The Predictive Power of Managerial Confidence: A Dynamic Mechanism of Attention and Reliability in China’s Stock Market

Abstract

Based on the “Future Outlook” sections of annual and semi-annual reports from Chinese A-share-listed companies (2011–2024), we construct a novel measure of managerial confidence by quantifying the intertemporal shifts in textual sentiment. Using a sample of 76,923 observations, our analysis reveals that this measure exhibits dynamic predictive power for expected stock returns. Specifically, in the short term, managerial confidence serves as a valid predictor. A long-short portfolio sorted by managerial confidence yields a 7.05% cumulative return spread over the five post-disclosure trading days. Mechanism analysis suggests that this short-term predictability stems from high managerial confidence effectively attracting investor attention. Over the medium term (six months), however, its predictive power hinges on the reliability of the confidence signal: For managers whose historical confidence has aligned with fundamental performance, high confidence predicts positive expected excess returns; for those who are chronically overoptimistic, it becomes an inverse predictor of firm value. These findings indicate that financial markets dynamically assess both the intensity and the reliability of signals within managerial disclosures, offering a new perspective on the predictive power of managerial psychological traits in capital markets.

1. Introduction

Recent advances in Natural Language Processing (NLP) have opened new avenues for measuring latent managerial traits from textual data, offering a powerful lens to examine their role in strategic decision-making and asset pricing. Among these traits, the most established area is managerial sentiment, in which scholars construct sentiment indices by quantifying the tone of corporate disclosures (e.g., earnings call transcripts [1], 10-Ks [2] sections, and financial disclosures [3]) and link these indices to market reactions and firm performance. However, this method suffers from inherent ambiguity: it fails to distinguish whether managerial sentiment stems from a retrospective view of past facts or a prospective outlook on future developments. This conflation has led to divergent empirical findings: some studies report only short-term market underreaction to managerial sentiment [4,5,6], while others document its persistent effects on returns and fundamentals [3,7,8]. We argue that a key source of this inconsistency lies in the heterogeneity of textual sources used to construct sentiment. Documents focused on recent events (e.g., earnings calls) may trigger transient market responses, whereas those inherently forward-looking (e.g., MD & A sections) are more likely to contain signals with longer-term pricing implications.

To address the ambiguity inherent in existing sentiment measures and accurately capture managerial forward-looking expectations, we use the “Future Outlook” section from the MD&A of annual and semi-annual reports issued by Chinese A-share-listed companies (2011–2024) as our textual corpus. This source is particularly valuable due to its mandatory nature and inherent forward-looking perspective. Extant research shows that the MD&A section contains incremental information not covered by other corporate disclosure channels, thereby aiding investors in better interpreting a company’s fundamental performance [9]. Building on this, research has leveraged MD&A texts to measure managerial traits, such as macro-cognition levels [10] and managerial myopia [11]. We define our core construct, managerial confidence, as the proportional change in sentiment derived from this section across consecutive disclosures. This measurement approach establishes a clear conceptual distinction. Managerial sentiment reflects the static, aggregate tone of a single disclosure, which often conflates retrospective and prospective elements. In contrast, we define managerial confidence as the dynamic change in sentiment within forward-looking statements. This measure reflects shifts in managers’ expectations regarding the future and effectively filters out managers’ fixed expressive habits or persistent optimistic/pessimistic tendencies [12], thereby providing a more precise and stable proxy for management’s genuine forward-looking beliefs.

Empirically, we examine the impact of managerial confidence on expected stock returns over two horizons: the short term (within two weeks post-disclosure) and the medium term (within six months post-disclosure). For the short-term analysis, following the classical asset pricing paradigm, we construct a long-short portfolio sorted by managerial confidence. The portfolio (value-weighted) generates statistically significant excess returns of 2.769%, 1.560%, 1.448%, 0.866%, and 0.234% over the first five trading days post-disclosure (cumulatively 7.05%) before becoming statistically insignificant in the second week. This pattern of return predictability remains robust after adjustment for multiple factor models. The results of Fama–MacBeth regressions that include multiple control variables also corroborate the predictive power of managerial confidence for expected stock returns. Further mechanism tests indicate that this short-term predictability is attributable to investor attention. Specifically, disclosures with higher managerial confidence effectively attract greater investor focus, which in turn drives the subsequent predictability of expected returns.

In the medium term, we find significant structural differences in the predictive power of managerial confidence. Based on the historical reliability of managerial confidence, we partition our sample into three groups: high optimism bias, consistent, and conservative. The empirical results show that for managers with recurrent high optimism bias, there is a significant negative correlation between their confidence and medium-term stock returns; for managers in the consistent group, their confidence serves as a reliable signal predicting rising stock prices; for conservative managers, the predictive power embedded in their confidence measure is relatively limited. These findings reveal that the medium-term predictive power of managerial confidence is heterogeneous: Its ability to forecast returns derives not only from the level of confidence itself but, more critically, depends on the degree of its alignment with fundamental performance, reflecting the market’s ongoing assessment of managerial signals.

In summary, by measuring managerial confidence, which serves as a proxy for the positivity of managers’ expectations about the future, we systematically reveal its predictive power in asset pricing. In the short term, it acts as a catalyst for attention, attracting investor focus that is associated with expected stock returns; in the medium term, it serves as a litmus test for reliability, distinguishing the quality of managerial signals through a fundamental verification mechanism. Overall, our study deepens the understanding of textual information from corporate disclosures, provides a new perspective on the predictive power of managerial psychological traits in capital markets.

2. Literature Review

2.1. Theoretical Underpinnings

This section aims to construct a multi-level theoretical framework to provide theoretical support for analyzing the impact of managerial confidence on expected stock returns.

2.1.1. Efficient Market Hypothesis and Asset Pricing Models

The Efficient Market Hypothesis (EMH), established by Fama as the theoretical cornerstone of modern finance, posits that security prices in a competitive and informationally efficient market fully and instantaneously reflect all available information [13]. Building on this premise, asset pricing models provide a quantitative framework for analyzing the fundamental risk-return trade-off. The Capital Asset Pricing Model (CAPM), independently developed by Sharpe [14], Lintner [15], and Mossin [16], formalizes the linear relationship between an asset’s expected return and its systematic (market) risk, implying that market beta is the sole determinant of its risk premium.

However, subsequent empirical research has identified persistent anomalies that challenge the CAPM’s explanatory power, notably the size effect [17] and the book-to-market effect [18]. In response, Fama and French significantly enhanced the model’s explanatory power by introducing size (SMB) and book-to-market (HML) factors, forming the three-factor model [19]. This theoretical lineage was further extended by Carhart with the addition of a momentum (MOM) factor [20], and by Fama and French with profitability (RMW) and investment (CMA) factors, culminating in the five-factor model [21]. Collectively, these models represent an evolutionary theoretical framework: an asset’s expected return is compensation for its exposure to various systematic risk factors, and any persistent excess return must be attributable to exposure to some unidentified risk factor.

Overall, the Efficient Market Hypothesis and asset pricing models provide the benchmark framework for rational pricing and a rigorous paradigm for empirical testing in this study.

2.1.2. Behavioral Finance Theory

Behavioral finance theory fundamentally challenges the perfectly rational agent assumption underpinning the traditional EMH. It demonstrates that individual investors are prone to systematic biases, such as overconfidence, leading to excessive trading and subpar returns [22]. When aggregated, such psychological biases can lead to market-wide under- and overreactions to information [23]. Moreover, because noise trader risk deters full arbitrage [24], the resulting mispricing can persist as systematic anomalies, rather than being swiftly corrected.

