Investor Sentiment and Market Efficiency: Evidence from China’s A-Share Market

Abstract

This study examines how investor sentiment and information transparency jointly shape informational efficiency in China’s A-share market. Using a monthly panel of major CSI stock indices from 2008 to 2023, we measure market efficiency through a price delay framework that captures the speed of information incorporation into prices. The results show that heightened investor sentiment is associated with greater price delay, suggesting that sentiment-driven trading can impede informational efficiency in a retail-dominated market. Importantly, this effect is attenuated in environments with higher information transparency: the interaction between sentiment and transparency indicates that improved disclosure quality weakens the extent to which sentiment distorts price discovery. These findings are robust to instrumental-variable estimation and a range of additional checks. Overall, this study highlights information transparency as a key institutional condition that moderates sentiment-driven inefficiency and provides evidence on the role of disclosure reforms in supporting more efficient price formation in emerging equity markets.

1. Introduction

Despite rapid institutional reforms, China’s A-share market remains a setting where price formation is strongly influenced by investor sentiment rather than fundamentals, largely due to the dominance of retail trading, which accounts for more than 80% of total market volume (China Securities Regulatory Commission, 2023). The dominance of individual traders makes prices highly sensitive to collective emotions and weakens informational efficiency (Shiller, 1981; Baker & Wurgler, 2007). In response, the Chinese securities regulator has introduced stricter disclosure rules—such as the Information Disclosure Management Measures (2021)—to enhance transparency and protect investors. In theory, enhanced disclosure should improve the information environment, reduce reliance on speculative beliefs, and facilitate faster price discovery (Diamond & Verrecchia, 1991; Verrecchia, 2001). Yet a fundamental question remains unresolved: in a sentiment-driven market, can improvements in information transparency meaningfully mitigate sentiment-induced inefficiencies and enhance the speed of price discovery? This question lies at the center of the present study, which investigates how investor sentiment and information transparency jointly shape the efficiency of China’s A-share indices.

Although prior studies have extensively examined market efficiency and behavioral biases, the bulk of this literature is grounded in developed markets characterized by strong institutions, high disclosure quality, and dominant institutional investors, raising concerns about whether their conclusions can be generalized to sentiment-driven emerging markets such as China (Fama, 1970, 1991; Shiller, 2003). More importantly, much of the existing literature examines investor sentiment largely in isolation, implicitly assuming a homogeneous information environment. This approach overlooks a central institutional question: whether the information environment itself conditions how sentiment translates into price inefficiency. (Baker & Wurgler, 2006; Da et al., 2015). This omission is particularly consequential in China, where improving information transparency has been a central objective of ongoing market reforms. If transparency is intended to discipline pricing behavior, its effectiveness should be evaluated precisely in environments where sentiment-driven trading is most pronounced. In practice, transparency should foster rational pricing (Diamond & Verrecchia, 1991), while sentiment often drives markets away from fundamentals, creating a persistent tension in price formation. This study is motivated by that tension. By explicitly modeling information transparency as an institutional moderator, this study addresses this gap by testing whether improvements in the information environment can attenuate sentiment-induced inefficiencies, thereby refining traditional efficiency frameworks and extending their applicability to retail-dominated emerging markets.

This study makes three main contributions to the literature on investor sentiment, market efficiency, and information transparency. First, while prior studies link investor sentiment to return anomalies (Stambaugh et al., 2012) or to broader market conditions such as liquidity and volatility (Zhou, 2025), this study advances the behavioral finance literature by linking investor sentiment to the speed of information incorporation rather than to price levels or return predictability alone. By adopting a price delay framework, it characterizes sentiment as a force that can impair informational efficiency through slower price adjustment (Chordia et al., 2005; Hou & Moskowitz, 2005). Second, this study introduces information transparency as a key institutional boundary condition that moderates the sentiment–efficiency relationship. Rather than treating transparency as a background market feature, the analysis emphasizes that the effectiveness of price discovery critically depends on the information environment in which sentiment-driven trading occurs (Fang & Peress, 2009; Tetlock, 2007). In contrast to studies emphasizing macro-emotional shocks such as national pride (Abudy et al., 2023), this study focuses on how institutional transparency shapes the transmission of aggregate sentiment into informational inefficiency. Third, by focusing on China’s A-share market—where retail participation is high and disclosure reforms are ongoing—this study provides novel evidence on how institutional transparency shapes behavioral inefficiency in emerging markets. The findings offer insights that may be informative beyond the China setting, illustrating how disclosure regimes can discipline sentiment-driven mispricing in markets undergoing structural transition.

This study integrates insights from behavioral finance, which emphasizes investor sentiment, and the information-based view of markets, which highlights disclosure quality and institutional transparency (Baker & Wurgler, 2007; Tetlock, 2007; Diamond & Verrecchia, 1991; Verrecchia, 2001). We conceptualize transparency as an institutional condition that shapes how strongly investor sentiment translates into informational inefficiency. Using the price delay measure as a direct proxy for the speed of information incorporation, we examine whether the information environment moderates the sentiment–efficiency link in China’s A-share market (Fama, 1991; Chordia et al., 2005). Beyond the China setting, the analysis highlights the policy relevance of disclosure reforms in sentiment-sensitive markets: improving transparency may help limit the extent to which sentiment-driven trading distorts price discovery.

The remainder of this paper is organized as follows. Section 2 reviews the literature on market efficiency, investor sentiment, and information transparency and develops the research hypotheses. Section 3 describes the data sources, variable construction, and empirical model used to test these hypotheses. Section 4 reports the empirical results and robustness checks. Section 5 concludes the paper.

2. Literature Review and Hypotheses

2.1. Market Efficiency and Its Measurement

2.1.1. The Efficient Market Hypothesis and Its Challenges

The Efficient Market Hypothesis (EMH) provides the foundational framework for understanding price formation in financial markets. In its classical formulation, EMH posits that asset prices fully and instantaneously reflect all available information, implying that price changes follow a random walk and that systematic excess returns are unattainable (Fama, 1970, 1991). Under this framework, investors are assumed to be rational, information is processed without bias, and any deviations from fundamental value are quickly eliminated through arbitrage.

However, extensive empirical evidence suggests that market efficiency is neither universal nor constant. Financial markets frequently exhibit delayed reactions to public information, excess volatility, and predictable return patterns that are difficult to reconcile with the strict assumptions of EMH (Shiller, 1981, 2003). These deviations are especially pronounced in environments where arbitrage is constrained by short-selling restrictions, funding limits, or institutional frictions (Shleifer & Vishny, 1997). In such settings, mispricing may persist even when information is publicly available.

Importantly, recent research highlights the role of investor sentiment in amplifying these inefficiencies. Stambaugh et al. (2012) show that pricing anomalies become significantly stronger following periods of high investor sentiment, particularly when arbitrage is limited. Their findings suggest that emotionally driven trading can delay price correction and weaken informational efficiency, challenging the EMH assumption that information alone guarantees efficient pricing. Taken together, these studies imply that market efficiency should be viewed as conditional and state-dependent rather than a fixed property, motivating frameworks that allow efficiency to vary across markets and over time—an issue that is particularly relevant in sentiment-sensitive markets such as China’s A-share market. Recent evidence further extends the role of investor sentiment beyond cross-sectional return anomalies. For example, Abudy et al. (2023) show that macro-level emotional shocks, such as national pride, can significantly influence aggregate stock market performance through sentiment channels. Similarly, Zhou (2025) documents that fluctuations in investor sentiment are closely associated with changes in liquidity and volatility, suggesting that sentiment affects broader dimensions of market quality. While these studies provide important evidence on how sentiment shapes returns and trading conditions, relatively little is known about whether sentiment directly impairs the speed of information incorporation into prices. By shifting the focus from return outcomes to informational efficiency, this study contributes to the literature by examining how sentiment influences price adjustment dynamics under varying institutional transparency conditions. In the EMH framework, market efficiency is commonly classified into weak-, semi-strong-, and strong-form efficiency (Fama, 1970, 1991). Weak-form efficiency implies that all information contained in past prices is reflected in current prices, so returns should not be predictable from their own history and prices should follow a random-walk process. Semi-strong-form efficiency extends this view to all publicly available information, while strong-form efficiency assumes even private information is incorporated into prices. Because this study focuses on return predictability and delayed price adjustment at the index level, the analysis is most closely aligned with weak-form informational efficiency.

2.1.2. Measurement: Price Delay and Return Autocorrelation

To reconcile the tension between EMH and behavioral evidence, Lo (2002) proposes the Adaptive Market Hypothesis (AMH), which views market efficiency as an evolving, state-dependent outcome rather than a fixed property. Under the AMH, markets can exhibit alternating efficiency across different states (e.g., bull versus bear phases, or periods of high versus low sentiment), because investor behavior, competition, and institutional conditions change over time. This perspective is particularly relevant for China’s A-share market, where retail participation is high and the information environment is shaped by ongoing regulatory reforms. Consistent with the AMH perspective, this study evaluates informational efficiency through the speed of price adjustment and examines whether sentiment weakens this process and whether information transparency serves as an institutional condition that dampens sentiment-driven inefficiency.

