Application Of SVM Based On Information Granulation In Time Series Analysis Of Securities

With the rapid development of economy, researches in the field of finance have been paid more and more attention by the scholars at home and abroad. In recent years, there are so many methods of time series analysis. Traditional model, either the nonlinear mapping of model which is relatively simple and can not fully reflect the nonlinear law:or the structure which is too complicated. making it difficult for applying widely. One of the methods, Support Vector Machines (SVM) has provided a better model method, which had a solid theoretical foundation, making it no necessary to consider the systematic mathematical model. For data fitting. using structural risk minimization and the training rate is high. This thesis combines information granulation with SVM to make a regression analysis on securities time series.First of all, the thesis gives a relatively deep introduction to the theories of time series analysis. SVM and information granulation and applies SVM to time series analysis, then based on the model to make choices of independent variables and dependent variables of securities time series data, finally normalizes these data:Secondly, doing researches on the issues of adjusting SVM-related parameters. Three parameter optimization methods are introduced, that is. cross validation, genetic algorithm, particle swarm optimization, then the thesis compares the efficiency of these algorithms:Last, after the fuzzy information granulation of the Shanghai Composite Index and then makes a regression analysis of SVM, thus an effective forecast of variation trend and space of the Shanghai Composite closing Index is made.Experimental results show that SVM model reflects the variational regulation of the Shanghai Composite index very well, fitting and forecast are relatively ideal; while the combination of information granulation and SVM model allows SVM to play better results, furthermore the method is very practical and feasible.