The theoretical foundation of behavioral finance is bounded rationality, which posits that decision-makers are constrained by cognitive limitations and finite information-processing capacity, thus unable to achieve the fully rational optimization assumed in traditional economics [25]. Within this paradigm, the allocation of attention emerges as a central mechanism for explaining market anomalies. When attentional resources are scarce, investors systematically prioritize the most salient information, leading to predictable patterns in stock returns [26]. This mechanism is well-supported empirically. For instance, Hirshleifer et al. find that during periods of high information concentration (such as when many firms announce earnings simultaneously) investors’ reactions to firm-specific news are diluted due to cognitive constraints [27]. Da et al. further demonstrate that investor attention, proxied by Google Search Volume Index, predicts short-term stock returns, offering direct evidence of attention-driven trading [28].

In this study, we posit that managerial confidence acts as a highly salient signal. According to the theory of limited attention, when managers use strong and unambiguous positive language in public disclosures, such signals are likely to penetrate investors’ attention barriers. Facing cognitive constraints, investors prioritize processing this conspicuous information, which can trigger concentrated trading and short-term price reactions. This mechanism will be empirically examined in the subsequent sections.

2.2. Measuring Managerial Psychological Traits

Given the inherently unobservable nature of managerial psychological traits, early research primarily relied on proxy variables for indirect measurement, inferring internal psychological states from observable external behaviors. These proxies included managerial option holdings [29], media-based measure [30], and executive compensation [31].

As research progressed, scholars recognized limitations of this proxy-based approach: Behavioral signals are susceptible to contamination by exogenous factors such as market conditions and corporate governance, raising concerns about their reliability and validity as measures of psychological traits [32]. A more critical limitation is the inherent difficulty in differentiating true psychological bias from rational decisions based on private information, thereby constraining their ability to explain managerial intrinsic motivation.

The development of computational linguistics and textual analysis have enabled researchers to move beyond indirect “behavior-outcome” proxies, allowing them to directly extract managerial psychological traits from public disclosures such as earnings call transcripts [1], 10-K filings [2], and financial statements [3]. A common approach within this paradigm employs lexicon-based methods to compute textual tone as the net difference between positive and negative word frequencies [33], and this net tone is often directly equated with managerial sentiment [34]. Early research in this area primarily relied on the Harvard-IV-4 dictionary, a widely adopted general-purpose lexicon that was not specifically designed for financial contexts. Some scholars also employed the accounting-oriented Diction lexicon [6] or alternative methodological approaches [7]. Loughran and McDonald made a seminal contribution by developing the Loughran-McDonald (LM) dictionary [2], which is specifically designed for the financial domain, and has gained widespread adoption in subsequent research.

In summary, positive and negative word counts are typically used to quantify psychological traits. The principal methodological divergence is the selected sentiment dictionary, whose quality is a critical determinant of measurement accuracy. Given the absence of an authoritative financial sentiment dictionary for Chinese text analysis, this study employs a dictionary recombination approach to construct a tailored Chinese financial sentiment lexicon. The specific methodology is detailed in Section 3.

2.3. Managerial Psychological Traits and Stock Returns

Existing literature proposes two competing pathways to explain how managerial psychological traits (commonly proxied by textual tone) influence stock returns. The fundamental divergence arises from opposing assumptions regarding market efficiency and investor cognition: the rational expectations paradigm presumes unbiased market interpretation of signals, whereas the behavioral perspective attributes mispricing to cognitive limitations that lead signals to be filtered through an emotional lens.

The first strand of research interprets managerial tone within a rational expectations framework, viewing it as a strategic signal rooted in managers’ superior private information about future prospects. This perspective posits that variations in textual tone reflect rational anticipations of the firm’s operating conditions [35] and cash flows [36]. Empirical studies aligned with this view typically document persistent return predictability. For example, Li constructs a managerial sentiment index from MD & A disclosures and finds its positive correlation with abnormal returns endures for at least twelve months [7]. Similarly, Feldman et al. show that shifts in managerial tone within 10-K filings predict returns from the filing date until the subsequent earnings announcement [8], and Jiang et al. demonstrate that sentiment extracted from 10-Ks and earnings calls forecasts returns over horizons extending to 36 months [3].

In contrast, scholars adopting a behavioral finance lens argue that investors exhibit bounded rationality, with decisions constrained by attentional and cognitive limits. From this viewpoint, sentiment-laden signals from managers (such as overt optimism in earnings calls) trigger emotional resonance, leading investors to rely on affective heuristics rather than rational analysis, thereby inducing predictable, short-lived pricing errors [4]. Corresponding empirical evidence consistently identifies transient effects. Henry finds that managerial positivity in earnings press releases correlates with abnormal returns only within a narrow three-day announcement window [6]. Loughran and McDonald report that the frequency of negative words in 10-K filings negatively predicts abnormal returns over the subsequent four trading days [2], while Jegadeesh and Wu show that sentiment derived from 10-K texts influences expected returns only over the following two weeks, with no effect beyond one month [4].

The tension between the documented “long-term information effects” and “short-term sentiment effects” reveals a significant theoretical integration gap. This divergence may stem from critical, yet underexplored, boundary conditions. First, the underlying mechanisms display temporal heterogeneity: the same psychological trait can simultaneously precipitate immediate, attention-driven trading and a slower, fundamentals-based price discovery. Second, the reliability of the signal sender (i.e., management) serves as a key moderator, as markets conditionally interpret signals based on perceived trustworthiness rather than accepting them at face value. To reconcile these competing narratives, our analysis advances a dynamic framework that traces a managerial signal’s complete lifecycle (from its release and market attention capture to its ex-post verification), thereby systematically explaining the conditional predictability of managerial psychological traits for expected stock returns.

3. Variable Definitions and Model Specification

3.1. Sample Selection

We use the Future Outlook sections from annual and semi-annual reports of Chinese A-share-listed companies for the period 2011–2024 as our research sample. The textual data are obtained from the Juchao Information Network. All corporate financial and market trading data are obtained from the China Stock Market and Accounting Research database (CSMAR).

To ensure robustness and consistency with standard empirical practices in the finance literature focusing on the Chinese market [37,38], we apply the following filters to the initial sample: First, we exclude ST and *ST companies to remove the influence of firms with abnormal financial conditions. Second, we remove financial firms to control for the unique impact of their high leverage. Third, we eliminate companies listed for under two years to avoid the potential new listing effect. Fourth, we exclude companies listed on the STAR Market. Finally, we exclude observations with missing values for key variables. This yields a final sample of 76,923 valid firm-period observations.

3.2. Measuring Managerial Confidence

This study employs textual analysis to measure managerial confidence from the Future Outlook section in the annual and semi-annual reports. Given the absence of an authoritative financial sentiment dictionary for Chinese textual analysis, we integrate four established lexicons to construct a tailored financial sentiment dictionary: the Dalian University of Technology Sentiment Lexicon (DUTIR Sentiment Lexicon), the CNKI Sentiment Analysis Lexicon, the National Taiwan University Sentiment Dictionary (NTUSD), and a Chinese translation of the LM dictionary. The detailed compilation process is outlined in Table 1.

Table 1. Dictionary Construction Process.

Category	Positive Words	Negative Words	Total
DUTIR Sentiment Lexicon	10,566	14,254	24,820
CNKI Sentiment Analysis Lexicon	4562	4360	8922
NTUSD	2598	7741	10,339
LM Dictionary (Chinese Version)	378	2287	2665
Combined	18,104	28,642	46,746
After Deduplication	11,521	15,876	27,397

Building upon the tailored dictionary, we employ Python 3.12 to perform word segmentation on the Future Outlook sections, identifying positive and negative terms within these sections. Table 2 presents the top 10 high-frequency terms identified within our sample.