If market efficiency is conditional rather than constant, it must be evaluated by how quickly prices incorporate available information. Accordingly, the literature increasingly measures informational efficiency through the speed of price adjustment rather than assuming instantaneous incorporation.

Following Chordia et al. (2005), this study adopts the price delay measure to capture delays in information processing. In an informationally efficient market, current returns should primarily reflect contemporaneous information, leaving little explanatory power for lagged market returns. When past market movements continue to explain current returns, prices adjust gradually rather than immediately, indicating lower informational efficiency.

Empirically, price delay is proxied by the R² statistic from regressions of index returns on contemporaneous and lagged market returns. A higher R² implies greater dependence on past information and slower price adjustment. Compared with anomaly-based or volatility-based measures, price delay directly reflects the information incorporation mechanism emphasized in the Efficient Market Hypothesis and its behavioral critiques, making it particularly suitable for markets where information diffusion may be impeded (Chordia et al., 2005; Hou & Moskowitz, 2005).

Price adjustment, however, may also be influenced by broader economic conditions. Changes in inflation affect real returns, interest rate movements alter discount rates, and exchange rate fluctuations influence capital flows and external risk exposure, especially in emerging markets. To ensure that price delay captures informational frictions rather than macroeconomic fluctuations, this study controls for inflation (CPI), interest rates (IRs), and exchange rates (REXs) in the empirical analysis.

As a robustness check, return autocorrelation is employed as a complementary indicator of market efficiency. Persistent autocorrelation suggests gradual price adjustment and reinforces the interpretation of delayed information incorporation. Prior evidence indicates that such patterns are more pronounced in markets dominated by retail investors and subject to institutional frictions, including China’s A-share market (Hou & Moskowitz, 2005).

By combining the price delay measure with controls for macroeconomic conditions, this study provides a coherent framework for assessing informational efficiency and isolating the effects of investor sentiment and institutional features on price discovery.

This approach is particularly appropriate for China’s A-share market, where information diffusion is often impeded by retail-dominated trading, short-selling constraints, and episodic regulatory interventions, making price adjustment speed a natural indicator of time-varying market efficiency.

2.2. Investor Sentiment: Conceptual Definition and Empirical Construction

Departing from the EMH’s assumption of fully rational investors, behavioral finance emphasizes systematic cognitive biases and limits to arbitrage as key reasons why sentiment-driven mispricing and delayed information incorporation can persist. Following the investor sentiment literature, we construct a composite sentiment index using a small set of widely accepted market-based proxies that capture different dimensions of speculative behavior and investor optimism. Prior studies emphasize that investor sentiment is a latent psychological factor that cannot be directly observed and is therefore best proxied by multiple indicators reflecting mispricing, speculative demand, retail participation, and trading intensity (Baker & Wurgler, 2006; Stambaugh et al., 2012).

Specifically, the four proxies employed in this study—closed-end fund discounts, IPO first-day returns, new investor accounts, and turnover rate—have been repeatedly validated in both international and China-specific settings as core sentiment indicators. Together, they provide a parsimonious yet comprehensive representation of investor sentiment while avoiding redundancy and excessive noise that may arise from including highly correlated or institution-specific measures.

Investor sentiment refers to the collective mood and expectations of investors toward the market, which can drive prices to deviate from their fundamental values. It reflects how investors feel about future returns rather than what rational valuation models would imply. In practice, even when economic fundamentals remain stable, shifts in optimism or pessimism can still trigger large swings in asset prices.

Baker and Wurgler (2007) argue that investor sentiment is not simply random noise but a systematic psychological component of asset pricing. When sentiment is optimistic, investors tend to trade more aggressively and bid up prices beyond intrinsic values. Conversely, during pessimistic phases, selling pressure may push prices below fundamentals. Brown and Cliff (2004) further find that sentiment cycles typically create short-term momentum followed by reversals, suggesting that emotions can move markets away from equilibrium before corrections occur.

Behavioral finance theory offers explanations for these patterns. Barberis et al. (1998) describe representativeness bias, where investors overreact to recent trends and form overly optimistic expectations, and conservatism bias, where they adjust beliefs too slowly when confronted with new information. Shleifer and Vishny (1997) also point out that even rational investors face practical limits—such as financing constraints and short-selling restrictions—that prevent them from eliminating sentiment-driven mispricing immediately. Together, these behavioral and institutional frictions help explain why sentiment-induced inefficiencies can persist for extended periods.

These dynamics are particularly relevant in China’s A-share market, where retail investors account for the majority of trading volume. Individual investors tend to rely heavily on online discussions and social media—such as EastMoney Guba or Weibo—to interpret news and form market opinions. As emotional narratives spread rapidly across these platforms, herding behavior intensifies. This collective reaction can amplify market swings, leading to synchronized optimism or panic that pushes prices away from fundamentals.

Because investor sentiment cannot be directly observed, researchers typically rely on composite indices constructed from observable market variables. Baker and Wurgler (2006) pioneered such an approach in the U.S., combining variables like IPO activity, first-day returns, dividend premium, and turnover rate. Chen et al. (2014) adapted this framework to China, integrating turnover, closed-end fund discounts, new investor accounts, and share issuance. Later studies, such as Da et al. (2015) extended the concept by incorporating text-based measures derived from search activity and financial news sentiment.

Formally, the Investor Sentiment Index (ISI) is constructed as an equal-weighted composite of the standardized sentiment proxies. Following Baker and Wurgler (2006), each component is first standardized to have zero mean and unit variance, and the closed-end fund discount is multiplied by −1 so that higher values consistently indicate stronger optimism. The sentiment index in period t is defined as follows:

ISI_t = 1/4[Z(−DCEF_t) + Z(RIPO_t) + Z(NA_t) + Z(TURN_t)]

where Z(⋅) denotes the standardized value of each variable.

Similar composite constructions have been adopted in recent studies examining the effects of investor sentiment on stock return indices. Moodley et al. (2025) for instance, formalize investor sentiment as an equally weighted combination of standardized sentiment proxies, highlighting that such a formulation provides a clear and interpretable measure of market-wide sentiment dynamics. Following this established practice, we adopt the same mathematical structure to ensure comparability and methodological consistency.

This construction captures the common variation across multiple sentiment-related market indicators while avoiding reliance on any single proxy, consistent with the composite-index approach widely adopted in the sentiment literature.

2.3. The Impact of Investor Sentiment on Market Efficiency

Building on the preceding discussion of investor sentiment and market efficiency, this section examines how emotion-driven trading behavior distorts the process through which information is incorporated into prices. When investors rely more on collective mood than on fundamentals, prices may deviate from intrinsic values and adjust more slowly to new information, resulting in lower informational efficiency.

A growing body of research documents that investor sentiment plays an important role in shaping market efficiency. Rather than affecting only the direction of price movements, sentiment influences how efficiently markets process and reflect available information. Periods of elevated optimism or pessimism are often associated with delayed price adjustment and increased return predictability, particularly in markets where arbitrage is limited.

One strand of the literature attributes sentiment-driven inefficiency to cognitive biases in investor behavior. Behavioral finance models emphasize that investors do not always update beliefs rationally when new information arrives. Barberis et al. (1998) show that representativeness and conservatism biases can lead investors to overreact to recent trends or underreact to fundamental signals, causing prices to deviate persistently from intrinsic values. Consistent with this view, Brown and Cliff (2004) find that extreme investor sentiment is associated with predictable return patterns and subsequent reversals. Evidence from China’s A-share market further supports this mechanism. Yu and Yuan (2011) document that investor sentiment is closely related to return autocorrelation, suggesting that sentiment-driven trading contributes to delayed price adjustment in markets dominated by retail investors.

A second strand of research emphasizes limits to arbitrage as a key channel through which sentiment affects market efficiency. Even when mispricing is recognized, rational investors may be unable or unwilling to correct it due to capital constraints, short-selling restrictions, and the risk that sentiment persists longer than expected (Shleifer & Vishny, 1997). Under such conditions, sentiment-induced price distortions may survive for extended periods, slowing the incorporation of fundamental information into prices. Using market-based measures of informational efficiency, Chordia et al. (2005) show that markets exhibiting stronger return predictability tend to adjust more slowly toward efficiency. Hou and Moskowitz (2005) further demonstrate that such inefficiencies are more pronounced in emerging markets, where institutional frictions and information barriers are more severe.

