Internet postings and investor herd behavior: evidence from China's open-end fund market

The study investigates whether and how internet postings (as a form of network information exchange) and investor herd behavior are dynamically linked in China’s open-end fund market. The context is that emerging markets like China, dominated by retail investors and characterized by information asymmetry, are more susceptible to herding. Prior literature shows mixed evidence on whether online information exchange mitigates or exacerbates herding and highlights time-varying and nonlinear features of herding. This paper addresses the research gap by: (1) testing for dynamic, time-varying dependence between postings and herding; (2) examining the internal transmission mechanism in both directions; and (3) assessing asymmetries under different conditions. The authors employ CSAD as a herding proxy and use DCC-GARCH for dynamic correlations, TVP-SV-VAR for time-varying impulse responses, and later a NARDL framework for asymmetry. Findings contribute novel evidence that the linkage is dynamic, negative (postings reduce herding), stronger in the short term, and asymmetric, with herding also feeding back to postings.

Data and sample: Monthly data for Chinese open-end funds and internet postings from Eastmoney Stock Forum spanning January 2010–June 2021. Forum posts are collected via web crawling from fund-related boards. Data are seasonally adjusted using Census X-12 and standardized. Herding measure: Cross-Sectional Absolute Deviation (CSAD) following Chang et al. (2000): CSAD_t = (1/N) Σ_i |r_it − r_mt|, where r_it is fund i’s return and r_mt is the Shanghai Composite Index return. Monthly CSAD is the average of daily CSAD within the month. Lower CSAD implies stronger herding. Internet postings variable: Let S_t be the number of forum posts; define L_t = ln(S_t). Because L_t is non-stationary due to fixed maintenance posts and a growth trend, an autoregressive model L_t = μ + γ1 L_{t−1} + γ2 L_{t−2} + γ3 L_{t−3} + E_t is estimated, and the residuals E_t represent de-trended internet information exchange. Pre-tests: ADF and PP unit root tests indicate CSAD and E are stationary at 1% (ADF: CSAD −19.812, E −40.210; PP: CSAD −21.220, E −41.004). ARCH-LM tests confirm ARCH effects for CSAD and E (χ²=51.611 and 39.005; p=0.000), supporting GARCH usage. VAR lag selection favors lag 1 (LR=77.841). DCC-GARCH (1,1): Conditional covariance H_t = D_t R_t D_t, with univariate GARCH(1,1) for each series and dynamic correlation via Q_t = (1−α−β)Q + α ε_{t−1}ε'_{t−1} + β Q_{t−1}. Estimates summarize persistence and time-varying correlations between E and CSAD. TVP-SV-VAR: A time-varying parameter VAR with stochastic volatility is estimated under a Bayesian MCMC framework (10,000 draws). Parameters β_t, α_t, and log-volatility h_t follow random walks. Identification via lower-triangular contemporaneous impact matrix A_t. Impulse responses are analyzed over short (4 months), medium (8 months), and long (12 months) horizons, and at different time points. Convergence checked via Geweke statistics (<1.96) and inefficiency factors (max 66.23). Asymmetry via NARDL: Following Shin et al. (2014), postings and CSAD are decomposed into partial sums of positive and negative changes (E_u, E_d; CSAD_u, CSAD_d) to examine asymmetric short- and long-run effects of increases vs. decreases in postings and dispersion on each other.

- Dynamic correlation (DCC-GARCH): The correlation between internet postings (E) and market dispersion (CSAD) is time-varying, ranging from −0.144 to 0.704, with mean 0.073, median 0.051, SD 0.092. Persistence parameters show high volatility clustering: CSAD α=0.252, β=0.698, α+β=0.950; E α=0.123, β=0.789, α+β=0.912. - Internet postings → herd behavior: Postings have a significant negative effect on herding (i.e., increase CSAD), strongest in the short term and weakest in the long term. The short-run impulse responses show postings shocks increase dispersion quickly (reducing herding), with effects fading by medium-to-long horizons. - Herd behavior → postings: Increases in market dispersion (lower herding) tend to reduce postings initially; decreases in dispersion (higher herding) increase postings. The impact is time-varying: increases in dispersion have more significant short-term effects on postings, while decreases in dispersion have more significant long-term effects. - Asymmetry (NARDL/TVP-SV-VAR): Positive vs. negative shocks are asymmetric. Increases in postings have a larger (and short-run strongest) effect on reducing herding than decreases in postings. Likewise, positive dispersion shocks affect postings more than negative shocks. Short-, medium-, and long-term paths differ over subperiods (e.g., regime shifts around 2013 and mid-2015). - Robustness and preliminaries: Series are stationary (ADF/PP at 1%), ARCH effects present (p=0.000), optimal VAR lag order is 1, MCMC diagnostics acceptable (Geweke < 1.96; inefficiency factors allow ≥150 effective samples).

The study answers whether online information exchange influences herd behavior and vice versa in a large emerging market. Evidence shows postings reduce herding by mitigating information asymmetry and improving the information environment, thereby increasing market dispersion (CSAD) and enhancing market efficiency. The effects are strongest in the short run, consistent with rapid investor reaction to newly disseminated information online and the timeliness/decay of online signals. Conversely, periods of high dispersion (low herding) reduce investors’ incentive to interact online, while low dispersion (high herding) spurs online engagement as investors seek confirmation and community, indicating feedback from behavior to information exchange. The relationships are both time-varying and asymmetric: increases in postings exert greater anti-herding pressure than decreases, and positive dispersion shocks influence postings more than negative shocks. These findings justify using time-varying and nonlinear models and highlight that policy or platform changes can alter the strength and horizon of effects over time.

There is a significant, time-varying, and asymmetric linkage between internet postings and investor herding in China’s open-end fund market. Internet postings weaken herding most strongly in the short term, while herding feeds back to increase postings and, via this channel, further dampens herding. Asymmetries are evident: increases in postings have a more pronounced effect on reducing herding than decreases; dispersion shocks affect postings differently across horizons. Policy suggestions include: (1) investors should recognize the dynamic nature of online information and avoid trend-following; (2) regulators should strengthen oversight and transparency of online platforms to stabilize markets; and (3) listed firms and information providers should facilitate accessible communication channels to reduce herding-induced volatility.