Identifying herding effect in Chinese stock market by high-frequency data

Abstract

Herding behavior is thought to often occur during market frenzy, stock crashes, financial crises, as well as strong bull markets. The issue has been gaining increasing attention in recent years, in the hope that timely detection of herding behavior can be used to implement effective means to mitigate them, thus to make the market more rational. So far, herding behavior has been mainly studied using low-frequency data with methods such as LSV, PCM, CH, CKK, and HS. Such studies can only report whether herding behavior exists in a long time span, such as a few months to even a few years, and thus essentially renders all those studies irrelevant to the design of any policies for curbing herding behavior. To achieve the latter goal, it is important to realize that herding behavior is a dynamic process that may only last for a short time span, such as a few minutes. This dictates that to timely detect the herding behavior in a stock market, high frequency data must be used. Guided by this rationale, we show that computation of mutual information and cross correlation coefficient from high frequency data can indeed effectively identify herding behavior from Chinese stock markets.