Unveiling anti-herding behavior in the vietnamese stock market: a Quantile regression and Markov–Switching approach

This study investigates herding behavior in the Vietnamese stock market (HOSE) from 2020 to 2026, extending the empirical literature on the Vietnamese stock market beyond traditional linear models. By employing Quantile Regression and Markov-Switching Dynamic Regression, this paper captures non-linear dynamics and latent market regimes during extreme conditions. Contrary to conventional findings, our results find no statistical evidence of herding behavior across all market quantiles. Instead, we uncover strong evidence of ‘anti-herding’ behavior, characterized by extreme cross-sectional return dispersion. This phenomenon is highly asymmetric: during severe downturns, the dispersion pattern is consistent with a ‘flight to quality’ toward safer assets, while upturns appear associated with localized, sector-specific momentum. Furthermore, anti-herding is most consistently significant in highly liquid stocks across all quantiles, while low-liquidity equities exhibit comparable or larger dispersion point estimates but lose statistical precision at extreme quantiles, reflecting the heterogeneity of investor beliefs and trading motives in a retail-dominated market. The Markov–Switching model endogenously identifies two distinct regimes, demonstrating that anti-herding predominantly governs the stable state and persistently maintains extreme dispersion even during crises. This suggests that anti-herding acts as a persistent structural characteristic of the Vietnamese stock market during the 2020–2026 period. These findings offer fresh empirical evidence on frontier market microstructures and suggest that the identified non-linear breakpoints could serve as useful input features for predictive Machine Learning models.