Securities Investment Funds (SIFs) play a crucial role in China's capital markets, with a significant market share as of March 31, 2020. However, the herding behavior among SIFs in asset allocation is prevalent due to weak internal governance and influences such as fund ranking and personal interests. This herding behavior is characterized by the synchronized buying or selling of specific stocks by numerous funds, often observed in quarterly reports. This phenomenon is further exacerbated by the prevalence of “crosstown effect”, insider information, and alumni connections, reflecting China's relationship-driven culture. This paper investigates the hypothesis that information similarity among fund managers, gained through their social networks, drives the convergence of their asset allocation decisions, resulting in concentrated shareholding and herding behavior. The paper addresses several key questions: How can fund managers' social networks be effectively measured? What are the specific characteristics of the influence mechanism of fund managers' social networks on the herding behavior of Chinese SIFs? To answer these, the study utilizes information asymmetry theory, behavioral finance theory, and social relations theory to explore the influence of fund managers’ social networks on SIFs herding behavior and formulate hypotheses regarding network centrality, size, and constraint. It employs the CSAD model to measure herding behavior and constructs a regression model for empirical analysis, considering the heterogeneity across dimensions such as herding behavior degree, gender, diploma, and region. The main contributions of this paper are its exploration of fund manager social networks' impact on SIF herding behavior, its clarification of the theoretical mechanisms involved, its construction of a measurement method for fund manager social networks, and its investigation of the heterogeneity of this impact.

The literature review examines research on social networks and their impact on investor trading behavior. Studies show correlations between investment behavior among fund managers with close network ties (Colla & Mele, 2010; Pool et al., 2015), influences of alumni connections (Cohen et al., 2008; Shen et al., 2016), and the role of network centrality in information advantage (Berk & Binsbergen, 2015). Further, it reviews existing research on the formation mechanisms of SIFs' herding behavior, primarily focusing on reputation (Bolton & Scharfstein, 1990), compensation structure (Maug & Naik, 2011), and information flow (Bikhchandani et al., 1992). Existing studies on the herding behavior of Chinese SIFs highlight characteristics such as pronounced herding in bulk-holding stocks (Chu & Qin, 2008), relationships between herding, fund return rate, and outstanding shares (Chen, 2004; Zhang & Li, 2005), and the role of technical analysis in driving herding effects (Wang et al., 2012). The paper also provides a comprehensive overview of models used for measuring herding behavior, particularly the CSAD model and its application in previous studies (Chang et al., 2000; Ahmed, 2017; Lee et al., 2017; Zhou et al., 2019; Cui et al., 2019; Ukpong et al., 2021; Wang et al., 2021; Cheng et al., 2022). The theoretical analysis integrates information asymmetry theory, behavioral finance theory, and social relations theory to frame the relationship between fund manager social networks and SIF herding behavior.

This study uses daily return data from July 6, 2012, to July 7, 2022, for all open-end SIFs in China. After excluding money market, bond, and some commingled funds due to their low variability, the sample includes 6078 entries of equity and mixed partial stock funds, covering over thirty industries. Market returns are proxied by daily returns of the Shanghai and Shenzhen 300 indices. Fund manager social networks are constructed using data from resumes, deconstructed into alumni and colleague networks. The CSAD model is used to measure the degree of herding behavior (HB) for each fund manager's funds, using the quadratic coefficient (γ2) from a regression of CSAD on market returns. A negative and significant γ2 indicates herding. The regression model examines the impact of fund manager social networks on HB, using 'Freeman closeness centrality' (FreeClo), 'Valente-Foreman closeness centrality' (ValClo), network constraint (Const), and network size (Size) as explanatory variables and controlling for gender (Sex), diploma (Dip), number of schools attended (Jnum), and number of funds under management (Snum). The CSAD model is detailed. The cross-sectional absolute deviation (CSADt) is calculated as the average absolute deviation of individual fund returns from the average market return. Under the Capital Asset Pricing Model (CAPM), a linear relationship exists between individual fund returns and market returns in an efficient market. However, herding behavior destroys this linearity, leading to a non-linear relationship between CSADt and market returns. Therefore, the study employs a regression model of CSADn,t on market returns (Rm,t), with the quadratic coefficient (γ2) serving as the measure of herding behavior. To enhance robustness, a modified CSAD model incorporating an exponential function correction is used following Feng (2011). The data processing involves calculating daily CSAD values for each fund managed by the sample fund managers and averaging these values to obtain a CSAD index for each manager. Explanatory variables are calculated using UCINET software based on the network data extracted from fund managers' resumes. Control variables include gender, diploma, number of schools attended, and number of funds under management. Data for 100 fund managers with significant negative γ2 values are analyzed. Descriptive statistics are presented for all variables. Heterogeneity is investigated by dividing the sample into subgroups based on different levels of herding behavior, gender, diploma, and region. Separate regression models are run for each subgroup, and the results are compared across groups.

