Investor Sentiment and Market Volatility Across Quantiles: Evidence from Vietnam

Abstract

This study examines the role of investor sentiment in asset pricing within a frontier market, focusing on Vietnam. Using a comprehensive dataset covering the period 2015–2025 with 4018 observations, sentiment indices are constructed from both market-based and survey-based indicators. The study employs a quantile causality approach and a Quantile Vector Autoregression (QVAR) model to capture nonlinear, asymmetric, and state-dependent relationships among investor sentiment, stock returns, and market volatility. The empirical results provide several important findings. First, investor sentiment significantly influences stock returns, with stronger effects observed at extreme quantiles corresponding to bearish and bullish market conditions. Second, the impact is heterogeneous across firm sizes, with small-cap stocks exhibiting greater sensitivity to sentiment fluctuations. Third, the impact of investor sentiment on volatility is proxy-dependent and state-dependent. The market-based sentiment measure is generally associated with lower volatility at middle and upper quantiles, whereas the survey-based sentiment proxy shows stronger effects at lower quantiles, particularly during distress periods. Finally, robust bidirectional causality is identified between sentiment and market variables, suggesting the presence of feedback mechanisms between investor behavior and market performance. These findings highlight the importance of behavioral factors in shaping market dynamics in frontier markets characterized by high retail participation and limits to arbitrage. The study contributes to the literature by providing new quantile-based evidence on the nonlinear and asymmetric effects of investor sentiment in Vietnam.

1. Introduction

Understanding the role of investor sentiment in asset pricing remains a key challenge in financial economics, particularly in markets characterized by informational frictions and behavioral biases. Traditional finance theory, most notably the Efficient Market Hypothesis developed by Fama (1965), posits that stock prices fully reflect available information and are not systematically influenced by investor psychology. Within this framework, any mispricing induced by irrational investors is expected to be quickly corrected by rational arbitrageurs. However, this view has been increasingly challenged by the behavioral finance literature, which emphasizes the existence of limits to arbitrage and the role of noise traders. De Long et al. (1990a) argue that arbitrage is inherently risky, as rational investors may suffer losses when trading against sentiment-driven investors. Similarly, Shleifer and Summers (1990) suggests that investor demand for risky assets is influenced not only by fundamentals but also by sentiment, implying that mispricing can persist over time. Extending this line of research, recent studies, such as Pham and Pham (2025), highlight that noise trader demand in emerging and frontier markets is not only persistent but may also exhibit systematic patterns driven by sentiment, herding behavior, and limited financial literacy. In such settings, the interaction between noise traders and arbitrageurs can generate feedback mechanisms that amplify market volatility and sustain deviations from intrinsic values. This suggests that investor sentiment may play a particularly important role in shaping asset prices in markets with high retail participation. Empirical evidence on the sentiment–return relationship remains mixed. Brown and Cliff (2004) find that sentiment is strongly correlated with contemporaneous returns but exhibits limited short-term predictive power. In contrast, M. Baker and Wurgler (2006) document significant cross-sectional effects of sentiment, particularly for small, young, and hard-to-value stocks. These findings suggest that sentiment effects are more pronounced in segments of the market where arbitrage is more constrained. Despite the growing body of literature on investor sentiment and asset pricing, several important gaps remain. Most existing studies focus on developed and emerging markets, while evidence from frontier markets such as Vietnam remains limited. This is a critical omission, as frontier markets are typically characterized by lower informational efficiency, higher information asymmetry, and a dominant presence of retail investors. These features are likely to strengthen limits to arbitrage and amplify the impact of investor sentiment on both stock returns and market volatility. Although recent studies increasingly employ nonlinear and quantile-based approaches, the dynamic interactions among investor sentiment, stock returns, and volatility have rarely been examined within a unified empirical framework. In addition, the role of state-dependent and asymmetric effects across firm size segments remains underexplored in frontier market settings. Traditional linear models may fail to capture these dynamics, as they typically focus on average effects and ignore distributional asymmetries in financial data. From a theoretical perspective, the literature suggests that the impact of investor sentiment is inherently nonlinear and state-dependent. M. P. Chen et al. (2013) document that sentiment effects vary across market conditions and become more pronounced during periods of extreme market stress. This evidence is consistent with Prospect Theory (Kahneman & Tversky, 1979), which posits that investor behavior differs systematically between gain and loss domains, implying asymmetric responses to market outcomes. Motivated by these considerations, this study employs a quantile-based framework, combining quantile causality tests with a Quantile Vector Autoregression (QVAR) model, to examine the relationship between investor sentiment, stock returns, and market volatility in Vietnam. This approach enables a comprehensive analysis of sentiment effects across the entire distribution of returns, particularly under extreme market conditions where behavioral biases are more pronounced. Vietnam provides an appropriate empirical setting for this analysis due to its characteristics as a frontier market, including a high proportion of individual investors, limited market transparency, and relatively high volatility. In addition, variation across firm size segments (large-cap, mid-cap, and small-cap) allows for the examination of cross-sectional heterogeneity in sentiment effects. This study contributes to the literature in three important ways. First, it provides new empirical evidence from a frontier market setting, namely Vietnam, where high retail investor participation, stronger informational frictions, and more pronounced limits to arbitrage may amplify sentiment-driven price dynamics relative to developed and emerging markets. Second, unlike prior studies that typically focus on either return predictability or volatility effects separately, this study employs a unified quantile-based framework combining quantile causality and QVAR analysis to jointly capture nonlinear, asymmetric, and state-dependent interactions among sentiment, returns, and volatility. Third, the study documents clear firm-size heterogeneity, showing that small-cap stocks are significantly more sensitive to sentiment shocks, particularly under extreme bearish and bullish market conditions, thereby providing new evidence on cross-sectional sentiment transmission in frontier markets.

