Causality Testing Based On Self-Organizing Data Mining

Causality between variables is one of the focuses in economic time series analysis. Granger causality test is one of the most important and popular methods. Science proposed in 1969, this method has been improved on various aspects by researchers in different countries. However, it has been demonstrated that this method has shortcomings on complex system studying.From the point of complex system modeling, it is proposed the causality testing method based on Self-Organizing Data Mining, though combining the basic idea of the concept of Granger causality with the complex system oriented technique, the Self-Organizing Data Mining, to test the causality in complex system.Firstly, it is proposed the GMDH (Group Method of Data Handling)-based causality test:(1) It is established the concepts of GMDH-based causality and designed the testing procedure.(2)Compared with Granger causality test though theoretical analysis, numerical experiment and the example of RMB exchanging rate, it is showed the validity of GMDH-based causality test in complex system studying.GMDH-based causality test is suitable for single-input-multi-output complex system with certain characters. Further more, it is proposed the following causality testing methods by combining the basic idea of the concept of Granger causalitywith other techniques of Self-Organizing Data Mining:(DParametric OSA based causality test and the OSA based Interpretation StructuralModeling, both suitable for multi-input-multi-output complex system with certaincharacters,(2)Nonparametric FRI based causality test and its improves edition based on ACwhich takes the effects of data splitting into account, both suitable forsingle-input-multi-output fuzzy, complex system.(3)Nonparametric OSA based FRI causality test, suitable formulti-input-multi-output fuzzy, complex system.Complexity and complex system research have important meaning on theory and extensive prospects of application. Causality testing based on Self-Organizing Data Mining extended the Granger causality test on concepts, methods and results, which exhibits its features on identifying the hierarchical structure and evolving rules of complex system.