Taken together, these behavioral and institutional mechanisms imply that heightened investor sentiment impedes the timely incorporation of information into prices and increases return predictability. Accordingly, this study hypothesizes that stronger investor sentiment is associated with lower market efficiency, as reflected in greater price delay.

Although prior studies have established that investor sentiment influences return anomalies (Stambaugh et al., 2012) and broader market conditions such as liquidity and volatility (Zhou, 2025), most of this literature evaluates sentiment effects without explicitly considering the institutional information environment. Moreover, studies such as Abudy et al. (2023) emphasize the macro-psychological origins of sentiment but do not examine how institutional transparency conditions its market consequences. As a result, the moderating role of disclosure quality in shaping sentiment-driven informational inefficiency remains underexplored, particularly in retail-dominated emerging markets.

2.3.1. Cognitive-Bias Channel: Sentiment-Induced Predictability in Returns

Behavioral finance suggests that investors often rely on intuitive judgment rather than fully rational information processing when making trading decisions. Barberis et al. (1998) identify two key cognitive biases underlying such behavior: representativeness bias, whereby investors extrapolate recent trends into the future and overestimate their persistence, and conservatism bias, whereby investors adjust beliefs slowly in response to new information. As a result, prices may overreact to past signals or underreact to fundamental news, leading to gradual and predictable price adjustments.

These alternating waves of optimism and pessimism generate return dynamics that deviate from the random-walk behavior implied by the Efficient Market Hypothesis (Fama, 1970). When belief updating is distorted, past price movements may contain information about future returns, giving rise to serial correlation and short-term predictability. Brown and Cliff (2004) find that extreme investor sentiment in the U.S. stock market is associated with predictable return reversals, while Yu and Yuan (2011) document similar patterns in China’s A-share market, where heightened sentiment is linked to stronger return autocorrelation.

Together, this evidence suggests that cognitive biases embedded in investor sentiment weaken informational efficiency by slowing belief updating and price adjustment. In markets with strong retail participation, such as China’s A-share market, these effects are particularly pronounced, making sentiment-induced predictability an important source of informational inefficiency.

2.3.2. Limits-to-Arbitrage Channel: Sentiment-Induced Delays in Price Adjustment

Even when mispricing is recognized, it may not be corrected immediately. In principle, rational arbitrageurs should exploit such opportunities and restore prices to their fundamental values. However, as Shleifer and Vishny (1997) emphasize, arbitrage is inherently risky and constrained by capital limitations, short-selling restrictions, and the possibility that sentiment-driven mispricing persists or worsens before converging. Faced with these risks, even informed investors may refrain from acting aggressively, allowing mispricing to endure.

These constraints are particularly salient in China’s A-share market, where retail investors dominate trading activity and short-selling mechanisms remain limited. During periods of heightened optimism or pessimism, sentiment-driven trades can overwhelm rational forces, causing prices to adjust gradually rather than instantaneously to new information. Chordia et al. (2005) capture this phenomenon through the price delay coefficient (R²), which measures the extent to which current returns are explained by lagged market movements. A higher R² indicates slower information incorporation and lower informational efficiency.

Empirical evidence supports the relevance of this mechanism in emerging markets. Chen et al. (2014) show that stronger investor sentiment in China is associated with greater price delay and larger deviations from fundamentals, consistent with constrained arbitrage. Related studies further suggest that institutional frictions and retail investor dominance amplify such effects, allowing sentiment-induced inefficiencies to persist over time (Hou & Moskowitz, 2005).

Overall, limits to arbitrage provide an institutional explanation for why sentiment-driven mispricing is not quickly eliminated. When rational correction is impeded, investor sentiment can translate into sustained delays in price adjustment, reinforcing the link between sentiment and reduced market efficiency.

2.3.3. Hypothesis Development

Both theoretical mechanisms indicate that sentiment weakens informational efficiency, though in different ways. The cognitive-bias channel reduces accuracy by making returns more predictable, while the limits-to-arbitrage channel reduces speed by slowing price adjustments. Together, they imply that sentiment-driven markets will exhibit higher return autocorrelation and greater price delay, both reflected in higher R² values—signifying lower informational efficiency.

Accordingly, the first hypothesis is formulated as follows:

H1. 

Investor sentiment has a significant negative impact on the informational efficiency of China’s A-share indices, leading to higher price delay and stronger return autocorrelation.

2.4. Information Transparency as a Moderator Between Investor Sentiment and Market Efficiency

The information transparency indicator in this study is constructed based on firm-level disclosure rating data of listed companies. Since disclosure ratings are reported at the firm level and on an annual basis, to capture time variation in the overall information disclosure environment of China’s A-share market, we take the average disclosure rating of all listed companies in each year to form an annual, market-level measure of information transparency.

Specifically, for year Y, market transparency is defined as the arithmetic mean of disclosure ratings across all A-share listed firms in that year. The annual transparency value is then matched to all months within the corresponding year and assigned to the monthly observations of all indices.

This construction allows the Transparency variable to reflect dynamic changes in the overall disclosure environment of China’s A-share market rather than cross-sectional differences across indices, making it suitable for examining how the institutional information environment moderates the effect of investor sentiment on market efficiency.

Information transparency is widely regarded as a key institutional factor influencing market efficiency. A transparent information environment improves the availability, credibility, and interpretability of information, thereby reducing information asymmetry among market participants. In such an environment, prices are more likely to reflect fundamental information rather than noise or speculative beliefs.

Early theoretical studies emphasize the role of disclosure in enhancing market efficiency through improved information flow. Diamond and Verrecchia (1991) argue that higher disclosure quality reduces information asymmetry and increases market liquidity, facilitating faster incorporation of information into prices. Verrecchia (2001) further conceptualizes disclosure as a mechanism that shapes investors’ information sets and trading incentives, influencing how information is processed in financial markets. These studies establish information transparency as a fundamental determinant of informational efficiency.

Empirical research provides further evidence on the economic consequences of transparency. Leuz and Verrecchia (2000) show that improved disclosure quality reduces information asymmetry and estimation risk, while Hail and Leuz (2006) document that stronger disclosure regimes are associated with lower costs of capital across countries. Together, the literature suggests that a more transparent information environment improves the conditions under which investors form expectations and trade, thereby affecting the efficiency of price formation.

More recent studies highlight that the effect of information transparency may depend on investor behavior and market conditions, particularly in the presence of strong investor sentiment. When disclosure quality is high, investors are better able to distinguish fundamental signals from speculative or emotion-driven information. Evidence from China’s stock market indicates that firms with higher disclosure quality experience weaker sentiment-driven price distortions (Leuz & Verrecchia, 2000). Similarly, Fang and Peress (2009) show that a richer information environment reduces information asymmetry and weakens mispricing driven by non-fundamental factors. These findings suggest that transparency can constrain the extent to which sentiment-driven beliefs are reflected in prices.

In markets characterized by strong retail participation and episodic sentiment-driven trading, information transparency does not eliminate investor sentiment but limits its impact on market efficiency by anchoring investor expectations to credible information. By reducing uncertainty and reliance on informal or speculative signals, a transparent information environment weakens the channels through which investor sentiment distorts informational efficiency. Based on this line of research, the following hypothesis is proposed:

H2. 

Information transparency weakens the negative impact of investor sentiment on market efficiency. Specifically, when transparency is high, the positive association between investor sentiment and measures of informational inefficiency becomes less pronounced.

Overall, the literature suggests that investor sentiment is linked to return predictability and slower price adjustment, especially when arbitrage is constrained, while a more transparent information environment improves price discovery by reducing information asymmetry. Yet most empirical studies examine sentiment and transparency largely in isolation, offering limited evidence on how the information environment conditions sentiment’s impact on informational efficiency—particularly in retail-dominated emerging markets. By integrating these two strands and testing a moderation mechanism in China’s A-share indices, this study clarifies when sentiment-driven inefficiency is more likely to emerge and when institutional transparency can mitigate it.

3. Data and Methodology

3.1. Data

This study employs monthly panel data obtained from the CSMAR database, covering the period from January 2008 to December 2023. The sample consists of nine representative CSI indices in China’s A-share market: 000016 (SSE 50 Index), 000300 (CSI 300 Index), 000802 (CSI 500 Index), 000852 (CSI 1000 Index), 000951 (CSI 300 Value Index), 399908 (CSI Environmental Protection Index), 399915 (CSI 1000 Growth Index), 399918 (CSI 500 Growth Index), and 399919 (CSI 300 Growth Index). Data frequency and sample period. The dataset used in this study consists of monthly data from January 2008 to December 2023, covering nine CSI indices. While we refer to the data as a balanced panel, some months are missing data for a few indices due to limitations in the CSMAR database. These missing months are relatively few and do not significantly affect the overall sample. The total number of observations reported (1317) reflects the available data after excluding any months with missing data for individual indices. This clarification ensures that the sample used in the analysis is clearly understood in terms of index coverage and missing months. Detailed descriptions of all indices are provided in

Table A2. Index Descriptions. This table show the description of indices tested in the main text.