2. Literature Review

The relationship between investor sentiment, stock returns, and market volatility has been extensively examined in financial economics. The theoretical foundation is grounded in the noise trader framework of Black (1986), which argues that trading decisions may be driven by non-fundamental signals, leading to deviations of prices from intrinsic values. Such deviations may persist in the presence of limits to arbitrage, as formalized by De Long et al. (1990b) and Barberis et al. (1998), and further developed by Barberis et al. (1998) and Daniel et al. (1998). Within this framework, investor sentiment is interpreted as a systematic but noisy factor influencing asset prices. Empirically, sentiment has been measured using a wide range of proxies, including survey-based, market-based, and volatility-based indicators. A key contribution is the composite sentiment index of M. Baker and Wurgler (2006), which integrates multiple proxies into a unified measure and has become a benchmark in the literature. More recent advances have shifted toward alternative data sources and machine learning techniques. Studies using social media, search behavior, and textual analysis (Bollen et al., 2011; Da et al., 2015; Huang et al., 2020; Sethi & Gupta, 2025) have improved the timeliness and granularity of sentiment measurement. These developments suggest that sentiment is increasingly captured through high-frequency and unstructured data, although measurement heterogeneity remains an ongoing challenge. A substantial body of empirical research documents that sentiment contains information relevant for stock returns, although its effects are context-dependent. Early evidence suggests both contemporaneous and predictive relationships between sentiment and returns (Neal & Wheatley, 1998; Lee et al., 2002; Brown & Cliff, 2004). Cross-sectional evidence further shows that sentiment effects are stronger for stocks that are difficult to value or subject to higher arbitrage constraints (M. Baker & Wurgler, 2006; Smales, 2017; Z. Chen et al., 2025). This heterogeneity is particularly pronounced in retail-driven and speculative segments of the market. At the aggregate level, the impact of sentiment depends on market structure and institutional characteristics. Evidence indicates that sentiment effects are stronger in less efficient markets with weaker informational environments (Schmeling, 2009). Consistent findings from emerging and frontier markets suggest that limits to arbitrage and behavioral biases amplify sentiment-driven mispricing (Zhang et al., 2021; Shahzad et al., 2021; Negi et al., 2025), making these markets particularly suitable for analyzing sentiment effects. In addition to return predictability, sentiment has been widely shown to affect volatility dynamics. Theoretical contributions emphasize the role of non-fundamental shocks and noise trader risk in generating excess volatility (Shiller et al., 1984; De Long et al., 1990a). Empirical studies further document that sentiment is associated with asymmetric volatility responses and volatility clustering (Lee et al., 2002; Verma & Verma, 2007). Recent evidence shows that sentiment shocks can significantly amplify volatility during periods of market stress, including the COVID-19 crisis (S. R. Baker et al., 2020), and generate cross-market spillovers (Klemola, 2020; Del Nero & Giudici, 2025). These findings highlight the role of sentiment in shaping volatility dynamics, particularly during turbulent periods. Another strand of literature emphasizes the bidirectional relationship between sentiment and market performance. Empirical studies show that sentiment both responds to and predicts market returns, suggesting dynamic feedback effects rather than a unidirectional causality (Solt & Statman, 1989; Brown & Cliff, 2005; Fisher & Statman, 2000). This evidence has been reinforced by more recent studies documenting stronger feedback effects in high-frequency and algorithm-driven markets (Yang et al., 2022; Todd et al., 2024). Finally, recent research highlights that the sentiment–market relationship is nonlinear and state-dependent. Consistent with Prospect Theory (Kahneman & Tversky, 1979), investor behavior differs across gain and loss domains, implying asymmetric responses to market conditions. Empirical studies using quantile-based approaches show that sentiment effects are significantly stronger in extreme market states (M. P. Chen et al., 2013; Ni et al., 2015; Li et al., 2017; Shahzad et al., 2021; Dai, 2025). These findings suggest that linear models may fail to fully capture distributional heterogeneity in sentiment effects. Overall, the literature indicates that investor sentiment plays a meaningful but highly context-dependent role in asset pricing and volatility dynamics. However, empirical evidence remains mixed regarding its magnitude, persistence, and transmission channels, reflecting differences in market structure, measurement approaches, and econometric frameworks. These inconsistencies highlight the need for a unified empirical framework capable of capturing nonlinear, state-dependent, and distributional effects. Based on the theoretical and empirical literature, this study proposes the following hypotheses:

H1. 

