Futures Prices Forecasting Based On Neural Network Models

The futures market can be affected by such factors like politics, economy etc, and with a complicated internal running rules, the traditional technological forecast doesn't show a satisfying result. While neural network has non-linear characteristics, in particular to handle nonlinear system, so the neural network is an effective way of solving nonlinear problems about futures market forecasting. This paper mainly completed the following work: the first, data processing. Nonlinear characteristics of the market produce the datas with noise characteristics, in this paper by adjusting the rate of inflation, the elimination of periodic, random items to generate data series having regulatory sequences, thereby ensuring the accuracy of the forecasting. The second, model establishing. After analyzing the structure and algorithm of BP network and RBF network, then giving the method of neural networks to establish the prediction model. The third, doing numerical tests. The results show that BP network forecast better results in RBF neural network. The fourth, doing corresponding improvement to make the relative error between the predictive value and the actual value smaller against BP network algorithm requiring a longer training time, proning to inadequate local minimum. The fifth, designing a common platform of futures forecast for all users in order the general users can understand and use. Because only the professionals can use the MATLAB programming. In the design process MATCOM is used to achieve mixed programming between MATLAB and VC++, and the software of the neural network platform can run on the machines without installing MATLAB environment.