Index Code	Index Name	Description
000016	SSE 50 Index	Top 50 large-cap blue-chip stocks listed on the Shanghai Stock Exchange.
000300	CSI 300 Index	300 representatives large- and mid-cap stocks from Shanghai & Shenzhen markets.
000802	CSI 500 Index	Mid-cap stocks excluding CSI 300 constituents; reflects broader market trends.
000852	CSI 1000 Index	Small-cap stocks providing exposure to the broader innovative sector of A-shares.
000951	CSI 300 Value Index	Constituents of CSI 300 selected based on strong value characteristics.
399908	CSI Environmental Protection Index	Firms operating in environmental protection and green technology industries.
399915	CSI 1000 Growth Index	Small-cap companies with high growth potential and innovation capacity.
399918	CSI 500 Growth Index	Mid-cap firms characterized by strong earnings growth expectations.
399919	CSI 300 Growth Index	Large- and mid-cap firms with significant growth features and performance momentum.

.

Data Frequency and Sample Period

This study employs monthly data rather than daily observations. The choice of monthly frequency is motivated by both methodological and conceptual considerations. Daily returns in emerging markets such as China’s A-share market are often subject to substantial microstructure noise, short-term trading frictions, and liquidity effects, which may obscure the relationship between investor sentiment and market efficiency (French & Roll, 1986; Lo & MacKinlay, 1990). Monthly aggregation helps mitigate these high-frequency disturbances and allows sentiment and disclosure-related effects—typically slow-moving in nature—to be more clearly identified. Moreover, the monthly frequency aligns with the construction of our Investor Sentiment Index based on four market-based proxies measured at the monthly level, and with the availability of macroeconomic controls (CPI, interest rate, and exchange rate), which are typically released at a monthly frequency. Using a consistent monthly scale ensures proper time matching across variables and avoids mixed-frequency measurement noise.

The sample period spans January 2008 to December 2023, reflecting the longest period for which all required variables, including sentiment proxies and information disclosure ratings from CSMAR, are consistently available. This period covers multiple market cycles, regulatory adjustments, and episodes of heightened investor sentiment, providing a suitable setting for examining the dynamic relationship between sentiment and market efficiency.

This study focuses on nine major CSI indices rather than firm-level returns for two reasons. First, our core constructs—investor sentiment and the disclosure environment—are conceptually market-wide and are measured at aggregate frequencies (monthly sentiment; annual disclosure rating). Index returns provide a natural mapping from these market-level drivers to an outcome that reflects market-wide price formation, while reducing firm-specific noise that is orthogonal to broad sentiment and disclosure conditions. Second, these indices collectively represent a broad cross-section of the Chinese stock market—including large-cap, mid-cap, and small-cap portfolios as well as value, growth, and sectoral components—providing a comprehensive foundation for analyzing the relationship between investor sentiment, information transparency, and market efficiency.

We acknowledge that the cross-sectional dimension is limited (nine index panels), and therefore, inference must be interpreted with appropriate caution. To mitigate concerns about over-reliance on a small number of entities, all specifications include index and year fixed effects and report standard errors clustered at the index level, and we complement baseline results with multiple robustness checks.

The dependent variable is Market Efficiency, measured by the price delay coefficient (R²) following the framework of Chordia et al. (2005). The R² value reflects the degree of price delay, or how quickly index returns respond to market-wide information. A higher R² indicates slower price adjustment and thus lower informational efficiency.

The independent variable, the Investor Sentiment Index (ISI), is constructed to capture both trading enthusiasm and participation dynamics in China’s A-share market. The ISI is based on four monthly market-based indicators from 2014 to 2023: (1) Discount of Closed-End Funds (DCEF), reflecting deviations between market prices and NAVs and thus aggregate optimism or pessimism; (2) IPO First-Day Return (RIPO), indicating speculative demand for new listings; (3) New Investor Accounts (NA), representing retail participation intensity; (4) Turnover Rate (TURN), capturing overall trading activity and market exuberance.

Each variable is standardized to remove scale differences and then aggregated into a single composite sentiment index following Baker and Wurgler (2006) and its adaptation to China’s market by Chen et al. (2014). A higher ISI indicates stronger investor optimism and elevated speculative trading. Descriptive statistics and correlations for sentiment components and ISI are reported in

Table 1. Descriptive Statistics. R² represents market efficiency, measured using the price delay indicator. The Sentiment Index is a composite indicator constructed from four commonly used sentiment proxies: the discount of closed-end funds (DCEF), IPO first-day returns (RIPO), market turnover rate (TURN), and the number of new investor accounts (NA). Transparency reflects the annual, market-level mean disclosure rating across all A-share listed firms in China, obtained from CSMAR. This measure captures the overall information disclosure environment of China’s A-share market and allows the Transparency variable to reflect dynamic changes in disclosure quality over time.

	N	Mean	SD	Min	Median	Max
R2	1317	0.21	0.150	0.000	0.175	0.844
Sentiment Index	1317	0.051	0.566	−1.25	−0.013	2.043
Transparency	1317	1.991	0.034	1.921	1.994	2.045
Volatility	1317	0.197	0.179	0.017	0.113	0.979
LnTrading	1317	15.449	1.268	11.474	15.528	18.375
CPI	1317	1.093	1.148	−0.900	1.000	5.400
IR	1317	3.070	0.295	2.790	2.900	4.140
REX	1317	6.580	0.305	6.051	6.573	7.249

and

Table 2. Correlations of all variables. This table reports Pearson correlation coefficients based on 1317 monthly observations across nine Chinese A-share indices from 2008 to 2023. The variables include market efficiency (R²), investor sentiment (Sentiment Index), information transparency (Transparency), trading condition controls (Volatility, LnTrading), and macroeconomic indicators (REX, IR, CPI). *, **, and *** denote statistical significance at the 10%, 5%, and 1% levels, respectively.

	R2	Sentiment Index	Transparency	Volatility	LnTrading	REX	IR	CPI
R2	1	−0.042	0.062 **	−0.230 ***	0.113 ***	0.084 ***	−0.141 ***	0.186 ***
Sentiment Index	−0.004	1	−0.169 ***	0.083 ***	0.112 ***	−0.175 ***	0.080 ***	−0.066 **
Transparency	0.045 *	−0.165 ***	1	−0.575 ***	0.149 ***	0.251 ***	−0.317 ***	0.403 ***
Volatility	−0.208 ***	−0.021	−0.430 ***	1	−0.367 ***	−0.255 ***	0.291 ***	−0.610 ***
LnTrading	0.098 ***	0.163 ***	0.155 ***	−0.417 ***	1	0.052 *	−0.176 ***	0.294 ***
REX	0.069 **	−0.218 ***	0.304 ***	−0.111 ***	0.033	1	−0.536 ***	0.317 ***
IR	−0.173 ***	0.027	−0.383 ***	0.412 ***	−0.250 ***	−0.224 ***	1	−0.309 ***
CPI	0.171 ***	−0.023	0.448 ***	−0.562 ***	0.295 ***	0.343 ***	−0.348 ***	1

.

The moderating variable is Information Transparency, measured using the annual disclosure ratings of listed firms provided by CSMAR. The rating reflects the quality, credibility, and clarity of corporate information disclosure. A higher score indicates a more transparent and reliable information environment, consistent with the theoretical frameworks of Diamond and Verrecchia (1991) and Fang and Peress (2009). This variable serves as a market-level proxy for information quality and disclosure consistency. Specifically, because disclosure ratings are reported at the firm-year level, we construct a market-level transparency measure by averaging disclosure ratings across all A-share listed firms in each year. The resulting annual transparency indicator is then matched to all monthly observations within the corresponding year and assigned to each index. This approach allows the Transparency variable to capture time variation in the overall disclosure environment of China’s A-share market rather than cross-sectional differences across indices. Following the disclosure literature that treats disclosure quality as an economy-wide information environment shaped by securities regulation and enforcement (Leuz & Verrecchia, 2000; Hail & Leuz, 2006), we operationalize transparency as a market-wide annual indicator.

To account for market-wide factors that may influence market efficiency, we include a set of control variables capturing trading conditions and macroeconomic fundamentals. Specifically, Market Volatility (Volatility) is measured as the standard deviation of monthly index returns to proxy for overall uncertainty and market risk, while Trading Volume reflects liquidity intensity and the degree of investor participation. In addition, we control for key macroeconomic conditions, including the interest rate, inflation rate, and exchange rate, which may jointly affect investor behavior and the speed at which information is incorporated into prices.

These controls are introduced to ensure that the estimated relationship between investor sentiment and market efficiency is not confounded by general market conditions.

