Financial System Trend Prediction Research Based On Influence Calculating Model

Complex network is one structured research tool of complex system, which studies the internal energy mutual relation of whole system from a global perspective. In the stock market system, using the theory of history reproducibility hypothesis to establish the complex network with stocks which can describe the correlation between feature, and then analyzing the property of network structure and network function from the viewpoint of system, this method is one of the main directions of the stock market system research.In the dissertation, it includes two aspects of content:the one involves the research of influence relationship in communities from the aspect of application concerned, the other one analyzes the evolution characteristics of communities to study the influence relation inner and among communities, and then realizes the predicted goal of trend prediction of stock market system. The two details issues as follows:Firstly, presenting the research status of stock network regard the correlation coefficient as its weight and analyzing the deficiencies of existing community partition algorithms, thus introducing the conception of partial correlation coefficient. Furthermore, adding activity feature based on application context to define influence calculating model. Regarding the influence value from the influence calculating model as the weights between nodes to establish the stock network and putting forward an algorithm named BCNHC algorithm. The results of Testing BCNHC algorithm with individual stocks and Plate Index show that the definition of influence calculating model and the design of BCNHC algorithm is reasonable and effectively.Secondly, researching the influence among communities according to community evolution theory, and then using the influence among communities to predict the trend of Stock. The thoughts as follow:the first is construct the Bayesian network with stock market technical indicators to obtain the trend at currently time-sharing line; the second is add the trend of shorter time-sharing line into Bayesian network to construct influence calculating model, then measure the trend with influence at currently time-sharing line; the last is build the transmission model of subsystems with each time-sharing line to predict the whole trend of stock market system.Comparing with the real data to assess the algorithm and model that were designed in the paper shows that as soon as a reasonable influence calculating model is put forward, the trend that is predicted will become more accurately.