Investor sentiment has predictive power for stock returns, particularly under extreme market conditions.

H2. 

Investor sentiment significantly influences market volatility in a state-dependent manner.

H3. 

The relationship between investor sentiment, returns, and volatility is asymmetric across different quantiles and firm size segments.

H4. 

There exist bidirectional predictive relationships between investor sentiment and market variables.

3. Methodology

This study examines the asymmetric impact of investor sentiment on stock market returns and volatility using a quantile causality framework. The quantile causality framework builds on the nonparametric approach proposed by Jeong et al. (2012), which allows for testing predictive relationships across different points of the conditional distribution without imposing strict parametric assumptions. Unlike conventional approaches, this framework allows the causal relationship between variables to vary across quantiles of the conditional distribution. This feature is particularly useful for capturing market dynamics under different regimes, such as bearish, normal, and bullish conditions. Moreover, this approach is well-suited for identifying nonlinear relationships that are often overlooked in mean-based models.

3.1. Quantile Causality Test

The traditional Granger causality framework evaluates whether past values of one variable help predict the conditional mean of another variable. A standard linear specification can be expressed as:

$$y_t=\alpha +{\displaystyle \sum _{i=1}^p}{\beta }_i{y}_{t-i}+{\displaystyle \sum _{i=1}^p}{\gamma }_i{x}_{t-i}+{\epsilon }_t$$

where $x_t$ is said to Granger-cause $y_t$ if the null hypothesis $H_0:{\gamma }_1={\gamma }_2=\cdots ={\gamma }_p=0$ is rejected. However, this framework only captures average effects and may fail to detect heterogeneous relationships across different market conditions.

To overcome this limitation, the quantile causality approach extends the Granger causality concept to different quantiles of the conditional distribution. Specifically, instead of focusing on the conditional mean, the method evaluates whether past values of one variable influence specific quantiles of another variable.

3.2. Quantile Regression Specification

Quantile causality is tested using the quantile regression approach developed by Koenker and Bassett (1978).

$$Q_y(\tau \mid {\mathcal{F}}_{t-1})=\alpha (\tau )+{\sum }_{i=1}^p{\beta }_i(\tau )y_{t-i}+{\sum }_{i=1}^p{\gamma }_i(\tau )x_{t-i}$$

where $Q_y(\tau \mid {\mathcal{F}}_{t-1})$ denotes the conditional quantile of $y_t$ at quantile $\tau \in (\mathrm{0,1})$, and ${\mathcal{F}}_{t-1}$ represents the information set available at time $t-1$. The coefficients ${\beta }_i(\tau )$ and ${\gamma }_i(\tau )$ are allowed to vary across quantiles, thereby capturing potential nonlinear and asymmetric effects.

The null hypothesis of no causality at quantile $\tau$ is defined as:

$$H_0:{\gamma }_1\left(\tau \right)={\gamma }_2\left(\tau \right)=\cdots ={\gamma }_p\left(\tau \right)=0$$

Rejection of this hypothesis implies that $x_t$ Granger-causes $y_t$ at the specific quantile $\tau$, indicating that the predictive relationship depends on the state of the market.

3.3. Quantile VAR (QVAR) Model

To account for the joint dynamics between investor sentiment and stock market returns, this study employs a Quantile Vector Autoregression (QVAR) model, in which both variables are treated as endogenous:

$$\left(\begin{array}{c}{Q}_{r_t}(\tau )\\ Q_{s_t}(\tau )\end{array}\right)=A(\tau )+{\sum }_{i=1}^p{B}_i(\tau )\left(\begin{array}{c}{r}_{t-i}\\ s_{t-i}\end{array}\right)$$

where $r_t$ denotes stock returns and $s_t$ represents investor sentiment. The coefficient matrices $B_i(\tau )$ vary across quantiles, allowing the dependence structure between the variables to change under different market conditions. The analysis is conducted across multiple quantiles (e.g., $\tau =0.1,0.5,0.9$), corresponding to bearish, normal, and bullish market regimes.

3.4. Statistical Inference

The joint significance is evaluated using the sup-Wald test proposed by Koenker and Machado (1999). The null hypothesis of no causality is tested using the sup-Wald statistic:

$$\sup \text{-}\mathrm{Wald}=\underset{\tau \in T}{\mathrm{s}\mathrm{u}\mathrm{p}}W(\tau )$$

where $W(\tau )$ is the Wald statistic computed at each quantile and $T$ denotes the set of quantiles considered. Rejection of the null hypothesis at specific quantiles indicates the presence of state-dependent causal relationships. To ensure robustness, standard errors are estimated using bootstrap procedures, which are appropriate under heteroskedasticity and non-normal error distributions.

Market volatility is estimated using a conditional volatility model based on a GARCH-type specification. The model is estimated over the full sample period, and the resulting conditional variance is used as the measure of volatility. This approach captures the time-varying volatility dynamics commonly observed in financial markets.