After merging all variables, the final dataset forms a balanced monthly panel with 1317 observations across nine indices. This structure enables consistent comparison across markets and over time. To eliminate unobservable heterogeneity, all subsequent regressions include index fixed effects and time fixed effects, allowing the analysis to isolate the within-index effects of investor sentiment and information transparency on market efficiency.

3.2. Methodology

3.2.1. Baseline Regression Equation

Building on the previous sections, this study examines how investor sentiment and information transparency jointly affect market efficiency across nine major indices in China’s A-share market.

To achieve this, a panel regression model is employed, which integrates both time-series and cross-sectional dimensions:

$$Y_{it}=\alpha +{\beta }_1{X}_t+{\beta }_2{X}_{1t}+{\beta }_3{(X}_t\times X_{1t})+\gamma Z_{it}+{\mu }_t+{\lambda }_t+{ϵ}_{it}$$

(1)

Here, $Y_{it}$ represents the efficiency of index i at time t, measured by the price delay coefficient (R²). A higher R² value indicates slower price adjustment and, consequently, lower informational efficiency. Sentiment captures overall investor sentiment, while transparency reflects the average level of market transparency. The interaction term Sentiment × Transparency tests whether transparency moderates the effect of investor sentiment on market efficiency.

Given the panel structure of the data, we use individual fixed effects (${\alpha }_i$) to account for unobserved heterogeneity across indices and time fixed effects (${\lambda }_t$) to account for unobserved factors that vary over time. This ensures that the analysis controls for potential biases arising from index-specific and time-specific effects. We formally assess whether a fixed-effects (FE) or random-effects (RE) specification is more appropriate. We estimate both FE and RE models using the same set of regressors and conduct a Hausman specification test, where the null hypothesis is that the RE estimator is consistent, that is, the unobserved index-specific effects are uncorrelated with the explanatory variables. The test strongly rejects the null hypothesis (χ²(5) = 328.33, p < 0.001), indicating that the RE specification is inconsistent in this setting. Accordingly, the fixed-effects (FE) specification is adopted in all baseline and subsequent regressions. In this model, ${\alpha }_i$ represents individual fixed effects (e.g., unobserved differences across indices) and represents time fixed effects (e.g., unobserved variations over time).

Control variables $Z_{it}$ include market volatility, trading volume, and turnover rate, which account for market-wide risk and liquidity differences. Both index and time fixed effects (${\mu }_i,{\lambda }_t$) are incorporated to control for unobservable heterogeneity across indices and over time.

In this specification, a positive estimate of β₁ indicates that stronger investor sentiment increases price delay and thus reduces market efficiency. The coefficient β₃ captures the moderating role of transparency; a negative estimate of β₃ would indicate that higher transparency alleviates this inefficiency by weakening the sentiment effect—enhancing the speed and accuracy of price discovery.

3.2.2. Dummy Variable

To further test the moderating effect of information transparency, a dummy variable ($D_t$) is introduced to classify the observations into high- and low-transparency regimes. Specifically, $D_t$ = 1 when the transparency score is above its annual median, and $D_t$ = 0 otherwise. This approach allows for a straightforward comparison between periods characterized by different disclosure environments.

The model is then re-estimated by including the interaction term, which tests whether the effect of investor sentiment on market efficiency varies across transparency regimes. If the estimated coefficient of the interaction term becomes smaller or statistically insignificant during high-transparency periods, it indicates that the influence of investor sentiment on price delay weakens when information disclosure is stronger. This finding would suggest that information transparency mitigates sentiment-driven inefficiency, reinforcing its stabilizing role in the price discovery process.

3.2.3. Measurement of Key Variables

To ensure consistency and robustness, this study employs well-established measures for each key construct.

Market efficiency is measured by the price delay coefficient (R²) following Chordia et al. (2005). This indicator captures how quickly new information is incorporated into index prices. A higher R² value indicates a greater degree of price delay and thus lower informational efficiency. In robustness tests, return autocorrelation is also employed as a supplementary indicator to validate the consistency of the results, as higher autocorrelation implies slower price adjustment and persistent inefficiency.

Investor sentiment is measured using a composite index constructed from four widely used market-based indicators that capture different dimensions of investor optimism and speculative activity in China’s A-share market: the discount of closed-end funds (DCEF), IPO first-day returns (RIPO), the number of new investor accounts (NA), and the market turnover rate (TURN). To mitigate scale effects and distributional skewness, the number of new investor accounts (NA) is transformed using the natural logarithm.

Each component is standardized to have zero mean and unit variance. Following the composite index approach proposed by Baker and Wurgler (2006) and adapted to the Chinese market by Chen et al. (2014), the investor sentiment index is constructed as the simple average of the standardized components. A higher value of the index indicates stronger investor optimism and more speculative market conditions.

Consistent with Hypothesis 1, the coefficient on investor sentiment (β₁) is expected to be positive, as elevated optimism may slow the incorporation of information into prices and increase price delay.

Information transparency is proxied by the annual disclosure rating of listed firms in the CSMAR database, which evaluates the completeness, timeliness, and credibility of corporate information disclosure. A higher transparency score represents a more open and reliable informational environment. According to H2, transparency is expected to moderate the sentiment–efficiency relationship, with the interaction term (${\beta }_3$) anticipated to be negative, indicating that transparency weakens the detrimental impact of investor sentiment on efficiency.

To address potential omitted-variable concerns, the model incorporates two control variables that account for broader market conditions affecting price formation. Market Volatility (Volatility) captures fluctuations in overall market risk by reflecting the variability in monthly index returns, while Trading Volume proxies for liquidity conditions and the intensity of investor participation. Including these measures helps ensure that the estimated relationship between sentiment and efficiency is not confounded by differences in market uncertainty or liquidity across indices and over time.

In addition to the main control variables, we include a set of macroeconomic control variables, namely the interest rate, inflation rate, and exchange rate, to account for broad economic conditions that may jointly influence investor sentiment and market efficiency. These controls help isolate the sentiment effect from concurrent macroeconomic fluctuations and enhance the robustness of the baseline specification.

Overall, the final model integrates index-level efficiency measures (R²) with market-level sentiment and transparency indicators in a fixed-effects panel framework, allowing for an examination of both the direct impact of sentiment and the moderating role of transparency under varying market conditions.

4. Results and Discussion

4.1. Descriptive Statistics

4.1.1. Summary Statistics of All Variables

The analysis of the relationship between investor sentiment, information transparency, and market efficiency is presented in Table 1, which reports the descriptive statistics for all variables used in the empirical analysis, including the sample size (N), mean, standard deviation (SD), minimum value (Min), median, and maximum value (Max).

The standard deviation of R² (Market Efficiency) is 0.150, and the mean value is 0.210, indicating a moderate degree of variation in price delay across the nine stock indices. The R² variable ranges from 0.000 to 0.844, with a median of 0.175, suggesting that while certain indices exhibit relatively high informational efficiency, others display stronger return predictability and more pronounced delays in price adjustment.

On average, the Sentiment Index is 0.051, reflecting a slightly positive investor mood in China’s A-share market over the sample period. However, the relatively large standard deviation (SD = 0.566) indicates substantial volatility in sentiment levels. The distribution spans from a minimum of −1.25 to a maximum of 2.043, with a median of −0.013, suggesting that investor sentiment fluctuated sharply between pessimistic and optimistic extremes across market cycles, which may have important implications for market efficiency.

The Transparency variable exhibits a mean value of 1.991 with a very small standard deviation (SD = 0.034), implying that the overall information disclosure environment is relatively stable over time. The values range narrowly between 1.921 and 2.045, with a median of 1.994, reflecting a homogeneous and slowly evolving disclosure regime in China’s A-share market.

With respect to control variables, market volatility (Volatility) has a mean of 0.197 and a standard deviation of 0.179, indicating noticeable variation in market risk conditions over time. Trading activity, measured by the logarithm of trading volume (LnTrading), has a mean of 15.449 and a standard deviation of 1.268, suggesting substantial dispersion in market liquidity and investor participation across periods.

Macroeconomic conditions are captured by inflation (CPI), interest rates (IRs), and exchange rates (REXs). CPI has a mean of 1.093 and ranges from −0.900 to 5.400, reflecting periods of both deflationary pressure and elevated inflation. The interest rate variable shows limited variation, with a mean of 3.070 and a range between 2.790 and 4.140, consistent with China’s relatively stable monetary policy environment during the sample period. The exchange rate (REX) has a mean of 6.580 and varies between 6.051 and 7.249, indicating moderate fluctuations in external financial conditions.

Overall, the descriptive statistics highlight a clear contrast between the high variability of investor sentiment and market conditions, and the relative stability of information transparency and monetary policy indicators. This variation provides a suitable empirical setting to examine how investor sentiment affects market efficiency and whether information transparency moderates this relationship.

