Behavioural drivers of sustainable debt markets: direct and indirect investor sentiment in global green, social and sustainability bonds

GSS bonds have emerged as one of the most prominent financial innovations of the past decade (Maltais and Nykvist, 2021), designed specifically to finance environmentally and socially beneficial projects (Lee et al., 2023). Despite rapid growth, with cumulative issuance exceeding USD 5.6 billion by the end of 2024 (Climate Bonds Initiative, 2024), sustainable debt still accounts for only around 4% of the global debt market (Alan and Lee, 2025). This reflects an imbalance between supply and demand. The sustainable debt market remains largely supply-driven, with issuance led by institutions rather than by investor demand (Barua and Chiesa, 2019). Even investors sympathetic to sustainability objectives often perceive ESG instruments as ineffective, reinforcing the demand–supply mismatch (Hashimoto and Huettinger, 2024).

Psychological and perceptual perspectives on the environment have traditionally been prominent research topics in environmental sociology and psychology (Thi Tuyet Mai, 2019). As a key indicator of investor psychology, investor sentiment plays a central role in shaping sustainable investment decisions (Piñeiro-Chousa et al., 2021; Wu and Liu, 2023). This study argues that investor sentiment is critical for addressing the demand-side constraint in sustainable debt markets. Hence, we examine the impact of investor sentiment on the performance of global GSS bond markets, including green, social and sustainable bond segments. The breakdown of GSS bonds is presented in Supplementary Table A.1.

Based on global GSS bond market data from November 1, 2021 to February 28, 2026, the findings show that both direct and indirect investor sentiment positively influence the volatility of sustainable debt returns, although indirect sentiment has a stronger effect. The impact is not uniform across bond categories, with green and social bonds displaying greater sensitivity than sustainability bonds. Moreover, the QQ analysis indicates that sentiment effects are state-dependent and vary across different return and sentiment quantiles.

Our research makes four key contributions to sustainable finance and behavioural finance literature. First, it extends the literature beyond green bonds by providing evidence on the role of investor sentiment in the broader sustainable debt market represented by global GSS bonds. In doing so, it advances the understanding of sustainable debt as a comprehensive asset class that incorporates both environmental and social objectives, offering insights that cannot be obtained from a green-bond-only perspective. Second, the study demonstrates that sustainable debt is not a homogeneous asset class. Examining green, social and sustainability bonds separately reveals heterogeneous responses to investor sentiment across bond segments, highlighting differences in investor behaviour within sustainable debt markets. This finding challenges the common assumption that sustainable debt instruments react uniformly to behavioural factors. Third, the study adds to behavioural finance research by showing that investor sentiment is multidimensional. By incorporating both direct and indirect sentiment measures, it provides evidence that different sentiment channels contain distinct information about sustainable bond market performance, thereby extending prior studies that rely on a single sentiment proxy. Finally, the study reveals that the relationship between investor sentiment and sustainable bond returns is non-linear and state-dependent. By applying a QQ approach, it uncovers behavioural dynamics that remain hidden in conventional mean-based analyses, offering a deeper understanding of how sentiment affects sustainable debt markets under varying market conditions. Hence, our study is significant for regulators and investors as it provides useful policy implications for regulating sustainable debt markets, developing comprehensive investor sentiment indicators and shaping sustainable investment strategies.

The remainder of the paper proceeds as follows. Section 2 reviews the literature, outlines the theoretical background and proposes hypotheses. Sections 3-5 present the data, methodology, results and discussion. Section 6 provides a robust test. Section 7 concludes. Section 8 provides research limitations.

With the rapid growth of sustainable finance, recent studies have incorporated sentiment analysis of green, social and sustainable (GSS) bonds (Bajra and Wagner, 2024). Piñeiro-Chousa et al. (2021, 2022) first document a positive relationship between investor sentiment derived from Twitter and green bond returns in global and Chinese markets. Subsequent research highlights the role of investor sentiment in spillover effects across green finance markets (Wu and Liu, 2023), green bond premiums (Fu et al., 2024), volatility transmission and risk spillovers (Su et al., 2024; Bouteska et al., 2024; Man et al., 2024; Le et al., 2025). Alternative sentiment measures distinguish moral sentiments such as self-love and sympathy, helping differentiate ESG and social and sustainable investment markets (Hashimoto and Huettinger, 2024). In parallel, investor attention has also been shown to predict green bond returns and volatility (Pham et al., 2020; Pham and Cepni, 2022; Gao et al., 2023; Uckun-Ozkan, 2024).