It is important to note that the term “causality” in this study is used in the Granger predictive sense rather than implying true structural causation. The quantile causality tests assess whether lagged investor sentiment contains predictive information about future returns and volatility across different market states. In this framework, the analysis is nonparametric and captures state-dependent relationships across the entire conditional distribution.

4. Data

This study employs a comprehensive dataset covering the Vietnamese stock market over the period 2015–2025, which captures multiple market cycles, including periods of expansion, correction, and heightened volatility. This time span is particularly suitable for examining asymmetric and nonlinear dynamics in the relationship between investor sentiment, stock returns, and market volatility. Investor sentiment is proxied using both market-based sentiment measure and survey-based sentiment proxy to ensure robustness. The primary proxy is a composite market sentiment index constructed from multiple market indicators reflecting trading behavior and overall market conditions. Specifically, the index incorporates variables such as market momentum, trading activity, market breadth, and price strength, and in robustness checks, volatility-related measures. These components are equally weighted to form an aggregate sentiment index, capturing the overall “mood” of the market. The sentiment index is constructed as an equally weighted composite of standardized indicators. Each component is first normalized using z-score transformation to ensure comparability across variables. The choice of equal weighting follows prior literature and avoids imposing subjective importance on individual indicators. To mitigate potential mechanical overlap with dependent variables, alternative sentiment measures excluding return-based and volatility-related components are also considered in the robustness analysis. To address the potential concern of mechanical overlap between sentiment measures and market outcomes, the construction of the market-based sentiment index is carefully designed to minimize direct inclusion of return and volatility information. Specifically, each component of the index, including market momentum, trading activity, market breadth, and price strength, is standardized prior to aggregation. In addition, diagnostic checks are conducted to ensure that the sentiment index is not mechanically driven by stock returns or volatility. The correlation between the sentiment index and contemporaneous returns and volatility is found to be moderate, suggesting that the index captures broader market sentiment rather than directly reflecting price movements. Furthermore, alternative sentiment indices excluding volatility-related and return-based components are constructed for robustness checks. To complement this measure, a survey-based sentiment index is also employed. This proxy reflects investors’ expectations and perceptions regarding economic and market conditions, thereby providing an alternative perspective that is not directly derived from observed market data. The combination of these two sentiment measures helps mitigate potential measurement bias and enhances the robustness of the empirical analysis. The study focuses on the Vietnamese stock market, represented by indices listed on the Ho Chi Minh Stock Exchange. Market performance is primarily captured by the VN-Index, while cross-sectional effects are examined using size-based indices, including large-cap, mid-cap, and small-cap portfolios. This classification enables the analysis of heterogeneity in the impact of investor sentiment across firms of different sizes. The lag length in the empirical models is selected based on standard information criteria, including the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC). The selected lag structure balances model fit and parsimony, ensuring robust and reliable estimation results. Daily closing prices and trading-related data are collected from multiple reliable sources, including the Ho Chi Minh Stock Exchange, FiinPro, and Investing.com. The use of multiple data providers ensures data accuracy and consistency through cross-verification. The sample period spans from January 2015 to December 2025. The final dataset comprises 4018 daily trading observations, where each observation corresponds to an actual market trading session after excluding weekends, public holidays, and other non-trading days, rather than calendar days. The survey-based investor sentiment series is obtained from a proprietary Vietnam-based investor sentiment survey database, which provides periodic measures of market expectations and investor mood. As this sentiment series is available at a lower frequency than the daily stock market data, a last-observation-carried-forward step-wise matching procedure is employed to align the two datasets. Specifically, each released survey value is carried forward and assumed to remain effective until the next survey release becomes available. This approach is consistent with the assumption that prevailing investor sentiment persists between consecutive survey updates and continues to influence trading behavior until new sentiment information is released. This procedure avoids look-ahead bias and ensures that only information available at each point in time is used in the empirical analysis. This relatively large sample size enhances the reliability of quantile regression estimates and improves the statistical power of the quantile causality tests. All models are estimated using daily data, and statistical inference is conducted using bootstrap-based standard errors with 1000 replications to account for potential heteroskedasticity and non-normality in the error terms. Stock returns are calculated as continuously compounded returns:

$$r_t=\mathrm{l}\mathrm{n}(P_t)-\mathrm{l}\mathrm{n}(P_{t-1})$$

where $P_t$ denotes the closing price at time $t$. This transformation ensures time-additive properties and is standard in financial econometrics.

To model market risk, conditional volatility is estimated using the GARCH(1,1) framework:

$${\sigma }_t^2=\omega +\alpha {\epsilon }_{t-1}^2+\beta {\sigma }_{t-1}^2$$

where ${\sigma }_t^2$ represents the conditional variance of returns. The model is estimated over the full sample period, and the resulting conditional variance is used as the measure of volatility. This measure is interpreted as an ex-post proxy for market risk rather than a real-time forecast, and therefore does not introduce look-ahead bias into the empirical analysis. The estimated conditional variance is subsequently used as a proxy for market volatility.

The dataset—combining multiple sentiment proxies, cross-sectional indices, and advanced volatility modeling—provides a robust empirical foundation to examine the asymmetric and nonlinear relationship between investor sentiment, stock returns, and volatility in the Vietnamese stock market.

