For evil news rides fast, while good news baits later?—A network based analysis in Chinese stock market

Abstract

As stated by British poet John Milton “For evil news rides fast, while good news baits later”. Does this case exists in the stock markets? In order for investors and regulators to make a well-informed decision, it is essential to understand the propagation characteristics of good information and bad information in the stock market. However, most of the articles focus only on the relationships among markets from a micro perspective, which leads to a lack of systematic study in the whole market. Therefore, it is necessary to study the propagation characteristics of good information and bad information in the whole market. Based on the China Securities Regulatory Commission industry classification criteria, we use stock data of some listed Chinese financial firms to study the propagation characteristics of good information and bad information. Firstly, log return decomposition model is applied to extract good information series and bad information series from the daily log return. Secondly, the linear Granger causality test model is employed to construct good information propagation network and bad information propagation network. Then, the Dijkstra algorithm is used to find the shortest distance between each pair of nodes in the information propagation networks before the construction of good information Dijkstra network and bad information Dijkstra network. Finally, four indicators, including Number, Speed, Depth and Connectless, are proposed to compare and analyze the propagation characteristics of good information and bad information on the constructed Dijkstra networks. As revealed by the comparison results, among the listed Chinese financial firms, good information propagates farther than bad information. However, bad information propagates more easily, faster and more reachable than good information. Bad information propagates more easily, faster and more reachable within the same cluster than between different clusters, while good information propagates farther within the same cluster than between different clusters. From the perspective of information propagation, the performance of log return decomposition is considered to be better than realized semi-variance method.