Herding in mutual funds: A complex network approach

•We sample data of mutual funds investing in European stocks.•We build the network basing on the correlation matrix.•We calculate centrality measures.•We introduce different measures of herding.•We detect the role of the main centrality measures for herding. The paper investigates herding in mutual funds through a complex networks approach. The detection of significant correlation coefficients constitutes the basis for the construction of the network. Some centrality measures and the assortativity are added as explanatory variables in the regression analysis of two popular indicators of herding, largely applied in finance literature. Cross-Sectional Standard Deviation and Cross-Sectional Absolute Deviation are both considered since they emphasize the bulk and the extreme values of herding. Two dummy variables designed to capture differences in investor behaviour in extreme up or down versus relatively normal markets are considered as independent variables. The results show a clear decrease of herding in stressful periods of the market. Moreover, the prevailing explanatory power of the betweenness is well evidenced, thereby highlighting the role of the network structure. In line with the literature on herding, the results also evidence a flight to safety effect.