Study On Application Of Support Vector Regression In Combined Forecast

Bates and Granger proposed a concept of combined forecast in 1969 and this idea has been well accepted by scholars. But with the development of our society, the forecast object is affected by lot of intricate elements. Obviously, those combined forecast models which are short of flexibility and self-learning can not forecast the object in complex environments accurately. Meanwhile, the researches on the single forecast model selection oriented by combined forecast are so rare that it will affect the stability and reliability of forecast.In view of above analysis, based on the analyses and comparison in existing literature, this paper proposes principle and process of single forecast model selecting which can raise the stability and reliability of combined forecast and put forward principle and process of combined forecast model based on support vector regression. The research of this paper will promote the selecting quality of single forecast models, accuracy, stability and reliability of forecast and reduce forecast risk, which can make the ability of managerial decision-making more exact and increase the decision level of our nation's enterprises.There are five main contents of this paper.Firstly, discuss the theory and models of combined forecast. Through analysis and comparison of exiting literature, review the basic concepts, theories, methods and models of combined forecast. At the same time, conclude the research achievement of combined forecast's theory which can lay the foundation for principle and methods of this paper.Secondly, study on the basic theory and arithmetic of Support Vector Machine. At first, research basic theory and model structure of Support Vector Machine. Then, conclude the superiority of Support Vector Machine and lay the foundation for principle and methods of this paper.Thirdly, study on single forecast models selection in the background of combined forecast. Based on deep analysis of single forecast models'features and according to the environment of combined forecast, build a standard of single forecast models selection and set up an appraisal system then apply advanced TOPSIS to evaluate single forecast models and realize single forecast models selection.Fourthly, study on building combined forecast models based onε-Support Vector Regression. On the basis of single forecast models selecting, applying Statistical Learning Theory to propose the theory and process of buildingε-Support Vector Regression combined forecast model and the framework of forecast. This combined forecast model can improve the forecast accuracy and reduce forecast risk.Fifthly, apply theε-Support Vector Regression combined forecast model to empirical research. According to above research, apply the theory of single forecast models selection andε-Support Vector Regression combined forecast model to the field of the dollar/yen rate. Through the analysis and comparison, the performance of the application has been evaluated and the validity of the researched result can be confirmed.Innovation has been accomplished in the following two aspects.(1)According to the features of combined forecast models and environment of combined forecast, build a evaluation system of single forecast models selection then apply advanced TOPSIS to promote the quality of single forecast models selection.(2)Based on single forecast models selection, propose theε-Support Vector Regression combined forecast model and research further on the theory and method of this combined model. This model promotes the accuracy and stability of combined forecast.