Table 2. High-Frequency Vocabulary List.

Positive Vocabulary		Negative Vocabulary
Development	Development	Loss	Volatility
(556,326)	(386,511)	(592,615)	(295,617)
Enhancement	Increase	Challenge	Severity
(474,672)	(376,519)	(415,167)	(273,181)
Achievement	Improvement	Fall	Pressure
(424,819)	(321,987)	(376,316)	(253,162)
Innovation	Optimization	Limitation	Aggravation
(416,151)	(315,179)	(316,156)	(243,193)
Growth	Advancement	Cause	Suffer from
(401,615)	(307,916)	(305,619)	(231,579)

Note: Figures in parentheses show the term’s occurrence frequency in the full sample.

It is noteworthy that in our sample, the term “development” appears 556,326 times, while the term “four consecutive championships” appears only 3 times. This hardly suggests that the positive impact of “development” is hundreds of thousands of times greater than that of “four consecutive championships”. To address this, we adopt Loughran and McDonald’s methodology [2] and implement the Term Frequency–Inverse Document Frequency (TF-IDF) algorithm to weight word frequencies, thereby enhancing their discriminative capacity across categories.

$$Adjus{t\text{\_}TF}_{i,j,t}=[1+\log \left({TF}_{i,j,t}\right)]\mathrm{l}\mathrm{o}\mathrm{g}({\displaystyle \frac{N_t}{F_{i,t}}})$$

(1)

Here, TF_i,j,t denotes the frequency of term i in text j during period t, N_t represents the total number of texts in period t, and F_i,t represents the number of texts containing term i during period t.

The frequencies of all positive and negative words are then, respectively, aggregated as follows:

$${Positive}_{j,t}={\displaystyle \frac{\sum Adjus{t\text{\_}TF}_{P,j,t}}{1+\mathrm{l}\mathrm{o}\mathrm{g}\left({Total\text{\_}Word}_{j,t}\right)}}$$

(2)

$${Negative}_{j,t}={\displaystyle \frac{\sum Adjus{t\text{\_}TF}_{N,j,t}}{1+\mathrm{l}\mathrm{o}\mathrm{g}\left({Total\text{\_}Word}_{j,t}\right)}}$$

(3)

Equations (2) and (3) define the frequencies of positive and negative words in text j, respectively. The numerators represent the adjusted cumulative counts of these words. To mitigate the potential issue of mechanical volatility in proportional metrics that could arise from extremely small baseline values, the denominator is defined as the logarithm of the total word count of the text.

Following the standard paradigm in sentiment research [2,3,4], managerial sentiment is calculated as the difference in frequencies between the positive and negative words:

$${Sentiment}_{j,t}={Positive}_{j,t}-{Negative}_{j,t}$$

(4)

where j denotes a Future Outlook section text.

Finally, we calculate the proportional change in managerial sentiment between consecutive annual reports and define this measure as managerial confidence.

$$M{C}_{j,t}={\displaystyle \frac{Sentimen{t}_{j,t}-Sentimen{t}_{j,t-1}}{Sentimen{t}_{j,t-1}}}$$

(5)

To intuitively demonstrate the calculation process of managerial confidence, Table 3 presents a concrete example using Vanke A (stock code 000002) based on its 2024 interim and annual reports. As shown in the table, although the company exhibited positive sentiment in both reports (1.651 and 1.547, respectively), the measure of managerial confidence stands at −0.063, indicating a marginal decline in confidence from the interim to the annual report. This contrast highlights the key advantage of managerial confidence over static sentiment metrics: while the latter merely captures the tone at a single point in time, the former, by quantifying the change in sentiment, provides a more sensitive measure of shifts in management expectations.

Table 3. Calculation Example of Managerial Confidence.

Future Outlook Text Source	Positive Word Frequency	Negative Word Frequency	Total Word Count	Sentiment Value	Confidence Value
2024 Interim Report	305	286	37,472	1.651	—
2024 Annual Report	496	477	78,904	1.547	−0.063

3.3. Research Methodology

This section outlines the analytical framework and specific techniques employed in the study.

3.3.1. Portfolio Analysis and Related Factor Models

In empirical asset pricing research, portfolio analysis is a widely used statistical method, often employed to test the predictive power of one or more variables on expected stock returns. The most notable advantage of this method lies in its nonparametric nature: it does not require assumptions regarding the functional form of variable relationships or their cross-sectional distributions. By simply grouping and sorting, it can intuitively reveal the relationship between explanatory and explained variables. For example, Fama and French demonstrated the influence of market capitalization and book-to-market ratio on expected stock returns using just three groups [18].

Following the classical research paradigm, we first employ portfolio analysis to examine the predictive ability of managerial confidence on expected stock returns. The return spread between high- and low-confidence portfolios is then risk-adjusted using factor models to test whether this spread stems from systematic risk compensation. The specific factor models employed include the CAPM [14], the Fama-French three-factor model [19], the Carhart four-factor model [20], and the Fama-French five-factor model [21].

3.3.2. Fama-MacBeth Regression

We employ Fama–MacBeth regressions [39] to examine the predictive power of managerial confidence. The model specification follows Equation (6).

$${Ret}_{i,t+h}=\alpha +\beta {MC}_{i,t}+{\delta }^{T}Controls+{\epsilon }_{i,t+h}$$

(6)

where Ret_i,t+h is the excess return (raw stock return minus the risk-free rate) of stock i on the h-th day following the release of its Future Outlook information. The subscript h ranges from 1 to 5; that is, we test the impact of managerial confidence on expected stock returns within one week (five trading days) post-announcement. To examine whether a price reversal effect occurs after the initial adjustment, we additionally employ the excess returns of the second week post-announcement as a dependent variable.

Additionally, we include the following control variables: At the firm level, following Fama and French [18], we select contemporaneous financial indicators from the same annual report, including market capitalization (Size), book-to-market ratio (BM), leverage ratio (Lev), return on equity (ROE), and standardized unexpected earnings (SUE), to control for the impact of these concurrently disclosed financials on stock returns. Among market trading characteristics, following Jegadeesh and Titman [40], we select reversal effect (Rev), and momentum effect (MOM). Table 4 details the construction of the aforementioned variables.

Table 4. Variable Definitions and Measurement.

Variable	Measurement
Reti,t+h	Excess return of stock i on the h-th trading day after the release of the Future Outlook.
MC	Measure of managerial confidence, as defined in Section 3.2.
Size	Natural logarithm of market capitalization at the end of the fiscal year.
BM	Book value of equity divided by market value of equity at the fiscal year-end.
Lev	Total liabilities divided by total assets at the fiscal year-end.
ROE	Net income divided by shareholders’ equity at the fiscal year-end.
SUE	Earnings surprise standardized by forecast dispersion.
Rev	Cumulative stock return over the prior month.
MOM	Cumulative stock return over the prior 2 to 12 months.

Table 5 presents the descriptive statistics for the main variables used in this study. To mitigate the influence of extreme values, all continuous variables are winsorized at the 1% level. The mean MC of our sample is 0.093, indicating an overall positive managerial confidence in Chinese A-share market. Moreover, managers exhibit a growing tendency to use positive wording in their future outlook disclosures over time.

Table 5. Descriptive Statistics.