Based on the extant literature on the relationship between investor sentiment and sustainable debt markets, we identify four research gaps that our study aims to address. Firstly, most current research highlights the role of investor sentiment in green bond markets, while other categories of sustainable financial instruments, including social and sustainability bonds, remain underexplored. Additionally, we recognize that existing studies in this area mainly employ a single direct sentiment indicator, paying less attention to the impact of indirect measures. Given the complexity of observing investor sentiment (Fernandes et al., 2016; Reis and Pinho, 2020; Wu and Liu, 2023), it is essential to consider both types of indicators for a more nuanced analysis of their influence on GSS bond market performance. Furthermore, current research primarily focuses on China and the US, whose results may not be fully representative. Consequently, global coverage of GSS bonds is highly in demand to provide greater coverage of this nexus. Lastly, most studies focus on average (mean) effects, overlooking how the relationship between sentiment and returns may vary across sentiment levels and market conditions.

To examine the impact of investor sentiment on the performance of sustainable debt markets, our research is framed within the theories of behavioural finance, prospect theory and the concept of socially responsible investment (SRI).

The theory of behavioural finance captures the human side of decision-making (Pompian, 2011; Prosad et al., 2015) and considers the influence of psychology on investor behaviour (Sewell, 2007). Accordingly, investors are not perfectly rational but are influenced by sentiments, cognitive biases and heuristics (Statman, 1999). Thus, behavioural finance theory incorporates psychological and emotional factors into market participants’ financial decision-making (Pompian, 2011; Sewell, 2007).

The prospect theory, introduced by Tversky and Kahneman (1979), argues that individuals evaluate outcomes relative to a reference point and are loss-averse, leading to deviations from rational behaviour under uncertainty, which helps explain the influence of investor sentiment on asset pricing, especially in volatile markets.

To better understand investor behaviour in sustainable debt instruments, this research also incorporates the concept of SRI, which is defined as “consider the social and environmental consequences of investments, both positive and negative, within the context of rigorous financial analysis” (Anastasia, 2003). Prior studies, including Adam and Shauki (2014) and Paetzold and Timo (2014), identified investor intention as a key driver of SRI decisions, consistent with the theory of planned behaviour, which states that actions are shaped by intentions reflecting individuals’ effort and commitment. Pro-environmental consumption behaviour, which is predicted by consumers’ psychological and social well-being regarding green products (Nguyen et al., 2024), may also capture this investor intention.

Existing research already shows that investor sentiment influences green bond market performance. For example, He and Shi (2023) document that sentiment stirs environmental awareness and triggers green preference among investors, contributing to green bond return dynamics. Chen (2021) notes that overinvestment behaviour may increase default risk and generate asymmetric payoffs, and that high-sentiment periods may shift capital flows between equities and bonds. These mechanisms imply that sentiment affects bond returns through overinvestment, shifts in preferences, and capital flow channels. When considering sustainable debt markets more broadly, these channels may become even more pronounced, as sustainable investors’ decisions involve additional behavioural components, such as moral intention and a preference for environmental or social outcomes.

Because sustainable bonds possess both financial and non-financial attributes, their valuation requires investors to assess not only expected risk and return but also environmental and social benefits. These non-financial outcomes are often difficult to quantify objectively, leading to greater valuation uncertainty than in conventional bond markets. Behavioural finance theory suggests that when information is ambiguous to evaluate, investors tend to rely more heavily on heuristics, emotions and affective judgements (Statman, 1999). Consequently, investor sentiment may play an important role in sustainable debt markets, where investment decisions are influenced not only by financial considerations but also by sustainability preferences, moral values and environmental awareness. As sentiment becomes more optimistic, investors may revise their expectations regarding the prospects of sustainable assets, leading to stronger market participation and greater fluctuations in prices.

Given the empirical evidence and the theoretical justification that sustainable bonds may be more sentiment-sensitive than conventional bonds due to their dual financial and non-financial natures, we propose the following hypothesis:

Investor sentiment and investor attention are usually complex to observe (Fernandes et al., 2016; Reis and Pinho, 2020; Osabuohien-Irabor, 2021; Wu and Liu, 2023) and can be reflected by both direct and indirect proxies. Direct measures of sentiment rely on survey-based assessments of investors’ perceptions of financial markets or the broader economy (Bouteska, 2020; Anand et al., 2021). In contrast, indirect measures derive sentiment from market and economic data, based on the premise that investors reveal their expectations through their trading behaviour (Fernandes et al., 2016).