5. Empirical Results

5.1. Descriptive Statistics and Preliminary Tests

Table 1. Descriptive Statistics and Unit Root Tests.

Variable	Mean	Min	Max	Std. Dev	Skewness	Kurtosis	JB (p-Value)	ADF	PP
Sentiment Index	52.84	18.23	89.67	14.92	0.21	2.35	0.000	−6.842 ***	−6.731 ***
VN-Index Return	0.00052	−0.041	0.038	0.0112	−0.18	6.12	0.000	−12.384 ***	−12.102 ***
Large Cap Return	0.00049	−0.039	0.036	0.0105	−0.12	5.98	0.000	−11.982 ***	−11.745 ***
Mid Cap Return	0.00071	−0.052	0.045	0.0138	0.34	7.21	0.000	−10.447 ***	−10.322 ***
Small Cap Return	0.00079	−0.061	0.058	0.0165	0.51	8.94	0.000	−9.884 ***	−9.765 ***

Source: Author’s calculation. *** p < 0.01.

presents the descriptive statistics of investor sentiment proxies and stock market returns. The results show that mid-cap and small-cap stocks generate higher average returns compared to the aggregate market index (VN-Index) and large-cap stocks. However, these higher returns are accompanied by greater risk, as indicated by higher standard deviations, supporting the conventional risk–return trade-off. In addition, all return series exhibit excess kurtosis and deviate from normal distribution, which is a stylized fact in financial time series. The stationarity of the variables is verified using both Augmented Dickey–Fuller (ADF) and Phillips–Perron (PP) tests. The results confirm that all series are stationary, thereby ensuring the validity of subsequent econometric analyses.

5.2. Quantile Causality from Investor Sentiment to Stock Returns

The results of quantile causality from investor sentiment to stock returns are reported in

Table 2. Quantile Causality from Sentiment to Stock Returns.

Quantile	VN-Index	Large Cap	Mid Cap	Small Cap
0.05–0.20	14.82 ** (+)	18.95 *** (+)	32.47 *** (+)	48.12 *** (+)
0.20–0.40	5.27 (ns)	6.14 (ns)	21.33 *** (+)	19.84 ** (−)
0.40–0.60	8.11 ** (+)	6.92 (ns)	5.88 (ns)	7.45 ** (−)
0.60–0.80	22.63 *** (+)	26.74 *** (+)	11.92 ** (+)	6.88 (ns)
0.80–0.95	35.91 *** (−)	34.77 *** (−)	24.63 *** (−)	33.58 *** (−)

Note: *** p < 0.01, ** p < 0.05, ns = not significant. (+) and (−) denote positive and negative causal relationships, respectively. Source: Author’s calculation.

, where different quantile intervals represent distinct market regimes ranging from extreme bearish to extreme bullish conditions. The results indicate that investor sentiment has predictive power for stock returns, particularly at the extreme lower and upper quantiles. This suggests that sentiment plays a more critical role during periods of market stress and exuberance. Furthermore, the impact is heterogeneous across firm sizes. Specifically, the effect is more pronounced at higher quantiles for large-cap stocks and the aggregate market index, while for small-cap stocks, the effect is stronger at lower quantiles. This supports the argument that smaller stocks are more susceptible to sentiment-driven mispricing. The direction of the relationship also varies across quantiles, as illustrated in Figure 1.

At lower quantiles (bearish markets), sentiment has a positive effect on returns, implying that declining sentiment leads to lower returns due to fear-induced selling. At moderate and higher quantiles, sentiment positively affects returns, consistent with optimism-driven demand as documented by S. R. Baker et al. (2020). However, at extreme upper quantiles, the relationship becomes negative, indicating that excessively high sentiment leads to lower subsequent returns. This reversal reflects over-optimism and potential overvaluation, where informed investors may exit the market, triggering price corrections. This interpretation is in line with Abudy et al. (2022), who demonstrate that exogenous mood variations driven by non-fundamental events can significantly influence market outcomes, highlighting the behavioral nature of sentiment-driven price movements. This further supports the interpretation that the observed relationships should be viewed as predictive rather than causal in a structural sense. This finding aligns with prior evidence on return reversals following periods of elevated sentiment. Results based on survey-based sentiment proxy are broadly consistent, although some discrepancies emerge at lower quantiles. These differences may be attributed to variations in data frequency and the relatively slower adjustment of survey-based sentiment proxy compared to market-based measures.

5.3. Reverse Causality: Stock Returns to Investor Sentiment

Table 3. Quantile Causality from Stock Returns to Sentiment.

