Research On Forecasting Of Stock Market Based On Neural Network

Stock market is a very complex nonlinear dynamic system. Neural network has the capability of approximating any nonlinear system and speciality of self-learning and self-adpting. The experiments prove that the method of modeling stockmarket using neural network has a satisfying result in near-period stock prediction.The thesis analyses the theory of stock market prediction based on BP neural network and the prediction model of stock market has been established using three-layer feedforward neural network. The problems including the structure of network, the number of hidden nodes, the choose and pretreatment of swatch datum and the determination of preliminary parameters have been discussed. In order to avoid local extremum and promote convergence speed, Levenberg-Marquardt BP algorithm has been adopted. Shanghai stock index have been applied to train the established network model, then stock datum of future 1 week have been predicted using the trained network and good effect has been gained.The thesis presents genetic-BP algorithm for the high-nonlinear speciality of stock market and the shortcoming of basic BP algorithm that includes the slow convergence speed and local extremum. The genetic algorithm pays attention to unknown area search . It has high speed and relative low precision. It will not get into local extremum. The BP algorithm searches the area that include whole-space minimum. It can improve speed and recision. Theoretical analysis and experiment result show that the method of stock prediction using neural network is feasible and efficient. It has favorable foreground. The thesis also shows that the genetic-BP algorithm can not improve the speed and credibility.Through lots of experiments of stock prediction, this paper investigates the influence on the results of stock prediction that is exerted by the change of parameter. It also presents some advices on how to improve the performance of neural network.