The mechanism through which indirect sentiment affects financial markets can be explained by the liquidity-as-sentiment framework of Baker and Stein (2004). They argue that, in the presence of short-sale constraints, optimistic investors exert a greater influence on market activity, causing higher investor sentiment to be reflected in increased trading and market liquidity. Consequently, sentiment-driven demand may exert upward pressure on asset price fluctuations. Wang et al. (2022) provide further evidence that liquidity movements are closely associated with investor sentiment. In the sustainable debt market, where valuation involves both financial and non-financial considerations, stronger indirect sentiment is expected to amplify fluctuations in bond prices and increase market uncertainty. Hence, we hypothesize that:

Although investor sentiment may affect the aggregate GSS bond market, its influence may vary across GSS bond categories. Green, social and sustainability bonds differ in the nature and measurability of the non-financial benefits they provide. Compared with the environmental outcomes of green bonds, the social benefits of social bonds are often more difficult to quantify and evaluate objectively. In contrast, sustainability bonds combine multiple sustainability objectives, potentially further increasing the valuation complexity. Investors rely more on sentiment when asset valuation is subject to greater uncertainty and ambiguity, as behavioural finance theory suggests. Hence, GSS bond categories with less easily measurable non-financial outcomes may exhibit stronger sentiment effects than those with more clearly defined sustainability characteristics. Therefore, we propose the following hypothesis:

The research design section introduces the data and variables used in our study as well as the research model used to test the proposed hypotheses.

Our research employs GSS bonds and investor sentiment indicators. For GSS bonds, we use the Bloomberg Global Aggregate GSS Bond Index, which is decomposed into the Bloomberg Global Aggregate Green Bond Index, Social Bond Index and Sustainability Bond Index. This decomposition allows us to examine the role of investor sentiment in sustainable debt markets at both the aggregate and individual bond category levels. Consistent with existing studies on sustainable bond performance, we use daily returns of the GSS bond indices.

For investor sentiment, we employ the World Sentix index and the market turnover ratio to proxy for direct and indirect sentiment, respectively.

Regarding the Sentix index, according to Wang et al. (2021), investors form expectations about the future based on objective factors such as political stability and macroeconomic conditions, which are closely related to anticipated economic and market performance. Survey responses from investors’ expectations are aggregated by Sentix GmbH into a qualitative diffusion index ranging from −100 (strong deterioration) to 100 (strong improvement), with zero indicating no change in expectations. Accordingly, the Sentix index directly reflects investor sentiment, with a positive (negative) value indicating that, on average, respondents are optimistic (pessimistic) about future market performance. Chen (2021) suggests that investor sentiment influences bond markets through two main channels: overinvestment and capital flows. On the one hand, during periods of high (optimistic) sentiment, investors become more confident about market performance, which can encourage overinvestment.

On the other hand, optimistic sentiment can be reinforced over time, prompting investors to engage in return-chasing behaviour consistent with feedback trading. As a result, capital may flow across asset classes, including shifts away from bonds. The Sentix index is a popular proxy for investor sentiment in prior research such as Bethke et al. (2017), Wang et al. (2021) and Panagiotou et al. (2023). Thus, it serves as a reliable proxy for direct investor sentiment in our study.

In terms of the market turnover ratio, the transmission channels through which it affects asset prices can be theoretically explained by the “liquidity-as-sentiment” approach developed by Baker and Stein (2004). Accordingly, they propose that liquidity, as reflected by market turnover, increases with investor sentiment when short-sale constraints are incorporated into their model. The reason is that when market liquidity is high, asset pricing is more likely to be driven by irrational investors who underreact to information in order flow and equity issuance. Wang et al. (2022) further validate the above “investor-sentiment-driven movements in liquidity.” Short-sale constraints may lead irrational investors to trade when they are optimistic, thereby causing their sentiment to be reflected in market activity. As a result, investor optimism (or pessimism) tends to correspond with an increase (or decrease) in the turnover ratio. This suggests that turnover is not only a measure of trading activity but also reflects sentiment-driven behaviour. In the literature, prior studies also adopt the market turnover ratio as a proxy for investor sentiment. Specifically, Baker et al. (2012) constructed investor sentiment indices for six international stock markets based on four market-based sentiment measures, including the market turnover ratio. They report that the market turnover ratio has the highest loading in the composite sentiment index. Additionally, Wang et al. (2022) utilize the market turnover ratio as a direct proxy for investor sentiment to investigate its impact on returns across 40 stock markets. Drawing on prior literature, we employ the market turnover ratio to reflect indirect investor sentiment.

