Prediction Of Financial Time Series Fluctuation Range Based On Weighted SVM And Granular Computing

The fluctuation of financial time series reflects the status of economicdevelopment and the distribution firm. It is the economic barometerattracting many investors. So the research of financial time series can helpus to understand the global economic change, and help investors to avoidrisk. But because the financial time series which is influenced by manyfactors, has characteristics of high noise, nonlinearity, traditional methodscan hardly accurately predict on the single value.This paper shows a way from a more practical perspective, whichuses a reliable range solution instead of a single value. It fuses SVM andGrC to predict the fluctuation range of financial time series and improvethe standard SVM to weighted SVM according to the characteristics offinance time series.The penalty parameter, kernel function parameter and loss functionparameter of SVM have important effects on the prediction accuracy.Study through experimental analysis of parameters is used to preliminaryscreen parameters to determine the loss function parameters. After thecomparison of several different optimization algorithms, the crossvalidation method is used to optimizing the parameters, helping to improvethe prediction accuracy.This paper also introduces the basic theory of granular computing andanalysis the method of fuzzy information granulation. The combinationof information granulation and SVM can get the approximate solution fortargets at lower costs. According to the characteristics of the financial timeseries, the weighed SVM gives recent data more weights than the long term data.The weighed SVM has a better generalization performance infinancial time series.Finally, according to the experiments on the three major stock indextime series, the algorithm based on weighed SVM and GrC can wellpredict the fluctuation range of financial time series and the precision isbetter than that of standard SVM.