Panel B Indices are classified into large-, mid-, and small-cap groups based on their underlying market capitalization characteristics. Large-cap indices include the SSE 50 (000016), CSI 300 (000300), CSI 300 Value (000951), and CSI 300 Growth (399919). Mid-cap indices include the CSI 500 (000802), CSI 500 Growth (399918), and the CSI Environmental Protection Index (399908). Small-cap indices include the CSI 1000 (000852) and CSI 1000 Growth (399915). Information transparency is measured using annual disclosure ratings from CSMAR, averaged across listed firms and years. Panel B reports mean transparency levels for each size group and is intended for descriptive comparison only.

Market Capitalization Group	Mean Transparency
Large-cap Indices	1.9810
Mid-cap Indices	1.9858
Small-cap Indices	1.9705
Overall Mean	1.9970

4.1.2. Correlations of All Variables

The correlation analysis provides preliminary insights into the relationships among the key variables. As reported in Table 2, the correlation between Market Efficiency (R²) and the Sentiment Index is close to zero (−0.004), indicating that investor sentiment does not exhibit a simple linear association with market efficiency at the aggregate level. This weak unconditional correlation suggests that the impact of sentiment on efficiency may be conditional on other market characteristics—such as the information environment—or may operate through nonlinear channels. This pattern motivates the subsequent regression analysis that explicitly tests the moderating role of information transparency.

Turning to the control variables capturing trading conditions, market volatility (Volatility) is positively and significantly correlated with both R² (−0.208) and the Sentiment Index (0.163). This finding is consistent with the view that sentiment-driven trading activity is more pronounced during volatile market periods, underscoring the importance of controlling for time-varying market risk when examining the sentiment–efficiency relationship. The logarithm of trading volume (LnTrading), by contrast, exhibits only a negligible correlation with R² (0.113), but shows a moderate and statistically significant positive correlation with investor sentiment (0.163) and volatility (0.149). This pattern suggests that heightened sentiment is typically accompanied by increased trading activity, even though trading intensity alone does not mechanically translate into greater price delay. Together, these results justify the inclusion of Volatility and LnTrading to account for risk and liquidity effects.

With respect to information transparency, the correlation matrix shows that Transparency is negatively and significantly associated with both market inefficiency (−0.575) and investor sentiment (−0.165). These relationships indicate that a more transparent information environment tends to coincide with faster price adjustment and lower sentiment-driven fluctuations, consistent with the notion that high-quality disclosure can dampen behavioral distortions in price formation.

The macroeconomic variables—exchange rate (REX), interest rate (IR), and inflation rate (CPI)—also display economically meaningful correlations with the main variables. Investor sentiment is negatively correlated with the exchange rate (−0.218) and positively correlated with interest rates (0.251), suggesting that external financial conditions and domestic monetary environments are closely linked to sentiment dynamics. Transparency exhibits significant correlations with all three macro variables, indicating that broader macro-financial conditions may co-move with the overall disclosure environment. Importantly, correlations between these macroeconomic controls and R² remain modest in magnitude, suggesting that they capture distinct channels through which macro conditions may influence market efficiency.

Overall, the correlations reported in Table 2 are moderate and well below conventional thresholds for multicollinearity concerns. This indicates that the variables represent different dimensions of market behavior, providing a suitable basis for multivariate regression analysis. These preliminary findings support the empirical strategy adopted in the subsequent sections.

4.2. Baseline Regression

Table 3. Baseline regression. Panel A reports variance inflation factors (VIFs) for the explanatory variables. All VIF values are below 2 (Mean VIF = 1.430), indicating no serious multicollinearity concerns. Panel B presents fixed-effects regression results, including index and time fixed effects. Volatility denotes market volatility, and LnTrading captures trading activity. REX, IR, and CPI are included as macroeconomic control variables. Robust t-statistics are reported in parentheses. *** and * denote statistical significance at the 1% and 10% levels, respectively.

Panel A: VIF Diagnostics
Variable	VIF	1/VIF
Sentiment Index	1.110	0.903
Transparency	1.480	0.676
Mean VIF	1.430
Panel B: Baseline Regression
	(1)
Variable	R2
Sentiment Index	5.983 ***
	(3.31)
Transparency Sentiment Index	−3.015 ***
	(−3.29)
Transparency	−0.040
	(−0.10)
Volatility	−0.122 *
	(−1.80)
LnTrading	−0.010
	(−0.56)
REX	0.054
	(1.21)
IR	−0.047 *
	(−1.82)
CPI	−0.001
	(−0.06)
Constant	0.219
	(0.23)
Observations	1317
Number of Indices	9
Adj. R-squared	0.200
Indices FE	YES
Year FE	YES

presents the baseline fixed-effects regression results examining the influence of investor sentiment and information transparency on market efficiency. The model incorporates both index and time fixed effects, and robust t-statistics are reported in parentheses. Multicollinearity was tested prior to estimation, with all variance inflation factors below 2 (Mean VIF = 1.430), well below the conventional threshold of 5, indicating that collinearity is not a concern. In addition to market volatility (Volatility) and trading activity (LnTrading), the specification further includes three macroeconomic controls—REX, IR, and CPI—to account for broader macro-financial conditions that may co-move with market efficiency.

The coefficient of the Sentiment Index is positive and statistically significant at the 1% level (β = 5.983, t = 3.31), indicating that higher investor sentiment is associated with greater price delay (higher R²), implying lower informational efficiency. This result supports H1, suggesting that sentiment-driven optimism amplifies speculative trading behavior and slows the incorporation of information into prices.

The interaction term Transparency × Sentiment Index is negative and statistically significant (β = −3.015, t = −3.29), indicating that higher transparency weakens the adverse impact of investor sentiment on market efficiency. In other words, the marginal effect of sentiment on R² decreases as transparency improves, suggesting that transparent information environments are less prone to sentiment-induced mispricing. This finding supports H2 and confirms that transparency plays a moderating role in stabilizing price discovery under sentiment-driven market conditions.

The coefficient of Transparency itself is negative but statistically insignificant (β = −0.040, t = −0.10). In the presence of an interaction term, the main effect of transparency reflects the conditional effect when investor sentiment equals zero and therefore has limited standalone economic interpretation. The primary focus remains on the interaction effect capturing transparency’s moderating influence.

Among the control variables, Volatility enters negatively and is weakly significant (β = −0.122, t = −1.80), suggesting that higher market volatility is associated with greater price delay. The coefficient of LnTrading is negative but statistically insignificant (β = −0.010, t = −0.56). Regarding the macroeconomic variables, the REX, IR, and CPI are not statistically significant in the baseline specification, indicating that the sentiment–transparency relationship is unlikely to be driven by concurrent macroeconomic fluctuations.

The adjusted R-squared of 0.200 indicates that the model explains approximately 20% of the variation in market efficiency within indices over time. Overall, these results demonstrate that investor sentiment significantly deteriorates market efficiency by increasing price delay, while information transparency mitigates such distortions.

4.3. Moderating Effect: The Role of Information Transparency

Table 4. The Moderating Effect of Information Transparency: Column (1) reports the full-sample fixed-effects regression including the interaction term between investor sentiment and information transparency. Columns (2) and (3) split the sample into high- and low-transparency regimes based on the annual median transparency score. All models include index and year fixed effects. Volatility and LnTrading are included as market condition controls, and REX, IR, and CPI are included as macroeconomic control variables. Robust t-statistics clustered at the index level are reported in parentheses. *** and **, denote statistical significance at the 1% and 5%, levels, respectively.

	(1)	(2)	(3)
	Full Sample	High Transparency	Low Transparency
VARIABLES	R2	R2	R2
Sentiment Index	0.059	16.425 ***	−0.630
	(0.13)	(4.09)	(−0.80)
Transparency × Sentiment Index	−0.031	−8.081 ***	0.316
	(−0.14)	(−4.10)	(0.79)
Transparency	−3.726 ***	−2.917	−4.775 ***
	(−4.35)	(−0.91)	(−7.38)
REX	0.069	0.190 **	−0.041
	(1.48)	(2.48)	(−0.76)
IR	−0.112 ***		−0.107 ***
	(−3.50)		(−3.44)
CPI	0.003	−0.004	0.003
	(0.62)	(−0.98)	(0.53)
Volatility	−0.055	−1.392 **	−0.084 **
	(−1.41)	(−2.39)	(−2.50)
LnTrading	0.018	−0.010	0.035 **
	(1.31)	(−0.45)	(2.57)
Constant	7.175 ***	5.180	9.760 ***
	(3.90)	(0.79)	(6.90)
Observations	1317	399	918
Adj. R-squared	0.150	0.179	0.135
Number of Indices	9	8	9
Indices FE	YES	YES	YES
Year FE	YES	YES	YES

reports the moderating effect of information transparency on the relationship between investor sentiment and market efficiency using three regression specifications. Column (1) presents the full-sample estimation including the interaction term, while Columns (2) and (3) split the sample into high- and low-transparency regimes based on the annual median transparency score. All models include index and year fixed effects, and robust t-statistics clustered at the index level are reported in parentheses.

