The Research Of SVM Model And Parameter Optimization For Inancial Time Series Forecasting

Along with the vigorous development of the market economy, the financial marketinvestment analysis already became an important part in the people’s daily life. There are alot of financial time series forecasting technical, which including ARMA and ARCHmethod applied in linear regression models and statistic regression models etc. However,the financial time series forecasting data is characterized with noise, randomness andnon-linear etc., the traditional forecasting model is not enough to obtain satisfactory effect.Although the neural network possesses excellent non-linear character, the research in theseseveral years indicate that sometimes it will occur overfitting or generalization ability etc.and affect the accuracy of forecasting and velocity of convergence due to the effect of thecomplicated network and uncertain nodes and so on.Support Vector Machine based on structural risk minimization rule, showing itsexcellent performance in solving the problems of small sample, nonlinear and highdimension, overcoming the problems of overfitting and dimension disasters etc., it hasbecome the focus in machine learning research. In this paper, it will be deeply andsystematically investigated.Firstly, this paper summarized the concept, basic idea and reconfiguration method ofphase space reconstruction theory, proved the theory basis and calculation process formutual information method and Cao method, and applied to calculate the importantparameter optimal delay time and the minimum embedding dimension of phase spacereconstruction theory.Secondly, this paper will discuss the calculation theory, structure and model ofsupport vector machine; introduce the kernel function, which is commonly used in theprocess of application model. Through analyzing the effect of support vector machinemodel parameter selection on model performance, this paper introduces the theory ofgenetic algorithm to optimize the model parameters, and emphatically expounds the basicfactors which affect the genetic algorithm.Finally, establish a kind of genetic algorithm to optimize the financial time seriesforecasting model of SVM based on phase space reconstruction. Carry out simulationexamination for BP neural network, which is based on phase space reconstruction, SVR,GASVR separately, and evaluate the experimental result.The experimental results show that the prediction model based on GA-SVR betterreflects the inherent law of financial time series, it is a kind of scientific and effective financial time series forecasting model, whose forecasting accuracy is higher than BPneural network and SVR method.