Variable	Mean	SD	Min	Med	Max	N
MC	0.093	0.066	−0.272	0.089	0.459	76,923
Ret1D (%)	0.644	3.171	−20.004	0.732	19.996	76,923
Ret2D (%)	0.337	2.537	−17.621	0.286	19.996	76,923
Ret3D (%)	0.186	1.728	−16.162	0.173	18.671	76,923
Ret4D (%)	0.206	1.484	−18.614	0.159	18.542	76,923
Ret5D (%)	−0.083	1.256	−17.651	−0.027	16.511	76,923
Ret2W (%)	0.061	6.516	−36.216	−0.108	45.376	76,923
Size	24.879	1.522	19.684	24.284	28.051	76,923
BM	0.660	0.157	0.159	0.656	1.156	76,923
Lev	0.487	0.211	0.027	0.466	0.991	76,923
ROE	4.227	4.542	−13.551	4.583	26.242	76,923
SUE	−0.018	2.162	−9.512	−0.056	10.201	76,923
Rev	0.277	12.221	−25.862	0.676	54.329	76,923
MOM	2.098	27.372	−38.856	−0.491	185.617	76,923

Regarding return patterns, the standard deviation peaks on the first trading day post-announcement and subsequently decays, which aligns with price discovery in response to new information. In terms of financial characteristics, the sample covers a wide spectrum of firm sizes (log market cap: 19.684 to 28.051), with an average leverage of 48.7%, an average ROE of 4.227%, and a near-zero average SUE (−0.018). As for trading characteristics, the large standard deviations of the reversal (Rev) and momentum (MOM) effects reflect significant heterogeneity in cross-sectional return patterns.

4. Baseline Analysis

4.1. Portfolio Analysis

Following the convention in asset pricing research, we employ portfolio analysis to initially investigate the relationship between managerial confidence and expected stock returns. Following each release of Future Outlook information, we sort all firms into five equally sized groups (quintiles) based on ascending managerial confidence.

Table 6 presents the characteristics of these sorted portfolios. MC exhibits a monotonic increase from portfolio P1 (0.028) to P5 (0.161). Conversely, none of the other variables show a discernible monotonic trend corresponding to this increase in MC. Thus, managerial confidence may constitute a relatively distinct dimension of information.

Table 6. Portfolio Characteristics Sorted by Managerial Confidence.

	P1	P2	P3	P4	P5
MC	0.028	0.059	0.091	0.127	0.161
Size	24.904	24.867	24.889	24.847	24.890
BM	0.641	0.634	0.705	0.682	0.638
Lev	0.513	0.476	0.479	0.498	0.468
ROE	4.307	4.152	4.127	4.319	4.228
SUE	−0.016	−0.018	−0.017	−0.015	−0.022
Rev	0.271	0.275	0.257	0.305	0.281
MOM	1.979	2.158	1.965	2.241	2.145

After sorting by managerial confidence, Table 7 presents the returns for the equal- and value-weighted portfolios. During the short window following information release (the first three days), the portfolio returns for both equal-weighted and value-weighted schemes exhibit a strictly monotonic increasing pattern as managerial confidence rises. Specifically, the first-day return of the equal-weighted portfolio spans from −1.119% (lowest confidence) to 1.766% (highest confidence), yielding a spread of more than 2.8 percentage points. However, this monotonic relationship gradually weakens over subsequent trading days and eventually disappears by the fifth trading day. Here, returns no longer follow any monotonic pattern across portfolio groups. The same phenomenon is similarly observed in the value-weighted portfolios.

Table 7. Portfolio Returns.

Panel A: Equal-Weighted Portfolio Returns
	P1	P2	P3	P4	P5
EW Ret1D	−1.119	0.312	0.886	1.376	1.766
EW Ret2D	−0.418	−0.103	0.384	0.671	1.152
EW Ret3D	−0.521	−0.111	0.239	0.402	0.921
EW Ret4D	−0.281	0.183	0.146	0.427	0.554
EW Ret5D	0.101	0.169	−0.421	−0.578	0.315
EW Ret2W	−0.039	0.213	0.346	−0.167	−0.048
Panel B: Value-Weighted Portfolio Returns
	P1	P2	P3	P4	P5
VW Ret1D	−1.006	0.286	0.924	1.221	1.763
VW Ret2D	−0.436	−0.111	0.397	0.692	1.124
VW Ret3D	−0.543	−0.122	0.251	0.414	0.905
VW Ret4D	−0.295	0.168	0.153	0.401	0.571
VW Ret5D	0.093	0.155	−0.398	−0.601	0.327
VW Ret2W	−0.142	0.196	0.221	−0.182	−0.051

The evidence presented above preliminarily indicates that managerial confidence possesses predictive power for short-term expected stock returns. After observing the monotonic pattern in portfolio returns, we perform a t-test on the return spread between the high and low portfolios (P5-P1) to assess its statistical significance and employ mainstream asset pricing models to examine whether this return spread can be explained by common risk factors.

Table 8 reports the excess returns adjusted by various asset pricing models, including the CAPM [14], the three-factor model [19], the four-factor model [20], and the five-factor model [21]. The results show that, irrespective of the weighting method, the high-low portfolio spread peaks on the first day following the information release, with both raw excess returns and risk-adjusted returns being statistically significant at the 1% level. Then, the excess returns exhibit a decay in both magnitude and statistical significance, becoming statistically insignificant in the second week.

Table 8. Tests on High-Low Portfolio Excess Returns.

Panel A: Risk-Adjusted Returns of Equal-Weighted Portfolios
	Excess Returns	CAPM Alpha	3-Factor Alpha	4-Factor Alpha	5-Factor Alpha
EW1D P5-P1	2.885 ***	2.371 ***	2.125 ***	2.087 ***	2.055 ***
	(6.74)	(6.86)	(5.45)	(5.95)	(5.62)
EW2D P5-P1	1.570 ***	1.273 ***	1.161 ***	1.047 ***	1.139 ***
	(5.39)	(5.35)	(5.35)	(5.41)	(4.75)
EW3D P5-P1	1.442 ***	1.298 ***	1.121 ***	1.050 ***	1.054 ***
	(4.56)	(4.62)	(3.91)	(3.82)	(3.56)
EW4D P5-P1	0.838 **	0.569 **	0.451 **	0.389 **	0.428 **
	(2.43)	(2.47)	(2.19)	(2.21)	(2.32)
EW5D P5-P1	0.214 **	0.102 **	0.084 *	0.056 *	0.067 *
	(2.31)	(2.18)	(1.91)	(1.87)	(1.84)
EW2W P5-P1	−0.009	−0.008	−0.007	−0.005	−0.008
	(−0.62)	(−0.58)	(−0.67)	(−0.53)	(−0.58)
Panel B: Risk-Adjusted Returns of Value-Weighted Portfolios
	Excess Returns	CAPM Alpha	3-Factor Alpha	4-Factor Alpha	5-Factor Alpha
VW1D P5-P1	2.769 ***	2.317 ***	2.041 ***	1.814 ***	1.935 ***
	(7.63)	(6.68)	(6.35)	(6.84)	(6.52)
VW2D P5-P1	1.560 ***	1.285 ***	1.092 ***	1.081 ***	0.968 ***
	(6.12)	(6.38)	(6.08)	(5.16)	(4.81)
VW3D P5-P1	1.448 ***	1.214 ***	1.063 ***	0.872 ***	0.937 ***
	(3.82)	(3.15)	(3.59)	(3.64)	(4.02)
VW4D P5-P1	0.866 ***	0.702 ***	0.518 **	0.382 **	0.428 **
	(2.99)	(2.86)	(2.43)	(2.22)	(2.14)
VW5D P5-P1	0.234 **	0.174 **	0.122 **	0.085 *	0.097 *
	(2.47)	(2.21)	(2.17)	(1.82)	(1.86)
VW2W P5-P1	0.091	0.087	0.088	0.087	0.089
	(1.07)	(0.88)	(0.93)	(0.87)	(0.71)

Note: *, **, *** denote statistical significance at the 10%, 5%, and 1% levels, respectively. t-statistics are reported in parentheses.