In the full-sample regression (Column 1), the coefficient of the Sentiment Index is positive but statistically insignificant (β = 0.059, t = 0.13), suggesting that investor sentiment does not exhibit a strong unconditional effect on market efficiency in the aggregate specification. Similarly, the interaction term Transparency × Sentiment Index is negative but statistically insignificant (β = −0.031, t = −0.14), indicating that the moderating role of transparency is not clearly identified when the full sample is considered without distinguishing different transparency regimes.

Subsample results reveal clear heterogeneity across transparency regimes. In high-transparency markets (Column 2), the coefficient of sentiment is positive and strongly significant (β = 16.425, t = 4.09), suggesting that prices react more strongly to sentiment shocks when disclosure quality is high. At the same time, the interaction term remains negative and significant (β = −8.081, t = −4.10), implying that transparent environments facilitate faster correction of sentiment-induced mispricing.

In contrast, in low-transparency markets (Column 3), the sentiment coefficient becomes negative and statistically insignificant (β = −0.630, t = −0.80), and the interaction term is also statistically insignificant (β = 0.316, t = 0.79). This suggests that in opaque environments, sentiment-driven price movements are weaker and less systematically related to the information environment, indicating that transparency plays a limited moderating role when disclosure quality is low.

Taken together, these findings indicate that transparency does not simply dampen investor sentiment; rather, it reshapes how sentiment affects market efficiency. In more transparent markets, prices respond more strongly to sentiment shocks but also adjust more effectively, consistent with the “transparency as a stabilizer” mechanism proposed in H2.

4.4. Robustness Check: Nonlinear Effect of Investor Sentiment

To further validate the robustness of the baseline model, this section examines whether the effect of investor sentiment on market efficiency is nonlinear. Specifically, the squared term of the sentiment index is added to the baseline specification to test whether sentiment effects intensify at higher levels. Importantly, the dependent variable remains the price delay measure (R²), consistent with the main analysis.

Table 5. Robustness Check: Nonlinear Effect of Investor Sentiment. This table examines the nonlinear effect of investor sentiment on market efficiency by including the squared term of the sentiment index. The dependent variable remains the price delay measure (R²). Both index and year fixed effects are included, and robust standard errors are clustered at the index level. ***, denotes significance at 1% levels.

	(1)
VARIABLES	R2
Sentiment Index	0.062
	(0.14)
Sentiment Index2	0.007
	(1.20)
Transparency Sentiment Index	−0.032
	(−0.14)
Transparency	−3.738 ***
	(−4.37)
REX	0.070
	(1.50)
IR	−0.114 ***
	(−3.66)
CPI	0.003
	(0.65)
Volatility	−0.055
	(−1.39)
LnTrading	0.017
	(1.24)
Constant	7.214 ***
	(3.93)
Observations	1317
Number of Indices	9
Adj. R-squared	0.150
Year FE	YES
Indices FE	YES

reports the regression results from the nonlinear specification that incorporates the squared term of investor sentiment. The coefficient of the Sentiment Index is positive but statistically insignificant (β = 0.062, t = 0.14), suggesting that once the nonlinear component is included, the linear sentiment effect becomes weaker in this specification.

The quadratic term, Sentiment Index², is also positive but statistically insignificant (β = 0.007, t = 1.20), indicating that the nonlinear amplification effect of sentiment is not statistically supported in this model. These results suggest that although sentiment may exhibit nonlinear dynamics, the evidence for such an effect is limited in the current specification.

The interaction term between Transparency and the Sentiment Index is negative but statistically insignificant (β = −0.032, t = −0.14), implying that the moderating role of transparency is not clearly identified when the nonlinear specification is introduced.

The coefficient on Transparency remains negative and statistically significant (β = −3.738, t = −4.37), suggesting that higher information transparency is associated with lower price delay and improved market efficiency.

Among the control variables, IR remains significantly negative (β = −0.114, t = −3.66), while Volatility (β = −0.055, t = −1.39) and LnTrading (β = 0.017, t = 1.24) are statistically insignificant. These results indicate that the nonlinear sentiment specification does not materially change the role of market volatility or trading activity.

Overall, the nonlinear specification does not provide strong statistical evidence of a nonlinear sentiment effect. However, the negative and significant coefficient on transparency continues to support the efficiency-enhancing role of the information environment.

4.5. Causality Check: Instrumental Variable Estimation (2SLS)

To address potential endogeneity concerns between investor sentiment and market efficiency, the robustness test employs an instrumental variable approach using the lagged sentiment index (L.Sentiment_Index) as the instrument. This method mitigates simultaneity bias and captures the exogenous variation in sentiment that is not contemporaneously correlated with the error term. The results of the two-stage least squares (2SLS) estimation are presented in

Table 6. Instrumental Variable Estimation (2SLS) Using Lagged Sentiment Index. This table reports the two-stage least squares (2SLS) estimation results using the lagged sentiment index (L.Sentiment Index) as an instrumental variable. Both index and time fixed effects are included. Robust t-statistics in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

	(1) First Stage	(2) Second Stage
VARIABLES	Sentiment Index	R2
L.Sentiment Index	0.512 ***
	(4.26)
Sentiment Index		9.907 *
		(1.74)
Transparency × Sentiment Index		−4.992 *
		(−1.74)
Transparency		0.348
		(0.78)
REX	−0.251	0.078 *
	(−1.39)	(1.77)
IR	−0.080	−0.049 **
	(−1.25)	(−2.25)
CPI	0.009	0.003
	(0.24)	(0.29)
Constant	1.907	−0.86
	(1.64)	(−0.73)
Observations	1308	1308
Adj. R-squared	0.303	0.595
Indices FE	YES	YES
Year FE	YES	YES

.

In the first stage, the coefficient of the lagged sentiment index is positive and highly significant (β = 0.512, t = 4.26), confirming the strong explanatory power of the instrument for the current sentiment index. This result indicates that investor sentiment exhibits persistence over time and that its lagged value serves as a relevant instrumental variable.

In the second stage, the coefficient of Sentiment Index remains positive and weakly significant (β = 9.907, t = 1.74), suggesting that higher investor sentiment continues to increase price delay and reduce market efficiency after accounting for potential endogeneity.

More importantly, the interaction term Transparency × Sentiment Index is negative and weakly significant (β = −4.992, t = −1.74), indicating that higher information transparency weakens the positive association between investor sentiment and price delay. This finding implies that information transparency mitigates sentiment-driven inefficiencies when endogeneity concerns are addressed.

The coefficient of Transparency itself is positive but statistically insignificant (β = 0.348, t = 0.78), suggesting that transparency mainly operates through its moderating role rather than exerting a direct effect on market efficiency in this specification.

Among the macroeconomic controls, the REX enters positively and is marginally significant (β = 0.078, t = 1.77), while the IR is negative and statistically significant (β = −0.049, t = −2.25). CPI remains statistically insignificant. The model exhibits reasonable explanatory power, with an adjusted R-squared of 0.303 in the first stage and 0.595 in the second stage.

Overall, the 2SLS estimation provides additional evidence that the relationship between investor sentiment and market efficiency is not solely driven by simultaneity or omitted variable bias. The direction of the key coefficients remains consistent with the baseline results, and the negative interaction term continues to support the moderating role of information transparency.

4.6. Robustness Check: Subsample and Additional Controls

To further examine the robustness of the baseline results, the sample is divided according to market capitalization, and additional control variables capturing market conditions and macroeconomic environments are included. Specifically, market volatility (Volatility) and trading activity (LnTrading) are incorporated to control for market risk and liquidity conditions, while exchange rates (REXs), interest rates (IRs), and inflation (CPI) are added to capture broader macroeconomic influences.

As reported in

Table 7. Subsample with Additional Controls: Market Capitalization and Robustness Tests. The table reports the regression results of the subsample and additional control analyses. Columns (1) and (2) show the results for large-cap and small/mid-cap markets, respectively. The dependent variable is R², representing the price delay measure of market efficiency, where a higher value indicates lower informational efficiency. Sentiment Index denotes investor sentiment, and Volatility represents market volatility, LnTrading denotes intraday trading activity, REX denotes the exchange rate, IR denotes the interest rate, and CPI denotes inflation. Both index and time fixed effects are included in all specifications. Robust t-statistics in parentheses. **, and * denote significance at the 5%, and 10% levels, respectively.