The Bloomberg Global GSS Bond Index is a daily data index launched in October 2022, with index history backfilled to January 1, 2018. However, the start date for daily data availability varies across bond categories. Thus, to obtain a consistent dataset, our sample covers the daily GSS bond index from November 1, 2021 to February 28, 2026. Table 1 presents an overview of our research data and variables.

To conduct the research, we employ a multi-method approach that comprises two econometric models. The GARCH-MIDAS model is employed to examine the numerical influence of investor sentiment on the sustainable bond return volatility. To gain a more comprehensive and nuanced view of the relationship between investor sentiment and GSS bonds, the QQ technique is used to illustrate the visual impact at different levels of sentiment and returns.

In our research, there was a mismatch of data frequency between dependent and independent variables. Specifically, the data frequency for sustainable bond returns is daily, while that for sentiment indices is monthly. To address this issue, we employ the GARCH-MIDAS model developed by Engle et al. (2013), which is well-suited to handling data-frequency mismatches. Additionally, GARCH-MIDAS captures fundamental shifts and accommodates structural breaks by accounting for movements in a long-term component for short historical samples (Engle et al., 2013).

In the original GARCH- MIDAS mode, we have:

Where $regbi,t$ is the daily returns of sustainable bond indices on day i in month t, $ζi,t|Ωt−1∼N(0,1)$ with $Ωt−1$ is the available information on day i-1; $gi,t$ is the short-term component; $mt$ is the long-term component; $μi$ is the unconditional mean. The short-term factor $gi,t$ follows the GARCH (1,1) model as:

Where $α$ and $β$ are ARCH and GARCH components, respectively, with $αi$ > 0, $βi$ >0 and

$α$ + $β$ < 1. When $α$ + $β$ < 1, conditional variance is mean-reverting and converges to its long-run level. As noted by Engle and Rangel (2008), values of $α$ + $β$ close to unity implies slow mean reversion and highly persistent volatility.

Regarding the long-term volatility factor, which is our parameter of interest, $mt$ it is a varying function of the exogenous variable, which is investor sentiment in our research, as follows:

In equation (3), γ is a constant intercept. $θ$ is a slope coefficient, capturing the impact of investor sentiment on the long-term volatility of bond returns. $sent$ represents investor sentiment. K is the maximum lag order of $thesent$⁠, while k is the index of the lag being weighted. $φk$ controls a beta weight, with $ω1$ and $ω2$ measuring beta weights for first and last lag observations, respectively. The beta weights are defined as follows:

Following Anderson and Burnham (2004), Wang and Duxbury (2021), we report the log-likelihood function (LLF) and Akaike’s Information Criterion (AIC) for each regression. We employ Ljung–Box on squared standardized residuals and the ARCH Lagrange multiplier for residual diagnostics as they are widely used in conditional heteroscedasticity models (Tse, 2002). Engle et al. (2013) suggest 4 lags for quarterly data and 12 lags for monthly data. As our sentiment indicators are monthly, we estimate the model with different lag lengths. The results show that lag 4 yields the best model, based on the maximum LLF and the minimum AIC. Therefore, the regression results in the next section are reported using lag 4.

To gain deeper insight into how investor sentiment affects sustainable bond returns across different distributional segments, we apply the QQ regression method developed by Sim and Zhou (2015). This technique extends the standard quantile regression framework by incorporating non-parametric estimation.

The QQ model is derived from the standard quantile approach, with the basic model being:

In equation (4), $regb$ is the daily GSS or individual bond return at time t. $sent$ is either direct or indirect investor sentiment. $τ$ is the $τth$ of the quantile of the bond returns. $utτ$ is the quantile error term. $βτ(·)$ is the slope of the relationship between GSS bond returns and sentiment.

To measure the impact of the $θth$ quantile of the sentiment on the $τth$ quantile of the return, proxied by $sentθ$⁠, $βτ$ can be linearized by the first-order Taylor rule as:

Where $βτ′$ is the partial derivative of $βτ(sentθ)$ and $βτ′(sentθ)$ in terms of $θ$ and $τ$⁠.

Equation (5) can be rewritten as

By substituting equation (6) into equation (4), the last equation is obtained as the QQ model to examine the impact of direct and indirect investor sentiment on daily GSS bond returns as follows:

In equation (7), $β0(θ,τ)+β1(θ,τ)(sentt−sentθ)$ is the conditional quantile of the GSS bond market performance. Following Sim and Zhou (2015), it sets the bandwidth parameter h = 0.05.

