A Study On The Prediction Of Stock Price Based Deep Belief Networks

In the development of modern society, stock market has gradually become a national or the local economy "barometer". And at the same time, it is often thought to be the one of the most important investment areas for financial investors all around the world. Moreover, many companies take it as one of the most important financial channels for equity financing. So it is much important for them to know something about patterns of the price change of some stocks.As we all know, however, stock system is which is highlynonlinear and dynamic, or in other words, chaotic. Changes in stock prices can be affected by many factors.The indicators in the macro level includes the related economic policies of local or national government and it's like. The indicators in the micro level include the actual performance of the companies and so on. Therefore, for a long time scholars have studied stock forecasting techniques. At the same time, many researchers have been paying many attentions to applying neural network algorithms, e.g., back propagation neural networks and basic radical neural network, to this area. Based on this, we consider the deep learning structure.In this paper, we will further study the application of deep neural network in the field ofstock market transaction based on the previous studies, and then also fully introduce the basic structure of the deep belief network and it`s substructure, namely, restricted Boltzmann machine. Then, we propose a continuous Restricted Boltzmann Machine, which can be applied to continuous numbers. We choose one listed Corporation from Shanghai stock market, and make price prediction by this method and other two models(BPNN&RBF) as benchmarks,based on these above analysis. Finally, we further analyze the intrinsic reasons for its performance, andidentify the future research direction.