Quantile	VN-Index	Large Cap	Mid Cap	Small Cap
0.05–0.20	82.41 *** (+)	71.33 *** (+)	55.12 *** (+)	58.27 *** (+)
0.20–0.40	69.25 *** (+)	63.11 *** (+)	52.77 *** (+)	49.18 *** (+)
0.40–0.60	64.88 *** (+)	67.45 *** (+)	48.92 *** (+)	44.61 *** (+)
0.60–0.80	51.73 *** (+)	70.22 *** (+)	53.66 *** (+)	42.08 *** (+)
0.80–0.95	58.96 *** (+)	104.37 *** (+)	97.84 *** (+)	62.45 *** (+)

Note: *** p < 0.01. (+) indicates positive causal relationships between stock returns and investor sentiment. Source: Author’s calculation.

reports the quantile causality from stock returns to investor sentiment. The results reveal strong and consistent causality across all quantiles, with a uniformly positive relationship. This indicates that higher past returns lead to increased investor sentiment, while lower returns reduce sentiment. The estimated coefficients, depicted in Figure 2, further confirm this positive feedback effect.

These findings support the presence of a bidirectional relationship, where sentiment is both a determinant and an outcome of market performance.

5.4. Additional Evidence from the QVAR Model

The Quantile Vector Autoregression (QVAR) model provides complementary evidence to the quantile-causality results by explicitly capturing the dynamic interactions among investor sentiment, stock returns, and market volatility across different market states. Consistent with the causality findings reported in Table 2, Table 3,

Table 4. Quantile Causality from Sentiment to Volatility.

Quantile	VN-Index	Large Cap	Mid Cap	Small Cap
0.05–0.20	0.81 (ns)	0.64 (ns)	1.12 (ns)	1.55 (ns)
0.20–0.40	2.44 (ns)	2.18 (ns)	1.96 (ns)	3.02 (ns)
0.40–0.60	38.74 *** (−)	40.11 *** (−)	27.53 *** (−)	26.89 *** (−)
0.60–0.80	41.92 *** (−)	47.55 *** (−)	33.21 *** (−)	49.37 *** (−)
0.80–0.95	48.63 *** (−)	54.28 *** (−)	59.74 *** (−)	102.55 *** (−)

Note: *** indicates statistical significance at the 1% level; ns denotes non-significance. (−) represents negative causal relationships between investor sentiment and volatility. Source: Author’s calculation.

and

Table 5. Quantile Causality from Volatility to Sentiment.

Quantile	VN-Index	Large Cap	Mid Cap	Small Cap
0.05–0.20	19.72 *** (+)	33.81 *** (+)	29.45 *** (+)	27.33 *** (+)
0.20–0.40	11.04 ** (+)	10.82 ** (+)	7.12 ** (+)	6.84 (ns)
0.40–0.60	5.92 (ns)	6.11 (ns)	5.03 (ns)	4.66 (ns)
0.60–0.80	8.77 ** (+)	9.33 ** (+)	78.12 *** (+)	35.66 *** (+)
0.80–0.95	4.88 (ns)	5.12 (ns)	120.45 *** (+)	138.72 *** (−)

Note: ***, ** indicate statistical significance at the 1% and 5% levels, respectively; ns denotes non-significance. (+) and (−) represent positive and negative causal relationships, respectively. Source: Author’s calculation.

, the QVAR estimates reveal substantial asymmetry across quantiles. In lower-tail market conditions (e.g., the 0.10 and 0.25 quantiles), negative sentiment shocks generate larger adverse effects on returns and stronger increases in volatility. These effects are economically more pronounced and persist for multiple lags, indicating that pessimistic sentiment has a stronger propagation mechanism during bearish states. By contrast, in median market states, the response of returns and volatility to sentiment shocks is more moderate, suggesting that fundamental information dominates market pricing during normal conditions. At upper quantiles, sentiment shocks continue to affect returns, but the persistence of the effect is weaker relative to lower-tail regimes. This asymmetric pattern supports the view that behavioral biases and noise-trader risk become particularly important during market stress episodes. The QVAR evidence reinforces the quantile-causality results by showing that the sentiment–market relationship is not only predictive but also dynamically state-dependent across different parts of the conditional distribution.

The Quantile VAR framework provides additional insights beyond the quantile causality tests by capturing the dynamic interactions between investor sentiment, returns, and volatility across different market states. The results indicate that sentiment shocks have asymmetric effects, with stronger and more persistent impacts during extreme market conditions.

In bearish states (lower quantiles), negative sentiment shocks amplify volatility and reduce returns more significantly, while in bullish states (upper quantiles), the effects are weaker and less persistent. This highlights the importance of state-dependent dynamics in understanding the role of investor sentiment.

Such dynamics are consistent with the behavior of trend-chasing investors and align with prior empirical evidence on feedback trading.

5.5. Investor Sentiment and Market Volatility

The relationship between investor sentiment and market volatility is presented in Table 4, with the corresponding quantile test statistic across market states illustrated in Figure 3. Importantly, the volatility effects differ across sentiment proxies, and therefore the interpretation should be proxy-specific rather than generalized across all sentiment measures.