As the QQ model does not account for data frequency mismatch, we follow Bethke et al. (2017), Wang and Duxbury (2021) and Le et al. (2025, 2026) to interpolate the sentiment variable from monthly to daily frequency for estimation.

This section analyses our research results. First, we report descriptive statistics for research variables. Next, we discuss the empirical findings of the role of direct and indirect investor sentiment in the performance of the aggregate GSS bond market. The last section presents this relationship in the context of decomposed GSS bond indices.

Table 2 presents the descriptive statistics of the variables used in our research. It shows that the average returns of bond indices are negative ranging from −0.93 to −0.31%, reflecting relatively low returns for these bonds. The coefficients of the Skewness test imply relatively symmetrical data, while the Kurtosis values appear to show heavier tails in the data distribution. The Jarque-Bera statistic also strongly rejects the null hypothesis of normality for our research sample. We employ the standard Augmented Dickey–Fuller (ADF) test to examine whether our time series exhibits a unit root. The results of the ADF tests reject the null of unit root existence for all the ADF test variables at high significance levels.

Table 3 reports GARCH-MIDAS results for the impact of direct and indirect investor sentiment on global aggregate GSS bond index returns. The short-run volatility dynamics are captured by the GARCH parameters α and β, whose sums are close to 1, especially for indirect sentiment (0.9994). This finding confirms the highly persistent nature of volatility and is consistent with the findings of Engle and Rangel (2008) and Engle et al. (2013), who document volatility processes characterized by persistence close to unity. The insignificant statistics of residual diagnostic tests fail to reject the null hypotheses of no residual autocorrelation and no ARCH effects, supporting the adequacy of the estimated GARCH-MIDAS model.

For direct investor sentiment, the θ coefficient is positive and statistically significant at the 1% level (θ = 0.2558), indicating that higher survey-based sentiment increases long-run GSS bond return volatility. The weight coefficient ω is positive but statistically insignificant (ω = 0.4908), suggesting that the weighting structure of past sentiment does not strongly drive persistence in the long-run component.

For indirect investor sentiment, the θ coefficient is also positive and significant at the 1% level (θ = 1.8209) and substantially larger than that of direct sentiment, indicating a much stronger long-run effect on GSS bond return volatility. The weighting scheme ω is positive and statistically significant (ω = 7.0064), implying that the weighting scheme of past turnover information plays an important role in shaping the long-run return component.

To better understand the relationship between investor sentiment and the returns of global GSS bonds, we estimate it using QQ regression. Figure 1 exhibits the heterogeneous effect of investor sentiment on the performance of the global sustainable debt market.

Figure 1 shows how the effects of direct (in Panel A) and indirect (in Panel B) investor sentiment on GSS bond returns vary across the quantiles of both dependent and independent variables. The Z-axis represents the coefficient values, indicating the relationship between investor sentiment variables (Quantiles of Sentix in Panel A and Quantiles of Market Turnover in Panel B on the X-axis) and GSS bond returns (Quantiles of GSS Bond on the Y-axis). Higher Z-axis coefficient values reveal a stronger positive impact on the returns. Both the X- and Y-axes range from 0 to 1, where higher values represent higher returns and more optimistic investor sentiment. The numerical parameters are reported in supplementary Tables A.2 and A.3.

For both sentiment measures, the positive coefficients are concentrated in the upper quantiles of GSS bond returns (approximately 0.6–1), while the negative coefficients mainly appear in the lower quantiles of the returns (around 0.2–0.4). This indicates that higher-return states of the GSS bond market tend to amplify the positive influence of sentiment, whereas lower-return states strengthen the negative relationship.

However, the magnitude and distribution of these effects differ between direct and indirect sentiment. The direct sentiment index shows its strongest effects at the very high return quantiles (around 0.85–1) and a range of lower quantiles (roughly 0.15–0.55). In contrast, the indirect sentiment measure exhibits a more even spread of coefficient strength across almost the entire quantile range (0.15–1), suggesting that indirect sentiment influences GSS bond returns more uniformly across different market conditions.

Given the significant role of investor sentiment in improving global aggregate GSS bond market performance, we explore whether this effect holds for individual bond markets. Accordingly, we separately regress equation (1) on the daily returns of global green, social and sustainable bonds. As in the previous section, we retain explanatory factors, including the global Sentix index and the turnover ratio.

