historical observation period should gradually decay. The existing technical models did not take into account the limitations of this point.In this case, we used a new exchange rate forecasting model:TF model of Harvey and Oryshchenko (2012) based on time varying kernel density estimation theory.The TF model was essentially a non parametric time-varying model. Firstly, this paper briefly summarized the domestic and foreign exchange rate forecasting, and introduces the main exchange rate forecasting model of technical analysis method. Then, the descriptive statistical analysis and related basic tests of exchange rate data were conducted. Next, rolling forecasts strategies was used and TF model, ARIMA model, BP model, model and STAR model was used to carry out empirical prediction of exchange rate respectively. In the end, the forecasting ability of the model was evaluated by the traditional method of loss function and RC test.Through the research and analysis, we can draw the following conclusions:(i) The exchange rate data contains complex dynamics of the nonlinear structure, which was not subject to normal distribution.(ii) The test results based on the loss function method and RC test show that the prediction ability of TF model based on the time varying kernel density estimation theory was better than ARIMA model, BP neural network model and STAR model. As a new forecasting model, the TF model can achieve particularly good prediction effect (iii) Unexpected situation had a big impact on the ability of the model prediction, in this case, the predictive power of the four models were all decreased, but the prediction ability and reaction speed of the TF model was superior to other models in the face of the unexpected situation.