From an economic perspective, a long-short portfolio constructed based on managerial confidence generates significant cumulative excess returns over a five-day holding period: 7.12% for the equal-weighted portfolio and 7.05% for the value-weighted portfolio. Even after risk adjustment using a five-factor model, these excess returns remain at 4.822% and 4.432%, respectively. This suggests that the predictive power of managerial confidence for stock expected returns is not merely statistically significant but may also generate economic value for investors, providing preliminary validation for its potential as an effective indicator in investment decision-making.

Overall, the portfolio analyses provide preliminary evidence that managerial confidence can predict short-term stock returns. The predictive effect is most pronounced on the first trading day, gradually decays thereafter, retains only marginal significance by day five, and does not lead to a subsequent price reversal. Moreover, from a risk-adjusted perspective, while conventional risk factors partially explain the raw return spread between high- and low-confidence portfolios, the strategy still yields a significant positive Alpha after controlling for market risk (CAPM), size and value factors (three-factor model), momentum (four-factor model), and profitability and investment patterns (five-factor model). This demonstrates that the informational value embedded in managerial confidence is not fully captured by established risk factors. The significant Alpha likely stems from a systematic pricing bias in the market’s response to such “soft information”. This finding suggests that we can examine the predictive mechanism of managerial confidence on stock expected returns from a behavioral finance perspective.

4.2. Baseline Regression

Building on the evidence from portfolio analysis, we implement Fama–MacBeth regressions to further examine the predictive power of managerial confidence for short-term expected stock returns. The results are presented in Table 9.

Table 9. Fama–MacBeth Regressions of Managerial Confidence on Expected Stock Returns.

	(1)	(2)	(3)	(4)	(5)	(6)
	Ret1D	Ret2D	Ret3D	Ret4D	Ret5D	Ret2W
MC	0.326 ***	0.292 ***	0.164 ***	0.128 **	0.101 *	0.031
	(6.01)	(4.81)	(4.42)	(2.42)	(1.88)	(0.88)
Size	−0.011 **	−0.021 **	−0.019 **	−0.025 *	−0.026 *	−0.031 **
	(−2.24)	(−2.17)	(−2.24)	(−2.01)	(−1.92)	(−2.41)
BM	0.056 **	0.057 **	0.055 **	0.054 *	0.052 **	0.056 **
	(2.49)	(2.25)	(2.43)	(2.06)	(2.22)	(2.31)
Lev	−0.016 *	−0.021 **	−0.022 **	−0.019 **	−0.021 **	−0.024 **
	(−1.90)	(−2.23)	(−2.15)	(−2.29)	(−2.27)	(−2.36)
ROE	0.031 ***	0.012 **	0.025 ***	0.022 ***	0.027 ***	0.011 **
	(3.76)	(2.35)	(3.26)	(4.63)	(4.51)	(2.47)
SUE	0.621 ***	0.517 ***	0.493 ***	0.262 ***	0.110 ***	0.086 **
	(8.84)	(8.02)	(7.89)	(5.81)	(4.39)	(2.41)
Rev	−0.018	−0.032 *	−0.024 *	−0.022 **	−0.030 **	−0.017 *
	(−1.62)	(−1.88)	(−2.03)	(−2.16)	(−2.48)	(−1.76)
MOM	0.002	0.006	0.007	0.008	0.005	0.009
	(0.54)	(1.18)	(1.08)	(0.75)	(1.21)	(1.08)
Constant	0.024 ***	0.029 ***	0.019 ***	0.117 ***	0.155 ***	0.265 ***
	(5.19)	(5.12)	(4.29)	(3.68)	(4.10)	(5.76)
N	76,923	76,923	76,923	76,923	76,923	76,923
Adj.R2	0.198	0.177	0.171	0.153	0.133	0.110

Note: *, **, *** denote statistical significance at the 10%, 5%, and 1% levels, respectively. t-statistics based on Newey-West standard errors with two lags are reported in parentheses.

As shown in Column (1) of Table 9, on the first trading day post-announcement, the coefficient on managerial confidence (MC) for next-day stock returns is 0.326 and highly significant at the 1% level (t = 6.01). Specifically, a one-standard-deviation increase in managerial confidence (0.066) is associated with an expected rise in next-day returns of approximately 2.152 percentage points (0.326 × 0.066). This result indicates that, after controlling for size, valuation, profitability, and other factors, managerial confidence retains an independent and economically significant predictive power for short-term returns.

Subsequently, this effect decays rapidly: by the fifth trading day, the regression coefficient drops to 0.101 and is only marginally significant at the 10% level. By the second week (Ret_2W), the impact of managerial confidence on stock returns is no longer statistically significant. This pattern reflects the characteristic dynamics of how the market reacts to new information.

The results for the control variables also lend support to the validity of the model specification, as they align with asset pricing theories: The negative coefficient for Size reflects the small-firm effect; the positive coefficients for BM and ROE are consistent with the value and profitability premiums, respectively; the strong SUE effect is consistent with post-earnings-announcement drift.

In summary, managerial confidence effectively predicts short-term expected stock returns in the post-announcement period. This evidence underscores the unique value of managerial forward-looking textual information in capital markets.

4.3. Robustness Tests

4.3.1. Robustness to Pre-Disclosure Regulation

Since the Future Outlook section is sourced from annual reports, does the predictive power of managerial confidence for expected returns actually originate from concurrently disclosed financial information? This identification concern persists despite our baseline regressions controlling for major corporate financial metrics.

According to the China Securities Regulatory Commission (CSRC) regulations, listed companies have two pre-disclosure channels available before releasing their annual reports: earnings forecasts and preliminary earnings announcements. The earnings forecast is a mandatory disclosure mechanism for companies with abnormal operating performance, which primarily provides an estimate of the firm’s net profit for the preceding fiscal year. The preliminary earnings announcement is a voluntary disclosure that presents key financial data (e.g., Operating Revenue, Operating Profit, Net Profit) from the listed company’s preceding fiscal year. Compared to earnings forecasts, preliminary earnings announcements provide more comprehensive financial data.

This institutional framework provides an ideal quasi-natural experiment setting for our study: For firms that have issued pre-disclosures, investors already know the key financial data before reading the Future Outlook section. Then, the managerial confidence conveyed becomes a purified signal about the company’s future expectations, cleansed of historical financial noise. Accordingly, we partition the sample into two groups based on pre-disclosure status and conduct separate Fama–MacBeth regressions. Thus, we can test whether the predictive power of managerial confidence remains robust after accounting for the confounding effects of financial information.

As presented in Table 10, Panels A and B, respectively, divide the full sample into two groups based on the two pre-disclosure channels. Again, managerial confidence exerts a significantly positive effect on expected stock returns, regardless of whether the firm has issued a pre-disclosure. Specifically, the coefficient on managerial confidence is slightly lower in the pre-disclosure subsample than in the non-disclosure subsample. This suggests that while concurrent financial disclosures have a modest diluting effect, they do not undermine the positive predictive power of managerial confidence for expected returns, confirming the robustness of our baseline results.

Table 10. Robustness Tests for the Pre-disclosure Institution.

