The impact of social media on stock market behavior is a growing area of research. This study focuses on the interaction between internet postings and herd behavior, particularly within the context of China's open-end fund market. This market is chosen for several reasons: emerging markets are more susceptible to herding due to information asymmetry; the Chinese stock market is heavily influenced by individual investors prone to irrational herd behavior; and the significant size of the Chinese market makes it globally relevant. The study employs the CSAD index to measure herd behavior, chosen for its ability to detect herd behavior on a large scale. The study also considers the role of online forums as important information hubs and disseminators of information, which can both influence and be influenced by herd behavior.

Existing literature presents conflicting views on the relationship between internet information exchange and stock returns. Some argue that it reduces information search costs and improves market efficiency, thus weakening herding behavior. Others contend that it exacerbates behavioral biases, leading to increased herd behavior. This study aims to address this gap by investigating the dynamic correlation and the internal transmission mechanism between investor behavior (herd behavior) and internet postings.

The study uses data on Chinese open-end funds and postings in internet financial forums from January 2010 to June 2021. Herd behavior is measured using the CSAD index (absolute deviation of cross-sectional returns), while internet postings are measured using the log of the number of postings in the Eastmoney Stock Forum. To address non-stationarity, an autoregressive model is used, and the residuals represent the internet information exchange variable. The data is seasonally adjusted using Census X-12 and standardized. Stationarity is tested using ADF and PP tests. The DCC-GARCH (1,1) model is employed to analyze the dynamic correlation between internet postings and herd behavior. This model is chosen based on the Lag Order Selection Criteria and ARCH effect tests which indicate the appropriateness of this model for this type of data. Univariate GARCH models are constructed for market dispersion and internet postings to get conditional variances and standardized residuals for DCC analysis. Subsequently, a TVP-SV-VAR model is used to investigate the interaction between internet postings and herd behavior over time and to capture any time-varying effects. The MCMC algorithm with Bayesian framework is employed for sampling and Geweke statistics are used to ensure validity. Finally, a nonlinear autoregressive distributed lag (NARDL) model is used to analyze the asymmetric interaction by dividing postings and market dispersion into high and low states, exploring heterogeneity. This allows the study to examine different states and conditions of the financial market.

The DCC-GARCH (1,1) model reveals a significant, dynamic correlation between internet postings and market dispersion (a measure of herd behavior), fluctuating between -0.1441 and 0.7042, with a positive mean and median suggesting some time-varying positive correlation. However, the TVP-SV-VAR model provides a more nuanced picture. The impulse response analysis shows that in the short term, a shock to market dispersion negatively impacts internet postings, while in the short term, a shock to postings positively impacts market dispersion (reducing herd behavior). This positive effect weakens over the medium and long term. The long-run effects of market dispersion shock on internet postings are negative. The response functions are time-varying, showing that the impact on internet postings from a market dispersion shock is significantly negative in the early periods and fluctuates around 0 in later periods. The NARDL analysis indicates asymmetric effects. Increases in postings have a consistently positive effect on reducing herd behavior, while decreases have a less consistent, yet still significant negative effect. Increases in market dispersion have a consistently positive effect on itself, indicating persistence, and a negative effect on postings. Decreases in market dispersion have a positive effect on postings. Overall, the findings demonstrate that the relationship between internet postings and herd behavior is dynamically correlated, time-varying, and asymmetric.

The findings of this study contribute to the literature by demonstrating the dynamic, time-varying, and asymmetric relationship between internet postings and herd behavior in China's open-end fund market. The negative relationship between internet postings and herd behavior in the short-term suggests that information disseminated online can counter the effects of herd behavior, particularly in immediate reactions. The asymmetry in the effects of increasing versus decreasing postings highlights the importance of considering the direction of information flow in understanding the impact on herd behavior. The time-varying nature of the relationship emphasizes the importance of understanding the dynamic context and how relationships change in different states of the market. This study provides novel empirical evidence for the intricate interplay between online communication and investor behavior, offering a more complex understanding than previously considered.

This paper finds a significant dynamic correlation between internet postings and herd behavior in China's open-end fund market. Internet postings negatively affect herd behavior, more so in the short term. Herd behavior influences postings, further moderating its influence. The relationship is time-varying and asymmetric. The study recommends that market participants should understand the dynamic relationship between online information and herd behavior. Regulators should focus on overseeing online platforms to maintain a transparent information environment. Firms can create communication channels to reduce price fluctuations caused by herd behavior. Future research could explore other emerging markets or investigate the role of specific types of online postings.

This study focuses solely on the Chinese open-end fund market, limiting its generalizability to other markets. The choice of Eastmoney Stock Forum may also introduce a bias, although it is one of the most influential financial platforms in China. While the models used are appropriate, potential model misspecification remains. Future studies could benefit from a broader range of data sources and alternative econometric models.