The Forecast Of Exchange Rate Time Series Based On The Combined Model Of The Phase Space Reconstruction And Kalman

Exchange rate becomes more and more important for the macroscopical economy policy, business work and individual decision-making, which makes the forecast of exchange rate become a hot issue for reaserchers both in china and other nations. However, the exchang rate system is very complex. After the breakdown of the Breton-Woods System, Exchange rate's fluctuation becomes more and more frequent and unstable. The forecast of exchange rate also becomes more and more difficult. And the traditional theories of exchange rate determination including"purchasing power parity","the balance of international payments of exchange rate determination"and"the method of asset maket analysis"based on linear model didn't explain the rule of exchange rate's change well. However, with the developmemt of nolinear system, exchange rate's reaserches make a great progress. Neural Network, Fractal, the theory of Chaos and the nolinear Combined Model have been used in exchange rate's forecast reaserch. Neural Network is a popular tool in nolinear system reaserch, which possesses a good approach ability of nolinear system. But Neural Network method has some disadvantages, for example, its structure is complex and the structure is built on its study of input and output data which lack of accurate mathematics expression; in the other words, Neural Network solves problem in the way of approaching in a full view. These disadvantages make Neural Network'constringency speed very slow and make Neural Network'exact rate very low. In addition, its slow speed of constringency leads Neural Network cannot be appled in the supper short-term exchange rate's online tracking and forecast.Therefore, this paper provided a new model established on the combined method of the reconstructed phase space and Kalman, and used this model to forecast the shortterm exchange rate and the supper short-term exchange rate. It aims to build a new model to fit the supper short-term exchange rate forecast and aims to have a better performance than the Neural Network model both the accrute rate and speed. The data set been selected in this paper includes three group: the daily closing price of euro to dollar from 2003-01-01 to 2003-05-16, the weekly average closing price of euro to dollar from 2003-01-01 to 2004-01-01 and the instant exchange rate selling price of dollar to Hongkong Dollar from 2006-05-26 13:55 to 2006-05-6 23:58. Then as a contrast, realize the GA Neural Network model and the BP Neural Network model and apply them to forecast the three groups of data above. Afterward, this paper compares the model based on the reconstructed phase space and Kalman with BP Neural Network and GA Neural Network. The result showed that the model based on the phase space reconstruction and Kalman is better than GA Neural Network and BP Neural Network both the forecast accrute rate and speed, and more fit to trade the exchange rate online than the Neural Network model. In the end, this paper applys .NET, C#.NET and MATLAB to realize the echange rate information service and forecast system, which provides users with exchange rate services and forecast.And it is a good practical utility.