Panel A: Earnings Forecasts
	with						Without
	(1)	(2)	(3)	(4)	(5)	(6)	(1)	(2)	(3)	(4)	(5)	(6)
	Ret1D	Ret2D	Ret3D	Ret4D	Ret5D	Ret2W	Ret1D	Ret2D	Ret3D	Ret4D	Ret5D	Ret2W
MC	0.288 ***	0.231 ***	0.141 ***	0.105 **	0.081 *	0.027	0.363 ***	0.310 ***	0.198 ***	0.142 ***	0.119 **	0.034
	(5.52)	(4.76)	(3.87)	(2.31)	(1.91)	(0.84)	(6.57)	(5.62)	(4.89)	(3.78)	(2.34)	(0.91)
Control	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
N	41,521	41,521	41,521	41,521	41,521	41,521	35,402	35,402	35,402	35,402	35,402	35,402
Adj.R2	0.189	0.154	0.148	0.127	0.101	0.096	0.212	0.191	0.172	0.151	0.137	0.104
Panel B: Preliminary Earnings Announcements
	with						Without
	(1)	(2)	(3)	(4)	(5)	(6)	(1)	(2)	(3)	(4)	(5)	(6)
	Ret1D	Ret2D	Ret3D	Ret4D	Ret5D	Ret2W	Ret1D	Ret2D	Ret3D	Ret4D	Ret5D	Ret2W
MC	0.275 ***	0.224 ***	0.131 ***	0.108 **	0.040 *	0.017	0.353 ***	0.319 ***	0.211 ***	0.156 ***	0.091 **	0.019
	(5.51)	(4.17)	(3.43)	(2.21)	(1.86)	(0.51)	(6.72)	(5.51)	(4.18)	(3.27)	(2.41)	(0.51)
Control	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
N	28,547	28,547	28,547	28,547	28,547	28,547	48,376	48,376	48,376	48,376	48,376	48,376
Adj.R2	0.172	0.161	0.145	0.121	0.099	0.091	0.194	0.173	0.152	0.128	0.114	0.102

Note: *, **, *** denote statistical significance at the 10%, 5%, and 1% levels, respectively. t-statistics based on Newey-West standard errors with two lags are reported in parentheses.

4.3.2. Robustness Tests on the Managerial Confidence Measure

The managerial confidence we construct is, in essence, a quantitative proxy derived through textual analysis of the qualitative narratives in the Future Outlook section. For its measurement, we employ the TF-IDF method, which uses a penalized weighting scheme to calculate weighted scores for positive and negative words. This approach was introduced in NLP research by Salton and Buckley [41] and was subsequently refined by Loughran and McDonald [2], establishing it as a standard methodology in financial text analysis. The core mechanism leverages the inverse document frequency by penalizing common terms that appear ubiquitously across the corpus and amplifying less frequent yet concentrated keywords, thereby enhancing the discriminant power of text-based sentiment measures.

However, since the TF-IDF method artificially weights raw word frequencies, this technical choice may have potential implications for constructing the managerial confidence measure. To test the robustness of our managerial confidence measure construction and ensure that our findings do not rely on a specific weighting scheme, this subsection reconstructs the managerial confidence measure using unweighted raw term frequencies (MC_Alt) and repeats the baseline regression specified in Equation (5).

The results in Table 11 demonstrate the robustness of our core findings. Although the managerial confidence measure based on unweighted raw term frequencies yields attenuated coefficients with lower statistical significance, it remains significantly positive throughout the first five post-announcement days. This indicates that the predictive power of managerial confidence for short-term returns documented in our baseline analysis stems not from the specific algorithmic choice of TF-IDF weighting, but from the market’s reaction to the underlying sentiment in the Future Outlook text. Furthermore, the alternative measure’s impact exhibits a decaying pattern that is consistent with the baseline results, with its effect strongest on day one and fading thereafter. This consistent temporal pattern provides further validation for the robustness of our baseline findings.

Table 11. Robustness Tests for the managerial confidence measure.

	(1)	(2)	(3)	(4)	(5)	(6)
	Ret1D	Ret2D	Ret3D	Ret4D	Ret5D	Ret2W
MC_Alt	0.283 ***	0.227 ***	0.132 ***	0.107 **	0.084 *	0.021
	(5.45)	(4.49)	(3.88)	(2.34)	(1.86)	(0.90)
Control	Yes	Yes	Yes	Yes	Yes	Yes
N	76,923	76,923	76,923	76,923	76,923	76,923
Adj.R2	0.183	0.161	0.144	0.132	0.121	0.097

Note: *, **, *** denote statistical significance at the 10%, 5%, and 1% levels, respectively. t-statistics based on Newey-West standard errors with two lags are reported in parentheses.

In summary, the pre-disclosure regulation or different text processing techniques may affect the regression coefficients and statistical power of managerial confidence on short-term stock returns to some extent. However, they do not alter the direction and fundamental pattern of its impact. Managerial confidence exhibits robust predictability for short-term stock returns.

5. Mechanism Analysis

5.1. The Investor Attention Channel

As shown in the baseline analysis, managerial confidence predicts short-term expected returns. Its predictive power peaks on announcement day and monotonically fades over subsequent days. However, this raises a subsequent question: Through what mechanism does managerial confidence drive such prompt and orderly market price reactions in an informationally saturated market? Based on the limited attention theory, we posit that a potential mechanism lies in managerial confidence’s ability to effectively capture investor attention.

In actual market environments, investors cannot simultaneously process all available information. Instead, they selectively focus on signals that are salient and diagnostic [26]. A substantial literature finds that when investor attention, as a scarce resource, is asymmetrically attracted to certain assets, it increases asset prices in the short run and leads to predictable returns [27,28]. The high confidence conveyed by the management in the Future Outlook sections is inherently salient due to its positive affective tone and forward-looking nature. If we can demonstrate that managerial confidence elicits investor attention, then its predictive power for returns can be interpreted as follows: The attention triggered by confidence drives the market’s reassessment of the firm’s future prospects, leading to predictability in expected stock returns.

To test whether managerial confidence effectively attracts investor attention, we follow Da et al. [28] by using online search volume (Search) and stock forum page views (View) as proxies for investor attention. The data are sourced from the Chinese Research Data Services platform (CNRDS). To mitigate potential bias in regression estimates caused by the right-skewed distribution, the dependent variables are log-transformed. Consistent with the temporal window in Equation (6) and to mitigate potential endogeneity concerns, we lag the dependent variable by one period and specify the panel fixed-effects model in Equation (7) to test the impact of managerial confidence on investor attention beginning the day after information release.

$$Y_{i,t+h}=\alpha +\beta M{C}_{i,t}+{\delta }^{T}Controls+u_{t+h}+v_i+{\epsilon }_{i,t+h}$$

(7)

where Y_i,t+h represents the online search index (Search) and stock forum page views (View) for individual stock i on the h-th day following the release of its Future Outlook information, the subscript h ranges from 1 to 5. The control variables include market capitalization (Size), return on equity (ROE), Tobin’s Q (TobinQ), State-Owned Enterprise (SOE), listing age (ListAge), institutional ownership (Inst), and managerial ownership (Mshare), with u_t+h and v_i denoting time and firm fixed effects, respectively.

The results in Table 12 reveal the dynamic facilitating process of managerial confidence on investor attention. The results for online search volume show significantly positive coefficients for managerial confidence from day 1 to day 5 post-announcement, with a declining trend in magnitude from 0.381 (significant at the 1% level) to 0.045 (significant at the 10% level). This dynamic pattern is corroborated by the stock forum page views results, where coefficients decrease from 0.428 (significant at the 1% level) to 0.085 (significant at the 5% level) over the same period, indicating that managerial confidence effectively drives investor attention.