The empirical analysis using data from 100 fund managers reveals the following: 1. **Overall Impact of Social Networks:** The regression results indicate that network size (Size) has a statistically significant negative impact on herding behavior (HB). In simpler terms, larger network sizes are associated with a lower degree of herding. However, network centrality (FreeClo and ValClo) and network constraint (Const) are not statistically significant. 2. **Heterogeneity Based on Herding Behavior Degree:** When the sample is divided into high and low herding behavior groups, network size (Size) demonstrates a significant negative effect on HB only in the high herding group. This suggests that the size of a fund manager’s network matters significantly for preventing herding in contexts where it is already a substantial problem, but not so much when herding is already low. 3. **Heterogeneity Based on Gender:** A significant negative relationship between network size and herding behavior is found only for male fund managers. Female fund managers show no significant relationship between social network characteristics (centrality, constraint, or size) and herding behavior. This difference suggests that the influences of social networks on herding behavior are potentially distinct for male and female fund managers, possibly due to differences in risk-taking behavior and social pressure. 4. **Heterogeneity Based on Diploma:** The study finds that the effect of network size on herding behavior is significantly negative only for fund managers with master's degrees, while it is not significant for those with Ph.D. degrees. The possible explanation is that Ph.D. fund managers might access and filter information better due to their more intricate and wider-ranging social networks. 5. **Heterogeneity Based on Region:** The study finds varying effects in different regions. In Beijing, the ValClo measure shows a positive and significant relationship with herding behavior, suggesting a “status advantage effect”. However, the impact of social networks on herding behavior is not consistent across regions, with no significant effects observed in Shanghai and Shenzhen.

The findings indicate that fund managers' social networks do influence their herding behavior, primarily through network size. Larger networks are associated with less herding behavior, potentially because of increased access to diverse information. However, network centrality and constraint did not show consistent significant effects, possibly due to factors not captured in the model. The heterogeneity observed highlights the complexity of the relationship. The differences between male and female fund managers might reflect gender-specific social dynamics and risk preferences. The contrasting results for master's and Ph.D. degree holders suggest that higher education levels, beyond a certain point, might not necessarily lead to greater susceptibility to herding. Regional variations point to the influence of local market conditions and information dissemination patterns. These findings add to the understanding of herding behavior in the context of social networks, particularly highlighting the importance of considering individual characteristics and market contexts.

This study contributes to the understanding of herding behavior in SIFs by highlighting the role of fund managers' social networks, especially the impact of network size. The heterogeneity analysis reveals significant differences based on gender, education, and region. The negative correlation between network size and herding emphasizes the importance of information diversity. Future research can investigate the impact of online social networks, incorporate other types of relationships, and examine intentional versus spurious herding through the lens of behavioral finance and principal-agent theory.

The study's limitations include the reliance on resume data for network construction, which might not capture the full complexity of social connections. The CSAD model might underestimate herding behavior in volatile markets. The focus on specific regions might limit the generalizability of findings. Future studies should use more comprehensive network data, investigate various herding types, and analyze a wider range of geographical areas.