Research And Development Of Foreign Exchange Trade System Based On Computational Intelligence

With development of the foreign exchange markets, the foreign exchange trading strategy, which is based on the computational intelligence, has become more and more important. On the other hand, foreign exchange rates forecast is the theoretical basis of the foreign exchange transaction.According to the present studies of the trading strategy, this paper introduced a neural network trading strategy, which is based on the foreign exchange rates forecast. The practical application of the strategy in the transaction shows that the transaction using the strategy could produce a better performance.With the background, purpose and significance of the subject, ARIMA model and GARCH model are introduced at the first part of the paper. Then, the building procedure of the ARIMA-GARCH model is described. At the second part, the ANN model is introduced. The VLRBP model and GRNN model are selected to establish the foreign exchange rates model. At the third part of the paper,a method of building SVM model is presented, which is based on the phase space reconstruction by using C-C means. At the forth part of the paper, a sample series of the euro exchange rate against dollar is used to build ARIMA-GARCH model, ANN model and SVM model. Through a study of real demonstration, the following conclusions are drawn:(1)The VLRBP neural networks model produces better performance than the ARIMA-GARCH model and the SVM model.(2)To improve the generalization performance of the ANN model, a bath-tube curve method is given when searching the size of the hidden neuron and the sliding window of the BP neural networks.(3)To improve the performance of the SVM model, a method, which is based on phase space reconstruction, is proposed when searching the embedded dimension and the time delay of the rates series.At last, through training the difference in historical exchange rate, the trading strate-gy is improved. Experiments show that the artificial neural networks can better impro-ve the performance of the trading strategy.