Table 12. Managerial Confidence and Investor Attention.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
	Search1D	Search2D	Search3D	Search4D	Search5D	View1D	View2D	View3D	View4D	View5D
MC	0.381 ***	0.272 ***	0.126 ***	0.105 **	0.045 *	0.428 ***	0.325 ***	0.261 ***	0.194 ***	0.085 **
	(7.39)	(5.24)	(4.08)	(2.44)	(1.91)	(8.17)	(6.45)	(5.12)	(3.87)	(2.43)
Size	0.156 ***	0.149 ***	0.135 ***	0.169 ***	0.153 ***	0.588 ***	0.579 ***	0.535 ***	0.569 ***	0.553 ***
	(8.69)	(7.78)	(8.35)	(6.63)	(6.63)	(9.24)	(8.51)	(9.18)	(7.45)	(7.32)
ROE	0.269 ***	0.228 ***	0.185 ***	0.113 **	0.107 **	0.322 ***	0.274 ***	0.222 ***	0.136 **	0.128 **
	(4.52)	(4.32)	(4.79)	(2.49)	(2.29)	(5.18)	(4.96)	(5.42)	(2.47)	(2.29)
TobinQ	0.093 ***	0.085 ***	0.089 ***	0.092 ***	0.087 ***	0.121 ***	0.125 ***	0.119 ***	0.122 ***	0.127 ***
	(3.91)	(3.84)	(3.75)	(3.89)	(3.79)	(4.35)	(4.28)	(4.17)	(4.32)	(4.25)
SOE	0.026	0.015	0.029	0.018	0.018	0.031	0.018	0.035	0.022	0.022
	(0.31)	(0.48)	(0.36)	(0.53)	(0.48)	(0.37)	(0.58)	(0.43)	(0.64)	(0.58)
ListAge	−0.051 **	−0.046 **	−0.038 **	−0.030 **	−0.030 **	−0.061 **	−0.066 **	−0.055 **	−0.063 **	−0.071 **
	(−2.27)	(−2.23)	(−2.31)	(−2.53)	(−2.49)	(−2.21)	(−2.37)	(−2.19)	(−2.46)	(−2.47)
Inst	0.114 ***	0.099 ***	0.063 **	0.078 **	0.051 **	0.137 ***	0.119 ***	0.128 ***	0.131 ***	0.151 ***
	(3.55)	(3.40)	(2.53)	(2.41)	(2.47)	(3.98)	(3.75)	(3.84)	(3.67)	(3.81)
Mshare	0.061	−0.031	−0.012	−0.033	−0.027	−0.073	−0.037	−0.014	−0.040	−0.032
	(0.12)	(−0.23)	(−1.01)	(−0.54)	(−0.54)	(−0.14)	(−0.28)	(−1.21)	(−0.65)	(−0.65)
Constant	0.056 ***	0.082 ***	0.106 ***	0.151 ***	0.198 ***	0.067 ***	0.098 ***	0.127 ***	0.181 ***	0.118 ***
	(4.56)	(8.12)	(4.90)	(5.42)	(4.37)	(5.47)	(9.74)	(5.88)	(6.51)	(5.25)
Year FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Firm FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
N	76,923	76,923	76,923	76,923	76,923	76,923	76,923	76,923	76,923	76,923
R2 (Within)	0.256	0.259	0.213	0.194	0.171	0.307	0.311	0.256	0.233	0.205

Note: *, **, *** denote statistical significance at the 10%, 5%, and 1% levels, respectively. t-statistics are reported in parentheses.

Importantly, the dynamic impact of managerial confidence on investor attention aligns closely with its impact on expected stock returns (Table 9), with both following an “immediate reaction-gradual convergence” pattern. These empirical results provide key evidence for the attention-based mechanism proposed earlier: High-confidence texts rapidly attract investors’ limited attention, motivating them to acquire additional information through searches and reading, which subsequently leads to a revision of their firm’s expectations and ultimately translates into stock price adjustments. Through its affective tone and forward-looking attributes, managerial confidence successfully acts as an attention anchor that guides investors to identify and process such signals within the information deluge, consequently resulting in return predictability.

5.2. The Role of Confidence Reliability

The portfolio analysis (Table 6, Table 7 and Table 8) and baseline regressions (Table 9) collectively establish that managerial confidence possesses significant short-term predictive power for expected stock returns, though this ability diminishes over time. This pattern reflects the market’s reaction to new information. Based on converging empirical evidence, we attribute it to an attention-driven effect. According to the Efficient Market Hypothesis, stock prices are ultimately driven by a firm’s fundamental performance [13]. Consistent with the Arbitrage Pricing Theory (APT) proposed by Ross, the no-arbitrage equilibrium condition in the capital market further implies that asset prices always tend to converge to their intrinsic values determined by fundamental risk factors and corporate operating performance, leaving no persistent risk-free arbitrage opportunities in an efficient market [42]. Managerial confidence represents an informed expectation of the firm’s future operating conditions, derived from managers’ informational advantage that includes private knowledge. If this expectation is accurate and reliable, the stock price should exhibit sustained growth over the medium term as expectations materialize. Conversely, if elevated confidence reflects mere hollow optimism disconnected from subsequent performance, the short-term price increase it drives, without fundamental performance support, will eventually reverse as prices converge to their fundamental value [13]. We define returns within the six-month period following information disclosure as medium-term returns. The selection of this window is based on two primary considerations. First, its duration is sufficient to effectively filter out short-term market noise caused by fluctuations in market sentiment and liquidity. Second, this interval roughly aligns with the reporting cycle between annual and semi-annual reports of Chinese listed companies, thus providing a natural window through which we can observe the market’s response across successive information releases. Accordingly, we hypothesize that the impact of managerial confidence on medium-term stock returns is inherently contingent upon the confidence’s reliability. To test this, we adapt Rogers and Stocken’s approach [43] by categorizing firms into subsamples based on managerial confidence’s historical reliability. We then perform subgroup regression analyses to examine the differential impact of confidence on medium-term returns across these groups.

Specifically, we construct a measure for managerial confidence’s historical reliability (MC_HR) over the past three years. This measure is calculated as the average annual discrepancy over this period between a firm’s within-industry percentile rank in managerial confidence and its within-industry percentile rank in ROE for the subsequent year, as shown in Equation (8).

$${MC\text{\_}HR}_{i,t}={\displaystyle \frac{1}{3}}\sum _{k=1}^3{(MC\text{\_}Rank}_{i,t-k}-{ROE\text{\_}Rank}_{i,t+1-k})$$

(8)

where ${MC\text{\_}Rank}_{i,t-k}$ denotes the within-industry percentile rank of firm i’s managerial confidence from prior disclosures (t−3, t−2, t−1) and ${ROE\text{\_}Rank}_{i,t+1-k}$ represents the within-industry percentile rank of its subsequent realized ROE (t−2, t−1, t) for the corresponding periods. This lagged pairing ensures market expectations at time t cannot retroactively affect historical performance or past confidence signals, eliminating endogeneity concerns.

This approach aims to evaluate the historical reliability of managers by quantifying the degree to which their past confidence has aligned with subsequent fundamental performance. Using this measure, we categorize the sample into three groups. Firms with a significantly positive score, whereby confidence persistently exceeds realized performance, are assigned to the High Optimism Bias group. Those with scores near zero, indicating alignment between confidence and performance, form the Consistent group. Finally, firms with significantly negative scores, where performance consistently outperforms prior confidence, constitute the Conservative group.