The results show that the effect of investor sentiment on volatility is both proxy-dependent and state-dependent. For the market-based sentiment index, the estimated test statistics are statistically significant primarily at the middle and upper quantiles (0.40–0.95), and the corresponding effects are predominantly negative in sign. This indicates that stronger market optimism, as reflected in trading activity, market breadth, and price strength, is associated with lower subsequent volatility during normal and bullish market conditions. In contrast, the survey-based sentiment proxy exhibits a more nuanced pattern. Its effect is more economically pronounced at the lower quantiles, particularly during distress periods, where the relationship is generally positive, suggesting that pessimistic investor expectations tend to amplify volatility under adverse market conditions. Therefore, the stabilizing effect of sentiment applies mainly to the market-based measure during middle and upper quantiles, whereas the volatility-amplifying effect applies more clearly to the survey-based proxy in lower-tail market states. This distinction suggests that the two proxies capture different dimensions of sentiment: the market-based index reflects contemporaneous trading optimism, while the survey-based proxy better captures downside expectations and perceived risk.

Notably, in relatively stable market conditions (i.e., low-volatility regimes), the effect of investor sentiment becomes statistically insignificant. This finding highlights the state-dependent nature of sentiment, implying that its influence is more pronounced during periods of market stress or turbulence.

These findings can be explained by behavioral mechanisms in financial markets. During extreme market conditions, investors are more likely to exhibit sentiment-driven behavior, leading to stronger deviations from fundamental values. In contrast, during normal market conditions, the impact of sentiment tends to be weaker as market participants rely more on fundamental information. This pattern is consistent with behavioral finance theories, which emphasize the role of bounded rationality and noise trading in driving asset price dynamics. The results are broadly consistent with previous studies in emerging and frontier markets, which document stronger sentiment effects under extreme conditions and highlight the importance of nonlinear dynamics (Koh, 2025).

5.6. Reverse Causality: Volatility to Investor Sentiment

Table 5 reports the reverse causality from market volatility to investor sentiment, with the corresponding quantile test statistics graphically illustrated in Figure 4.

The findings reveal a statistically significant causal relationship, particularly at the extreme quantiles. Market volatility exerts a pronounced influence on investor sentiment, especially during extreme market conditions characterized by fear and greed. The relationship is predominantly positive, indicating that heightened volatility intensifies investors’ psychological responses. This result provides strong support for the feedback mechanism, whereby market instability contributes to the formation and amplification of investor sentiment. For small-cap stocks, the effect exhibits a more pronounced nonlinear pattern: it is positive at lower quantiles but turns negative at upper quantiles. This suggests that investor reactions differ across market states, reflecting heterogeneous behavioral responses under varying levels of market stress. Similarly, market volatility significantly affects survey-based sentiment primarily at extreme quantiles, with a positive sign. This finding implies that investors become more sensitive to risk during periods of heightened uncertainty or exuberance, reinforcing the role of sentiment as a state-dependent construct.

The empirical findings provide strong support for the noise trader theory proposed by Black (1986), in which investor sentiment acts as a non-fundamental signal influencing both stock returns and market volatility. More importantly, the results are consistent with a growing body of recent empirical studies documenting the nonlinear and state-dependent nature of the sentiment–market relationship. For instance, studies such as Biswas & Sharma (2025) and Balcilar et al. (2017) show that investor sentiment exerts heterogeneous effects on returns and volatility across different quantiles using quantile causality approaches. Similarly, more recent evidence (e.g., Shahzad et al., 2021; Naeem et al., 2021) highlights that sentiment-driven dynamics are particularly pronounced during periods of market stress and exhibit strong asymmetry across market conditions. These findings collectively suggest that the relationship between sentiment, returns, and volatility is inherently nonlinear, asymmetric, and state-dependent. The impact of investor sentiment is concentrated primarily in extreme market conditions, reinforcing the importance of employing advanced econometric techniques such as quantile causality to fully capture market dynamics. This is in line with recent methodological advances that extend traditional causality frameworks to the entire conditional distribution rather than focusing solely on mean effects. The results are particularly relevant in the context of Vietnam, where limits to arbitrage and the dominance of retail investors amplify the influence of behavioral factors on asset pricing. In such an environment, sentiment-driven trading plays a crucial role in shaping market outcomes, especially during periods of heightened uncertainty.

6. Robustness Checks

To ensure that the empirical findings are not driven by the specific construction of the sentiment index or model specification, this study conducts a series of robustness checks. These tests are particularly important given the potential concern of mechanical overlap between sentiment measures and market outcomes. First, alternative sentiment indices are constructed by modifying the composition of the baseline index, in which volatility-related components and return-based indicators are excluded to mitigate potential mechanical bias. Second, the models are re-estimated using alternative lag structures, and lagged sentiment variables are employed to reduce contemporaneous feedback effects, thereby ensuring that the results reflect predictive relationships rather than mechanical contemporaneous correlations. The results of these robustness checks are reported in

Table 6. Robustness Checks Across Alternative Sentiment Specifications.

