Economic Forecasting Methods In Grey Support Vector Regression

With the development of social economy, the size and complexity of data are increasing, which has been a great challenge to the traditional forecasting method. The emerging statistical learning theory and grey system theory in the 1970s have provided new theoretical basis and technical supports for prediction of complex problems. On the combination of grey prediction theory with support vector machines, we put forward some proved practical methods of complex system applied in analysis and prediction of economic and financial data.By analyzing the social and economic data, the dissertation is focused on studying the performance and efficiency of support vector machine, as well as the theory and application of grey system. At first, the common grey model satisfying with verge value condition has been adopted to predict financial time series with some characteristics like poor information, high noise, non-stationary and non-linearity. Then, the model has been revised by support vector regression based on the calculation of the residual error sequence between predicted values and original data. Auto-adaptive parameters Ci have been adopted instead of C in the standard support vector regression to improve the forecasting accuracy. Meanwhile, an algorithm aboutεhas been proposed to smooth overshooting. Experimental results show that the composite model can achieve comparative accurate prediction as well as smoothing overshooting comparing to the other simple models in predictions of IGIP and TIOV.Reducing the size of training set is a direct approach to improve the learning efficiency. Due to much time consumption of global SVR, local methods come into existence as the situation requires. We introduce local grey SVR (LG-SVR) combined grey relational grade regarded as neighborhood function with local support vector regression. To optimize the machine, based on leave-one-out errors, pattern search method is adopted for model selection. Experiments are carried out on three real financial time series forecasting with LG-SVR and the results demonstrate that our approach can not only speed up the computing speed, but also improve the prediction accuracy.There are some systems with many correlation factors to predict difficultly in our life. To reduce the large error of prediction in the multivariable grey model (MGM(l,n)), based on the method of support vector machine combined with grey system, we put forward multi-variable grey composite support vector regression model (MGM-MSVR) to improve the multivariable time series prediction accuracy. Two experimental results in multi-factor economic status and stock sequence show that the compound model has an ideal effect. The method is suitable for analysis and prediction with many variables, which are influenced and restricted each other.Online prediction is the inevitable requirement of the social development. In terms of the batch learning research, in this dissertation, we put forward the adaptive on-line model with grey support vector regression. Making reference to the kernel research achievements of current online learning theory, we study the online adaptive grey prediction together with on-line kernel learning. A series of experimental results in economic data demonstrate that grey support vector regression model based on adaptive on-line can achieve high precision at the cost of more time. The learning time can be quickly reduced with choosing proper length in grey modeling data. Although explicit and implicit update in online learning and their SMD adjustment have the advantage in shortening the learning time, the generalization performance has no obvious improvement.Accurate and timely forecast methods can provide technical support for macro economic planning, and their construction and related technologies can also be applied to other research and application. The achievements in this dissertation will better the methods of data analysis and prediction, and have directive significance for many realistic problems.