Reseach On Forecast Techinology Of Exchange Rate Based On Neural Networks And Genetic Programming

Recently, Exchange rate determination and prediction has been very an important and controversial topic among existing international economic research references. But there have been not complete theory that can give convincing explanation. In this paper, because of the complexity of the exchange rate issue, according to the nonlinear relationship between exchange rates and exchange rate'nonlinear characteristics of data itself, calculated from the nonlinear model and evolutionary point of view, the RMB exchange rate predictability of some issues were analyzed which used neural networks and genetic programming as the main tool. According to the characteristics of RMB exchange rate itself, the forecasting model was built that was suitable for the RMB exchange rate. The main achievements are summarized as follows:1. The process of the RMB exchange rate forecasting model was designed. Through the analysis of influencing factors on the RMB exchange rate, USD, JPY, EUR, and KRW the four exchange rates were selected as input variables, and collection and pre-processing were achieved on the input data. Through the analysis of various empirical models, neural network and genetic programming model were selected. The fixed forecasting scheme was used to compare and evaluate the prediction accuracy such as2. Do some researches for BP network and algorithm in forecasting examples. Through the different structure, the number of hidden layers, learning rates, and transfer function which influence the convergence effects and test results were compared each other. After comparison and analysis, the best BP model structure was obtained and the forecasting result was optimum.3. Conventional BP algorithm was improved from the point of transfer function and networks structure. A new transfer function was used. The transfer function is bipolar after mathematical calculation and analysis. So it can be used as a transfer function in neural network. Through the forecasting results were compared with the traditional BP neural network, we can see that not only convergence speed but also forecasting accuracy of the improved BP network had been improved significantly. The superiority of the new ransfer function was reflected.4. According to the same training samples and forecasting samples, RBF and GRNN were used to forecasting exchange rate. Through the determination of the model's parameters, we can receive the best SPREAD of RBF and GRNN, the forecasting errors of the training and testing results were smallest. Then the forecasting results of the improved BP neural network were compared with RBF and GRNN network. It proved that the training speed and forecasting accuracy of the BP neural network were best.5. A RMB exchange rate forecasting model based on the genetic programming method was proposed and constructed. According the symbolic regression function of genetic programming, the results of the algorithm were analyzed and compared with other forecasting models. In this case the result of genetic programming was better than neural networks. The technology is a method innovation in forecasting of exchange rate field.