Specification	Return Q10	Return Q50	Return Q90	Vol Q10	Vol Q50	Vol Q90
Baseline	0.00001874	0.00001791	−0.00002086	−0.00003572	0.00001184	−0.00005361
No Vol	0.00001869	0.00001788	−0.00002079	−0.00003565	0.00001179	−0.00005348
No Return	0.00001742	0.00001796	−0.00000438	−0.00003987	0.00001267	−0.00004492
Lagged	0.00001851	0.00001642	−0.00002094	−0.00003974	0.00001123	−0.00004987

Source: Author’s calculation. Note: Reported values are estimated QVAR coefficients at the 0.10, 0.50, and 0.90 quantiles. Statistical significance levels are based on bootstrap standard errors with 1000 replications. The “No Vol” specification excludes volatility-related components from the sentiment index, while “No Return” excludes return-based components. Coefficients are reported to eight decimal places to highlight small numerical differences across alternative sentiment specifications. The minimal deviation between the baseline and the “No Vol” specification confirms that the volatility-related component contributes only marginally to the standardized equally weighted sentiment index. The key findings remain qualitatively unchanged across all alternative specifications. In particular, the asymmetric and state-dependent effects of investor sentiment on stock returns and market volatility persist. These results provide strong evidence that the main findings are not driven by the specific construction of the sentiment index or modeling choices, thereby reinforcing the reliability of the empirical conclusions.

.

As an additional robustness test, the full sample is further divided into two sub-periods to examine whether the main findings remain stable across different market regimes. Specifically, the sample is split into a pre-COVID-19 period (2015–2019) and a post-COVID-19 period (2020–2025), reflecting potentially different market environments in terms of volatility, investor participation, and behavioral dynamics. The results are presented in

Table 7. Robustness Check: Sub-sample Analysis across Market Regimes.

Sub-Sample	Return Q10	Return Q50	Return Q90	Volatility Q10	Volatility Q50	Volatility Q90
2015–2019 (Pre-COVID-19)	0.000016 **	0.000014 *	−0.000018 **	−0.000031 **	0.000010	−0.000047 ***
2020–2025 (Post-COVID-19)	0.000022 ***	0.000019 **	−0.000025 ***	−0.000042 ***	0.000013 *	−0.000059 ***
Difference in magnitude	Stronger	Moderate	Stronger	Stronger	Stable	Stronger

Source: Author’s calculation. Notes: The sample is divided into two sub-periods to assess the stability of the baseline findings across different market conditions. The pre-COVID-19 period covers January 2015–December 2019, while the post-COVID-19 period covers January 2020–December 2025. The reported coefficients are representative estimates from the quantile regression framework. *, **, and *** denote statistical significance at the 10%, 5%, and 1% levels, respectively.

.

Table 7 indicates that the main findings remain qualitatively consistent across both sub-periods. More specifically, investor sentiment continues to exhibit significant predictive power for stock returns at lower and upper quantiles, confirming the persistence of nonlinear and state-dependent effects across market regimes. Similarly, the impact of sentiment on market volatility remains statistically significant, particularly at upper volatility quantiles. Importantly, the magnitude of the sentiment effect is stronger in the post-COVID-19 period. This result is economically intuitive, as the post-COVID-19 period in Vietnam was characterized by heightened uncertainty, increased speculative trading, and a substantial rise in retail investor participation. These market conditions are likely to amplify sentiment-driven trading behavior and strengthen deviations from fundamentals. Overall, the consistency of the results across alternative sentiment constructions, lag specifications, and sub-sample periods provides additional support for the robustness and stability of the main empirical findings. Additional sub-sample analysis further confirms that the main findings remain stable across different market regimes, with stronger sentiment effects observed in the post-COVID-19 period.

7. Conclusions

This study investigates the asymmetric and cross-sectional relationship between investor sentiment, stock returns, and volatility in the context of a frontier market, namely Vietnam. Using both market-based sentiment measure and survey-based sentiment proxy, the analysis employs a quantile causality framework to capture nonlinear dynamics across different market regimes. The empirical results yield several key findings. First, investor sentiment significantly affects stock returns, particularly at extreme quantiles, confirming the presence of nonlinear and asymmetric effects. Second, the impact of sentiment on volatility is heterogeneous across sentiment proxies and market states. The market-based sentiment measure is associated with lower volatility at middle and upper quantiles, while the survey-based measure exhibits stronger amplification effects in lower-tail market conditions. Third, the results indicate strong bidirectional causality between sentiment and market variables, highlighting a feedback mechanism between investor behavior and market outcomes. Overall, these findings underscore the important role of investor sentiment in shaping market dynamics in frontier markets, where limits to arbitrage and a high proportion of retail investors amplify behavioral effects. The results have implications for understanding the mechanisms through which sentiment influences both return formation and volatility dynamics in less efficient markets. From a theoretical perspective, the results are consistent with behavioral finance theories, particularly those emphasizing state-dependent and asymmetric investor behavior. From an empirical perspective, the findings highlight the importance of accounting for nonlinear dependence structures when modeling financial markets. Despite these contributions, several limitations should be acknowledged. First, the measurement of investor sentiment may be subject to proxy-related biases inherent in both market-based and survey-based indicators. Second, the empirical analysis focuses solely on Vietnam as a representative frontier market, which may limit external validity. Future research could extend this framework by conducting cross-country comparisons, incorporating higher-frequency sentiment indicators, or examining sector-level heterogeneity.