To reasonably set the grouping threshold, we analyzed the statistical characteristics of MC_HR. Table 13 presents the descriptive statistical results. The data show that the mean (−0.009) and median (−0.013) of MC_HR are relatively close, and the skewness (0.077) is near zero, indicating that its distribution is roughly symmetric with no significant skewness. The kurtosis (2.650) is lower than the normal distribution value of 3, suggesting a platykurtic distribution. The standard deviation is 0.312, reflecting a certain degree of heterogeneity among the sample companies in terms of the historical reliability of management confidence.

Table 13. Descriptive Statistics for MC_HR.

Mean	SD	Min	Q1	Med	Q3	Max	Skewness	Kurtosis	N
−0.009	0.312	−0.929	−0.224	−0.013	0.201	0.945	0.077	2.650	71,498

Specifically, the minimum value of −0.929 corresponds to the case of North China Pharmaceutical in 2015 (stock code 600812). The company’s management confidence rankings for the past three years were 1/133, 1/133, and 3/144, while its ROE rankings for the same period were 125/133, 124/133, and 137/144, demonstrating a clear optimistic bias. The maximum value of 0.945 is observed for Weiming Environmental Protection (stock code 603568), where the management confidence rankings for the past three years were 36/36, 34/36, and 37/37, while the corresponding ROE rankings were 1/36, 1/36, and 2/37, indicating a significant conservative bias.

Based on the statistical results, we established grouping thresholds by taking the median value (−0.013) as the benchmark and defining a fluctuation range of half a standard deviation (0.156 = 0.312/2) above and below this central point. Specifically, the classification criteria are as follows: firms with MC_HR values lower than the median minus half a standard deviation (i.e., below −0.169) are assigned to the High Optimism Bias Group; those with MC_HR values falling between the median plus or minus half a standard deviation (i.e., from −0.169 to 0.143) are categorized into the Consistent Group; and firms with MC_HR values exceeding the median plus half a standard deviation (i.e., above 0.143) are placed in the Conservative Group.

Then, we conduct Fama–MacBeth regressions as specified in Equation (6) for each subsample. The results in Table 14 provide strong support for our hypothesis. In the High Optimism Bias group, managerial confidence exhibits a significant negative relationship with medium-term returns. This finding can be explained through two plausible channels: First, the systematic failure of subsequent performance to justify elevated confidence levels may trigger a value convergence mechanism [13]; second, such optimism may potentially lead to value-destroying corporate behaviors, including overinvestment and undisciplined expansion [44].

Table 14. Group-based Tests on Managerial Confidence and Medium -Term Stock Returns.

	High Optimism Bias	Consistent	Conservative
	Ret6M	Ret6M	Ret6M
MC	−1.07 ***	2.011 ***	0.321 **
	(−4.52)	(7.42)	(2.54)
Control	Yes	Yes	Yes
N	22,058	27,382	22,058
Adj.R2	0.191	0.283	0.163

Note: **, *** denote statistical significance at the 5% and 1% levels, respectively. t-statistics based on Newey-West standard errors with two lags are reported in parentheses.

For the Consistent group, managerial confidence demonstrates significant positive predictive power for medium-term returns (coefficients of 2.011 and significant at the 1% level). This pattern supports the interpretation that historically validated confidence can serve as a reliable forward-looking signal. The observed high confidence may reflect improving fundamental prospects, and as subsequent positive information is disclosed, market participants progressively verify management’s forecasting accuracy, establishing positive price feedback.

For the Conservative group, managerial confidence also positively predicts medium-term returns, though the effect is more subdued. This implies that while the signal remains informative, its predictive power is constrained by the cautious nature of these managers, who consistently understate future performance.

As a robustness check, we estimated the regressions using the first and third quartiles as grouping thresholds. The results, presented in Table 15, show that the significance and direction of the managerial confidence effect remain consistent with the earlier findings based on median and standard deviation groupings, indicating that the observed heterogeneity in its predictive power across subgroups is not driven by the specific choice of thresholds.

Table 15. Robustness Check with Quartile-Based Grouping.

	High Optimism Bias	Consistent	Conservative
	Ret6M	Ret6M	Ret6M
MC	−1.169 ***	1.914 ***	0.243 **
	(−5.24)	(7.09)	(2.49)
Control	Yes	Yes	Yes
N	17,875	35,748	17,875
Adj.R2	0.217	0.272	0.155

Note: **, *** denote statistical significance at the 5% and 1% levels, respectively. t-statistics based on Newey-West standard errors with two lags are reported in parentheses.

Overall, in the medium run, capital markets do not simply react to managerial confidence at face value. Instead, they discriminate its informational value and dynamically incorporate it into prices. This nuanced market behavior clarifies the boundary conditions of managerial confidence’s predictive power and underscores the importance of evaluating such signals through the lens of their historical reliability.

6. Conclusions

6.1. Main Findings and Research Implications

Based on the changes in sentiment expressed by managers in the Future Outlook sections, we construct a measure of managerial confidence and systematically examine its ability to predict expected stock returns. The results show that managerial confidence exhibits significant predictive power for stock returns, with notable heterogeneity in its underlying mechanisms and effect persistence.

In the short term, managerial confidence is a significant predictor of stock returns in the brief post-announcement window (up to five trading days). Mechanism analysis suggests that this short-term predictability is driven by investor attention, as higher managerial confidence effectively attracts greater market focus, which in turn drives the subsequent predictability of expected returns.

Over the medium run, the predictive power of managerial confidence exhibits significant heterogeneity. For managers with a consistent historical alignment between their confidence and fundamental performance (the consistent group), high confidence can reliably predict sustained medium-term price appreciation. Conversely, for managers with a persistent history of excessive optimism (the high optimism bias group), higher confidence instead has a negative predictive effect on medium-term returns. This suggests that the market does not merely respond to managerial optimism but dynamically evaluates and distinguishes the substantive value of their statements.

Based on our findings, we suggest that investors pay attention to the consistency between a firm’s historical confidence and actual performance when evaluating management’s forward-looking statements. For listed firms, we recommend adopting a cautiously optimistic communication approach that balances positive expectations with achievable performance to avoid damaging long-term corporate value. For regulatory bodies, it is advisable to improve disclosure frameworks, guiding companies toward more substantive and verifiable forward-looking disclosures in order to enhance disclosure transparency.

6.2. Limitations and Future Research

This study has several limitations, which delineate the scope of applicability of the conclusions and offer some tentative avenues for future research.

First is the endogeneity issue in textual disclosure. Our measure of managerial confidence is derived from the text of the Future Outlook sections. It is important to note, however, that these forward-looking statements may still be influenced by concurrent financial conditions, the business cycle, and the broader market environment. To address endogeneity concerns, we controlled for key concurrent financial variables (e.g., ROE and leverage) in our empirical analysis and conducted robustness tests based on the pre-disclosure system. Nevertheless, potential endogeneity issues may persist. Future research could use exogenous policy shocks or more refined text decomposition techniques to identify managerial confidence more accurately.

Second is the external validity of the findings. This study draws on data from China’s A-share market covering the period 2011–2024. Caution is warranted when extending these findings to other markets, such as mature capital markets or those operating under different information disclosure regimes. The distinct characteristics of China’s capital market, including its investor composition, regulatory framework for information disclosure, and corporate governance practices, may moderate the relationship between managerial confidence and market reactions. Consequently, the generalizability of this study’s findings requires further examination across varied institutional settings.

Third comprises the proxy variables for investor attention. In examining the mechanism of investor attention, we used online search volume and forum view counts as proxies for attention. While these indicators are commonly adopted in existing literature, they may still be affected by other concurrent information events. Future research could employ more refined experimental designs or more direct behavioral data, such as eye-tracking or institutional research records, to improve the credibility of the inferred mechanism.