	(1)	(2)
	Large-Cap Markets	Small/Mid-Cap Markets
VARIABLES	R2	R2
Sentiment Index	0.909	−0.228
	(2.48)	(−0.33)
Transparency Sentiment Index	−0.458	0.115
	(−2.47)	(0.33)
Transparency	−5.576 **	−2.355 **
	(−6.81)	(−3.11)
Volatility	−0.058 *	−0.052
	(−3.61)	(−0.91)
LnTrading	0.010	0.023
	(2.36)	(1.36)
REX	0.034	0.105
	(1.71)	(1.48)
IR	−0.142 *	−0.068
	(−4.94)	(−1.41)
CPI	0.006	0.001
	(0.76)	(0.22)
Constant	11.279 **	4.012 **
	(6.09)	(3.00)
Observations	506	802
Adj. R-squared	0.352	0.112
Number of Indices	3	6
Indices FE	YES	YES
Year FE	YES	YES

, the coefficient of the Sentiment Index remains positive and statistically significant in large-cap markets (β = 0.909, t = 2.48), indicating that stronger investor sentiment is associated with higher R² values—i.e., greater price delay and lower informational efficiency—consistent with H1. In contrast, the sentiment coefficient becomes statistically insignificant in small- and mid-cap markets (β = −0.228, t = −0.33), suggesting that the sentiment–efficiency relationship is primarily concentrated in large-cap market segments.

The interaction term Transparency × Sentiment Index remains negative and statistically significant in large-cap markets (β = −0.458, t = −2.47), indicating that higher information transparency weakens the adverse impact of investor sentiment on market efficiency. This result continues to support H2, suggesting that transparent information environments can mitigate sentiment-driven price distortions.

Regarding the control variables, Volatility is negative and statistically significant in large-cap markets (β = −0.058, t = −3.61), while LnTrading is positive and statistically significant (β = 0.010, t = 2.36), suggesting that market risk and trading intensity may influence price adjustment dynamics in large-cap indices. In contrast, these variables remain statistically insignificant in small- and mid-cap markets.

Among the macroeconomic controls, IR is negative and statistically significant in large-cap markets (β = −0.142, t = −4.94), while REX shows a weakly positive association with market efficiency (β = 0.034, t = 1.71). CPI remains statistically insignificant across both subsamples.

Overall, these results indicate that the sentiment–efficiency relationship is more pronounced in large-cap markets, where information environments and institutional participation are relatively stronger. Importantly, the negative interaction between transparency and investor sentiment remains robust in the large-cap subsample, reinforcing the conclusion that information transparency mitigates sentiment-driven inefficiencies in financial markets.

To further address potential concerns regarding reverse causality, we conduct a reverse causality test by regressing investor sentiment on lagged market efficiency (R²) and the same set of control variables used in the baseline specification. The null hypothesis is that past market efficiency does not influence current investor sentiment.

As reported in

Table 8. Reverse Causality Test: This table reports the results of a reverse causality test, where investor sentiment is regressed on lagged market efficiency (R²) and control variables. The null hypothesis is that past market efficiency does not influence current investor sentiment. The results show no significant effect of lagged market efficiency on sentiment, suggesting that reverse causality is unlikely to drive the baseline findings. Standard errors are clustered at the index level. *** p < 0.01, * p < 0.10.

	(1)
	Sentiment_Index
L. R2	−0.125 *
	(−1.96)
Transparency	−15.271 ***
	(−12.04)
Volatility	−0.051
	(−0.42)
LnTrading	0.271 ***
	−10.7
REX	0.103 ***
	−3.94
IR	−0.190 ***
	(−12.86)
CPI	−0.051 ***
	(−12.65)
Constant	25.779 ***
	−10.33
Indices FE	YES
Year FE	YES
Observations	1308
Adj. R-squared	0.544

, the coefficient of lagged market efficiency (L.R²) is negative and weakly significant (β = −0.125, t = −1.96). Although the coefficient reaches the 10% significance level, its magnitude is relatively small, suggesting that past market efficiency has only a limited influence on current investor sentiment.

Regarding the control variables, Transparency remains negative and highly significant (β = −15.271, t = −12.04), indicating that higher information transparency is associated with lower investor sentiment. LnTrading exhibits a positive and highly significant effect (β = 0.271, t = 10.70), suggesting that increased trading activity is associated with stronger investor sentiment. Volatility remains statistically insignificant (β = −0.051, t = −0.42), indicating that market risk does not significantly influence sentiment in this specification.

Among the macroeconomic variables, REX shows a positive and statistically significant relationship with investor sentiment (β = 0.103, t = 3.94), while IR (β = −0.190, t = −12.86) and CPI (β = −0.051, t = −12.65) are both negative and statistically significant.

Overall, although lagged market efficiency exhibits a weak association with investor sentiment, its economic magnitude is limited. These results suggest that reverse causality is unlikely to be the primary driver of the baseline findings, reinforcing the interpretation that investor sentiment primarily affects market efficiency rather than being driven by past market conditions.

5. Conclusions

This study investigates how investor sentiment and information transparency jointly shape the informational efficiency of China’s A-share market. First, investor sentiment is found to be negatively associated with market efficiency, as higher sentiment levels coincide with increased price delay and stronger return autocorrelation. This pattern suggests that optimism-driven trading is associated with slower incorporation of new information into prices, consistent with herding behavior and speculative overreaction. Second, information transparency appears to mitigate this adverse effect by improving the flow and credibility of information. The interaction between sentiment and transparency is consistently negative and significant, indicating that markets with better disclosure quality tend to be less vulnerable to sentiment-driven mispricing. In essence, transparency can function as a stabilizing mechanism within China’s capital market: it cannot eliminate emotional fluctuations entirely, but it buffers their disruptive impact on price discovery by anchoring investor expectations around more reliable information. Third, the relationship between sentiment and market efficiency remains robust across a range of validation exercises, including instrumental-variable estimation that addresses potential endogeneity concerns. Subsample analyses further suggest that sentiment-related inefficiency is more pronounced in small- and mid-cap indices, where information asymmetry and speculative participation tend to be higher. Additional controls for market volatility, liquidity, and trading volume also leave the main coefficients largely unchanged. Taken together, these results provide consistent evidence that investor sentiment is systematically related to informational efficiency in China’s equity market, with the strength of this relationship varying across market segments.

This study contributes to the literature by integrating insights from behavioral finance and information transparency in the context of China’s A-share market. Building on the price delay framework of Chordia et al. (2005), the analysis suggests that its explanatory power in emerging markets depends on transparency as a key institutional contingency shaping how investor sentiment translates into price efficiency. Rather than treating sentiment and transparency as separate forces, the findings indicate that transparency can act as a behavioral stabilizer by dampening the extent to which collective emotions distort price discovery. The results further suggest that sentiment-driven inefficiency is more pronounced in small- and mid-cap indices—where disclosure quality is lower and retail participation dominates—while large-cap markets, supported by analyst coverage and institutional oversight, exhibit greater resistance to emotional shocks. This distinction helps clarify how transparency defines the contextual limits within which behavioral biases influence price dynamics, offering additional perspective on inefficiency in emerging markets. Our findings also carry timely implications for ongoing regulatory efforts aimed at improving disclosure quality and market stability. From a policy perspective, the results suggest that a uniform disclosure regime may be insufficient in sentiment-sensitive markets. Instead, differentiated transparency requirements may be more effective for market segments that are particularly vulnerable to sentiment-driven mispricing.

Several limitations of this study should be acknowledged. First, the analysis is conducted at the index level using a limited number of major CSI indices. While index-level data help smooth firm-specific noise and capture aggregate market dynamics, this approach necessarily abstracts from heterogeneity across individual firms in terms of disclosure practices, governance structures, and trading behavior. Second, this study focuses exclusively on China’s A-share market. Although this setting provides a valuable laboratory for examining sentiment-driven trading in a retail-dominated environment, the findings should be interpreted with caution when extrapolating to other institutional contexts. Third, information transparency is measured using annual, aggregate disclosure ratings, which may not fully capture higher-frequency or firm-specific changes in the information environment. Finally, while the empirical strategy incorporates instrumental-variable estimation and extensive robustness checks, the results primarily speak to conditional relationships and moderating effects rather than establishing strong causal mechanisms.

These limitations also open several avenues for future research. First, extending the analysis to firm-level data would allow researchers to examine how heterogeneity in disclosure quality, corporate governance, and ownership structure shapes the interaction between investor sentiment and market efficiency. Second, cross-market or cross-country studies could assess whether the moderating role of information transparency observed in China’s A-share market generalizes to other emerging or transitioning financial systems. Third, future work could exploit regulatory events—such as the implementation of the Information Disclosure Management Measures in 2021—as quasi-natural experiments to provide sharper causal identification of transparency effects. Finally, incorporating additional market dimensions, including liquidity, volatility, or trading constraints, may help further illuminate the channels through which sentiment and information environments jointly influence price discovery.