Table 4 reports the estimated results from a GARCH-MIDAS regression on the impact of direct and indirect investor sentiment on the bond returns of three individual GSSs bonds: green, social and sustainability bonds. The GARCH parameters α and β are significant and their sum is close to unity across specifications, indicating slow mean reversion and strong volatility persistence. The Ljung–Box and ARCH-LM test results were insignificant, indicating no evidence of residual autocorrelation or remaining ARCH effects.

For green bond index returns, both direct and indirect investor sentiment shows positive and significant long-run effects through the θ coefficient. Direct sentiment has a positive effect (θ = 0.1208, significant at 5%), indicating that survey-based optimism increases the long-run volatility of green bond returns, although the magnitude is moderate. Indirect sentiment exhibits a much stronger effect (θ = 1.2294, significant at 5%), suggesting that trading-based sentiment plays a more important role.

Regarding social bond returns, both sentiment measures significantly affect long-run performance. Direct sentiment shows a positive and significant θ coefficient (θ = 0.1241, significant at 10%), while indirect sentiment has a larger and highly significant impact (θ = 1.3826). This suggests that social bonds are especially sensitive to market-based sentiment.

Meanwhile, the effects of investor sentiment on sustainable bonds are weaker. Direct sentiment shows a small and marginal θ coefficient (θ = 0.0376), while indirect sentiment is positive but statistically insignificant (θ = 0.1562).

Figure 2 illustrates the visual effect of direct and indirect investor sentiment on the performance of three individual GSS bond markets. The meanings of the axes are the same as in Figure 1 but applied to different subcategories of GSS bonds: Panels A, B and C for green, social and sustainability bonds, respectively, with both direct and indirect sentiment. The numerical coefficients of the QQ model for these bonds are presented in supplementary Tables A.4-9.

As shown in Figure 2, the effects of direct and indirect investor sentiment on the three individual GSS bond indices vary across different quantiles of both sentiment measures and bond returns. Although the magnitudes differ across green, social and sustainability bonds, each sentiment measure exhibits a broadly similar pattern across all three categories. More specifically, the strongest effects of direct sentiment on bond returns concentrate at the upper tails of the return distribution (around 0.85–1) and at the lower quantiles of the Sentix index (around 0.15–0.45). In contrast, the indirect sentiment measure (market turnover) also has the largest impact on the upper return quantiles. However, the effect spreads more evenly across market turnover quantiles rather than being concentrated in a narrow range.

This consistent structure across green, social and sustainability bonds resembles the pattern observed for the combined GSS bond index in the previous analysis, suggesting that both direct and indirect sentiment affect the overall GSS bond market and its sub-indices in a broadly uniform way, although with some differences in strength across quantiles.

This section discusses the empirical findings on the influence of both direct and indirect investor sentiment on the performance of global aggregate GSS bond indices and their decomposition.

The GARCH-MIDAS estimates in Table 3 indicate that both direct and indirect investor sentiment have significant long-run effects on the combined GSS bond index, though the magnitude differs across measures. The MIDAS coefficient of the direct sentiment (θ) is 0.2558; the weighting function with ω = 0.4908 puts 0.0554 on the first lag, and the maximum weight is 0.2554, indicating that a one-unit increase in direct sentiment, measured by the natural logarithm of the Sentix index, of this month is associated with $e0.2558×0.0554−1≈$ 0.0143 or 1.43% rise in the next long-run volatility of GSS bond returns. In the meantime, the slope coefficient of indirect sentiment is 1.8209 with a weight function ω = 7.006, putting weight on the first lag of 0.1232 and the greatest weight of 0.2732. In this case, a one-unit rise in indirect sentiment, proxied by the natural logarithm of market turnover ratio, leads to $e1.8209×0.1232−1≈$ 0.2515, or a 25.15% increase in the next long-run volatility of GSS bond returns. The greater impact of indirect sentiment on GSS bond returns reveals that the market-based indicator captures sharper swings in investor behaviour that translate more strongly into GSS bond returns. These results also imply that the combined GSS bond index reacts more intensely to changes in actual market trading activity than to shifts in stated investor expectations.

The significant effect of direct investor sentiment, proxied by the Sentix index, on the global GSS bond index is consistent with prior research on its impact on conventional financial assets. Bethke et al. (2017) found that higher Sentix sentiment reduced corporate bond credit and liquidity risks, improving bond performance. In equity markets, Wang and Duxbury (2021) reported that Sentix sentiment positively affected the Eurozone’s mean-variance relation but distorted it elsewhere.

