| Exchange rate is not only an important bridge to maintain economic ties between countries, but also a critical variable which can effectively and timely reflects the economic stability and financial security. In recent years, due to the rapid growth of trade relations between China and EU, the elevated international status of the euro and China's abandon of single USD-pegging exchange rate regime since the reform in 2005, the RMB/EUR exchange rate has drawn more and more attention from around the world. In particular, since its behavior has shown more complexity after the global financial tsunami, it become even more important to make realistic description and accurate prediction of the RMB/EUR exchange rate for China's foreign exchange mangagement.This paper chooses RMB/EUR series as the object, in consideration of the multi-scale complexity within its behavior as well as the limitations of ANN model applied in the current exchange rate prediction researches. Based on the tests of normality, nonlinearity and chaos existed in the RMB/EUR exchange rate series, this paper proposes a hybrid forecasting model by combining the multidimensional spatial UKW clustering algorithm with the dynamic RNN technique, which is supposed to optimize both the input selection and the learning mechanism of ANN model. From three aspects as the in-sample fit, out-of-sample forecasting and the significance test, in the empirical studies the prediction accuracy of four ANN models with various parameters are compared. The results show that UKW clustering based RNN2 model is the best one to describe and forecast the behavior of RMB/EUR exchange rate.With the combination of UKW clustering and RNN model, this paper has offered a better explanation and prediction of the RMB/EUR exchange rate behavior after the reform, which has an important significance in the exchange rate relevant decision making for both economic policies and risk aversion. |
PDF Full Text Request