Researches On Industrial Correlation Based On Granger Causality Networks

As an effective tool of the research on complex networks,correlation analysis benefits for the understanding of the behavior of network unities.Using methods of both partial Granger causality and complex networks,one can better understand the direct relationship among unities.This article utilizes Granger networks,partial Granger networks,and information flows to reveal the causality among industries based on Shenwan primary industry indexes.This article mainly focuses on the following points:(1)The preliminary analysis of industrial indexes relationship.Construct hierarchical structure trees based on industrial indexes,and then get the degrees of industrial indexes based on a thought of classification,and finally get the reliability of each link based on a bootstrap technique.Results show that in the observation period,nodes in hierarchical structure trees are hardly influenced by time dimensions,revealing that there are no large structure changes in industrial correlation.(2)Causality of industrial indexes.Construct a Granger causality network based on a Granger causality test of each two Shenwan primary industry indexes.The results show that there are some industries having causal relationship to the majority of the other industries,such as the steel industry,and serve as core industries to influence other industries.(3)The direct causality of industries.In the basis of the Granger causality network,a direct causality network is constructed based on a partial Granger causality test to investigate the direct causal relationship of the studied industries.And then the direct causality is distinguished from the indirect so the industries relationship is studied from different aspects.Results show that the main relationship between industrial indexes is indirect.(4)The information flow of industrial Granger and Granger causality networks.First,calculate the in-information flow and the out-information flow of each industrial index and get the industries which have great in-information,out-information flow and great asymmetry to study the characteristics of energy transfers of each industry.Based on a C hu-Liu-Edmonds MST algorithm,the minimal spanning trees of Granger and partial Granger causality networks are constructed with the node which haves the greatest information flow being the root node to identify the chain structures and ring structures of the networks.Results show that there is a global industrial correlation structure in the Granger causality network but not in the partial Granger causality network.However,the cyclic structure exists in some parts of the partial Granger network.