Although the role of the market turnover ratio in GSS bonds remains underexplored, the findings align with evidence from stock markets. Baker and Stein (2004) showed that the turnover ratio predicts future US stock returns. Similarly, Wang et al. (2022) found that its effect on stock returns is regime-dependent, with optimistic (pessimistic) sentiment increasing (decreasing) returns in bull markets and having the opposite effect in bear markets.

The QQ surface plots further show that sentiment effects vary across distributions of sentiment and returns. For direct sentiment, stronger positive effects appear at higher return quantiles and lower Sentix quantiles, whereas greater negative impacts are present at lower quantiles of both returns and sentiment. These findings can be explained through behavioural finance and state-dependent pricing mechanisms. Specifically, when the Sentix is at low quantiles, GSS bonds may be undervalued due to excessive pessimism. The strong positive effects observed at higher return quantiles may reflect a correction of this mispricing, as improving expectations drive prices upward. Conversely, when both Sentix and returns are at their lower quantiles, pessimism is reinforced, leading to herding behaviour. Together with limits-to-arbitrage and noise trading, this pattern suggests that sentiment-driven mispricing becomes more pronounced during periods of market stress. These results align with those of Pham and Cepni (2022), who found that investor sentiment plays a significant role in driving green bond returns, particularly during extreme market conditions.

Regarding indirect sentiment, the stronger positive effects observed in the upper quantiles of GSS bond returns suggest that trading activity may amplify positive returns by increasing investor attention. Notably, these impacts are more evenly distributed across turnover quantiles, indicating that indirect sentiment influences the market more uniformly. This reflects the role of turnover as a proxy for market-wide participation and liquidity, rather than sentiment polarity as measured by the Sentix index. Similarly, the negative effects shown in the lower quantiles of returns across all turnover levels imply that trading activity facilitates the transmission of downside risks during market stress, potentially through liquidity pressures and portfolio rebalancing. Overall, the findings from the QQ model imply that while direct sentiment shows asymmetric effects, indirect sentiment yields more uniform impacts on GSS bond returns. There are no prior studies that have employed a quantile-on-quantile framework to examine direct and indirect investor sentiment in the global GSS bonds; however, these results are closely related to Wang et al. (2022), who document state-dependent effects of sentiment on global stock returns.

The GARCH-MIDAS results in Table 4 indicate clear heterogeneity in sentiment effects across green, social and sustainability bonds. For green bonds, both direct and indirect sentiment positively affect long-run returns. However, indirect sentiment shows a much larger impact, highlighting the dominant role of trading activity and market liquidity. Specifically, the MIDAS coefficient for direct sentiment is 0.1208 with a weight scheme of 0.5775, putting 0.1566 weight on the first lag, meaning that a one-unit increase in current direct sentiment is linked with $e0.1208×0.1566−1≈$ 0.019 or 1.9% rise in the next long-term volatility of green bond returns. Using the same computation, there is a 28.55% increase in future green bond returns when indirect sentiment rises by one unit.

A similar pattern is observed for social bonds, where both sentiment measures are significant, but indirect sentiment exerts the strongest influence. In terms of economic magnitude, the slope parameter for direct sentiment is 0.1241 and the weight function is 0.4868, with the first lag bearing only 0.0636 weight. It means that a one-unit increase in present direct sentiment raises $e0.1241×0.0636−1≈$ 0.008 or 0.8% in the long-run volatility of social bond returns, while that of the indirect sentiment is 16.02% (⁠$e1.3826×0.1075−1≈$ 0.1602).

In contrast, sustainability bonds appear to be insensitive to investor sentiment, as the MIDAS coefficients for both direct and indirect investor sentiment are statistically insignificant.

The QQ surface plots support these results. For all three bond types, direct sentiment shows stronger effects at higher return quantiles and lower Sentix quantiles. Indirect sentiment also exhibits stronger effects at high return quantiles, but its influence is more evenly distributed across turnover levels. These patterns are consistent with the combined GSS index, indicating similar sentiment–return relationships across bond categories. Stronger sentiment effects for green and social bonds are consistent with behavioural finance and SRI theories. In contrast, weaker effects for sustainability bonds may reflect their broader, less targeted objectives.

