The Path Correlation:Empirical Evidence Andsimulation Test

Correlation analysis is a kind of method to research of the relation between variables.It is important for the random analysis of variables,and the results of correlation analysis can provide powerful support for excavating the information behind the data.From the application,the investment,risk control,information push and so on are closely related to correlation analysis.Scholars had studied the correlation,and proposed many methods to measure the correlation of variable s,but the correlation research focused on the analysis of the degree of correlation the related degree between variables,not on the statistical pattern between the variables.Some correlation coefficient,such as Pearson correlation coefficient,to measure the correlation degree between the variables but not statistical pattern,other correlation coefficient,such as Kendall correlation coefficient,Spearman correlation coefficient and so on,can reflect the correlation between variables partly,and also depict the correlation structure between the variables partly.The arrival of the age of the data brings challenges to the correlation research.In theory,the correlation relationship between multiple variables is very complex which is especially true for high dimensional data.With the deepening of the research,some scholars have found that the hypothesis of original research is not established,these inappropriate assumptions may lead to serious consequences.This paper,inspired by Xu Bing's path design(2010),references to some recent research.It provides a new method to analysis correlation by building a nonparametric path model system,which can analysis the correlation patterns and the degree of correlation.In this paper,we use Li and Racine' nonparametric variable selection method(2004)to classify the variables;build a nonparametric path model system based on the classification results,then analysis of the integral effect,direct effect and indirect effect between variables.We found(1)the proportion and the fluctuations of nonlinear component are greater than them of the linear component whether in the benchmark model or path model,and the nonlinear variables dominate the model system;(2)the overall effect of electricity consumption is the biggest among single path variable,while the overall effect of Electricity consumption and the use of credit line is the biggest among double path variable;(3)a simulation analysis was carried out on the specific data while extrapolation precision of benchmark model replaced the causal analysis between variables.