基于模糊聚类的羊群效应检测方法

Detection Method of Herding Effect Based on Fuzzy Clustering

摘要: 为检测股票市场中有无出现羊群效应，以及个股的羊群效应程度和股市不同时间段的羊群效应程度，提出一种基于模糊聚类的羊群效应分析方法。通过计算股票每天的开盘收盘价得到收益率，计算收益率一阶差分构建具有时间序列属性的收益变化特征。将每支股票作为单个样本，采用模糊C均值聚类方法得到隶属度矩阵。建立类别判断准则，计算第一类样本隶属度与第二类样本隶属度的最大/最小差值，将出现最大/最小差值的样本与收益变化特征均值的方法来判断哪一类是羊群效应。相比计算横截面标准差等传统方法再进行线性回归得到回归系数，该方法具有能够表现个股羊群效应优势；相比k-means聚类计算羊群效应，该方法用隶属度能够得到更多的聚类信息。

Abstract: In order to detect whether there is herd behavior in the stock market, the degree of herd behavior of individual stocks and the degree of herd behavior in different periods of stock market, this paper proposes an analysis method of herding effect based on fuzzy clustering. By calculating the daily opening and closing prices of the stock to get the yield, the first-order difference of the yield is calculated to construct the return change characteristics with time series attribute. Taking each stock as a single sample, the membership matrix is obtained by fuzzy c-means clustering. The maxi-mum/minimum difference between the membership degree of the first kind of sample and the membership degree of the second type sample is calculated by establishing the category judgment criteria. The method of the sample with the maximum/minimum difference and the mean value of the income change characteristics are used to determine which kind of herding behavior is. Com-pared with traditional methods such as calculating cross-section standard deviation and linear regression to get the regression coefficient, this method can show individual stock herding effect; compared with k-means clustering, this method can get more clustering information by member-ship degree.