Our findings of stronger effects of indirect sentiment on green and social bonds align with those on combined GSS bonds. Nevertheless, this is not the case for sustainability bonds, whose returns are not affected by investor sentiment, regardless of sentiment measures. The insignificant impact of investor sentiment on sustainability bond returns can be explained by weaker investor preference, given their broad, multi-objective nature. Unlike green or social bonds, which provide a focused thematic signal with specific targets, sustainability bonds combine both environmental and social projects (ICMA, 2019). This creates ambiguity in the use of proceeds and makes it more difficult to assess the intended impact, reducing their suitability for impact investing (ICMA, 2023). Consequently, investors are less favourable towards sustainability bonds than towards green and social bonds, leading to weaker demand responses to sentiment changes (Aruga, 2024), particularly during low-return periods (Aruga and Islam, 2025). This lower investor preference suggests that sustainability bond prices are more likely to be driven by fundamentals than by sentiment fluctuations, which explains the insignificant relationship observed in our research.

The empirical analysis and discussion above rely on a single exogenous sentiment variable, which may lead to biased findings. To address this, we conducted a robustness check by incorporating control variables into the GARCH-MIDAS model. Due to data limitations at the global level, we follow Piñeiro-Chousa et al. (2021, 2022) and Pham and Cepni (2022) in using US inflation expectations, VIX, term spread and conventional bond returns as proxies for global macroeconomic and financial conditions. Details of these variables are provided in Table A.10 in Appendix. The results for the impact of direct and indirect sentiment on combined and individual GSS bonds are reported in Tables A.11 and A.12.

As the ARCH and GARCH conditions (α + β < 1) are satisfied across all estimations, we report only the key parameters. The θ coefficients of direct and indirect sentiment are positive and statistically significant for green and social bonds, but insignificant for direct sentiment and marginal for indirect sentiment (θ = −0.0003) for sustainability bonds. These findings further strengthen our main results above.

Regarding the control variables, most are large in magnitude but statistically insignificant, except for conventional bond returns, whose coefficients remain close to zero across all specifications.

We employ global direct and indirect sentiment proxies and analyze their effects on both an aggregate GSS bond index and decomposed green, social and sustainable bond indices. Using GARCH-MIDAS and quantile-on-quantile (QQ) regression models, we examine how investor sentiment influences GSS bond performance across market conditions.

Our findings confirm that investor sentiment plays a significant role in GSS bond returns. First, aggregate GSS bond performance is driven by economic outlook and market liquidity, with liquidity-related sentiment exerting a stronger influence. Second, sentiment effects observed at the aggregate level are also present in individual GSS bond markets. Among them, green bonds are the most sensitive to sentiment, while sustainable bonds exhibit weaker responses. Third, QQ regression results reveal that sentiment effects differ across quantiles of both sentiment indicators and bond returns, indicating asymmetric, state-dependent relationships.

These results offer important policy implications. Given the strong influence of investor sentiment on GSS bond performance, regulatory frameworks should incorporate behavioural factors such as investor outlook and trading behaviour rather than relying solely on traditional financial indicators. Furthermore, our evidence shows that GSS bond returns are particularly sensitive to indirect sentiment captured by market turnover, highlighting trading behaviour as a key driver of market performance. Current regulatory approaches often emphasize survey-based sentiment measures, which may overlook the real-time behavioural signals embedded in market activity. To foster transparent and liquid markets, regulators should adopt a more comprehensive view of investor sentiment, one that includes both perception-based and market-based indicators, to better anticipate market movements and support the resilience of the sustainable debt market. In addition, the lack of standardized sentiment and performance measures tailored to sustainable finance underscores the need for unified authoritative indicators.

For investors, the documented sentiment effects suggest that incorporating behavioural signals can improve pricing and portfolio decisions in GSS bond markets. The heterogeneous sentiment sensitivity across green, social and sustainable bonds underscores the need for tailored strategies that account for liquidity and trading patterns.

This research is not without limitations. First, we employ the Sentix index and the turnover ratio to proxy for direct and indirect investor sentiment, respectively. Thus, these sentiment measures may not fully reflect sustainable investor behaviour. Second, in the QQ estimation, we interpolate monthly sentiment data to a daily frequency. This treatment may introduce substantial artificial smoothing, leading to biased findings by eliminating the inherent nature of sentiment data that characterizes real-world behavioural reactions. Third, this specific timeframe covers unprecedented macroeconomic shifts. We tested for structural breaks and found that our sample indeed had a break on October 13, 2022. Nonetheless, given that our data span from November 1, 2021, dividing a subsample from that date to the breakpoint yields too few observations. The thin sample size prevents us from partitioning the sample to examine this relationship in the presence of structural breaks. As data availability improves, future studies can explore this relationship more comprehensively by capturing the actual sentiment of sustainable investors across dimensions and frequency and considering time-variant parameters for major shocks.

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

The supplementary material for this article can be found online.