Do emotions matter? The impact of managerial emotion profiles on stock returns

Humans are not only rational beings but also possess a profound capacity for emotion. Emotions are discrete, intense, but temporary experiences elicited by a particular stimulus that often include action sequences (Repenning et al., 2022). Emotions can be helpful, harmful, influence behaviors or guide decisions. An emotion profile is a tool for understanding the complex landscape of a blend of emotions in individuals or groups. The psychology literature posits that individual emotion profiles have a fundamental effect on observers (Van Kleef and Côté, 2021). Drawing insights from these social implications of emotions, we examine how investors react to emotion profiles portrayed during earnings calls.

Conceptually, emotions are different from other affective states, such as moods (e.g. Johnson et al., 2016), sentiments (e.g. Ge et al., 2019) or attitudes (e.g. Nolder and Kadous, 2018). We differentiate our study from prior research examining the effect of sentiment, conveyed in conference calls, on investor reactions (Allee and DeAngelis, 2015; Davis et al., 2015). Sentiment analysis primarily focuses on determining the overall tone of language or opinion expressed in a text and categorizes them as negative, positive and neutral. In contrast, emotion detection delves into a deeper range of emotions expressed in text such as anger, sad, fear, joy and surprise. We also differentiate from prior work extracting managerial emotions or affective states from vocal cues (Mayew and Venkatachalam, 2012) and from facial expressions (Momtaz, 2021). Instead, our approach of detecting emotions, from the earnings call transcripts, is more intricate and novel as it seeks to determine subtleties within emotional expression. Importantly, we draw from scholarly consensus and from theoretical insights, which build around two social consequences of emotions. First, emotion profiles entail affective reactions (Sutton, 1991). Supporting this notion, CEO affective states can trigger emotional contagion among investors (Momtaz, 2021). For example, managers conveying negative emotions may cause the same negative emotions among investors. Second, observed emotions may be taken as cues of deeper motives (Van Kleef et al., 2004) and can be interpreted using inferential processes. We postulate that emotions serve as signals, offering insights into firm performance, future earnings prospects, risks and uncertainties, thereby facilitating investors’ judgment.

We extract emotions for the CEO and for Other Managers participating at the earnings conference call by using natural language processing (NLP) techniques, following Gamage et al. (2024). We use a validated psychological model (Plutchik, 1980), consisting of eight primary human emotions: anger, fear, sadness, disgust, joy, anticipation, trust and surprise, which are universal to all cultures, populations, ages and genders. We use a machine learning (ML) algorithm to classify text, from the earnings call transcripts, into one or more emotion categories. Using a sample from the USA from 2010 to 2020, we find that the emotion profiles are significantly different between the conference participants and between the presentation and Q&A sections of the conference call.

Our results support our assertion that emotion profiles are significantly associated with stock returns. Specifically, investors react favorably (unfavorably) to the positive (negative) emotion profile exhibited by the firm managers at conference calls, supporting the notion of social effect of emotions. Earnings news serves as a contextual signal that can shape how investors interpret emotional expressions during earnings calls. Our findings reveal that investors respond negatively when emotional expressions by CEOs or analysts are incongruent with earnings news – suggesting that such emotional signals may amplify or undermine the credibility of financial disclosures.

Our paper offers several contributions. First, we contribute to the academic literature by examining how manager emotion profiles impact investor behavior. Much of the early work in this area focused on the influence of demographic or other observable executive characteristics (e.g. personality traits, narcissism and extraversion) or sentiment (i.e. positive, negative or neutral) on firm outcomes. However, the influence of manager emotions on stock returns remains relatively unexplored. Emotional profiles may reveal new insights about managerial confidence or uncertainties that are not immediately apparent in the content of the disclosure. Drawing on the nature of the earnings conference calls, we use a robust emotion dictionary to better understand how emotions portrayed by managers influence receivers’ behaviors. Second, we contribute to the social effects of emotions literature. Research in psychology suggests that individual’s emotions are manifest through their words and actions, which are observed by outside actors who then use those observations to make judgments. Given the managers’ prominent role in the firm, how their emotions present to external market actors should not only affect their perceptions of the managers but should also inform the market’s view of the firm. We use a novel theoretical insight to support our assertion that emotions influence investor behavior through two primary mechanisms: inferential processes and affective reactions (Van Kleef, 2009). In the conference call setting, inferential processes allow investors to interpret shifts emotions as information cues, while affective reactions allow investors to mirror the emotions conveyed by managers. Finally, we use NLP techniques and ML methods to extract emotions from the earnings conference call transcripts. Frankel et al. (2022) finds that measures based on ML offer a significant improvement in explanatory power over dictionary-based measures. Specifically, adopting ML methods for emotion analysis could provide more accurate, dynamic and predictive insights into how emotions affect stock returns.

Earnings conference calls are an important disclosure mechanism that convey both quantitative results and qualitative cues to the market. These calls provide an interactive and unconstrained platform for managers to explain financial results and engage with analysts (Bushee et al., 2003; Matsumoto et al., 2011; Papadopoulou et al., 2024). The dual structure of these calls, consisting of a management presentation and a Q&A session, creates opportunities for firm-specific characteristics, including managerial communication styles and behavioral cues, to influence market perceptions.

One stream of literature on emotions examines how individual emotions affect their own behavior such as risk taking and strategic decision-making (Cascón‐Pereira and Hallier, 2012; Delgado‐García et al., 2010). Studies in accounting and finance show that features such as disclosure tone (Price et al., 2012; Davis et al., 2015), affective signals (Mayew and Venkatachalam, 2012) and linguistic patterns indicating deception (Larcker and Zakolyukina, 2012) can significantly affect investor responses. While this literature has largely focused on tone, fewer studies have examined the role of discrete emotional expressions. Using facial expressions to capture emotions, Akansu et al. (2017) find that fear-based emotions lead to short-term changes in the firm’ equity price, whereas anger-, disgust- and happiness-based emotions explain long-term (post-announcement period) firm performance. Momtaz (2021) find that investors use CEO affects, measured by facial muscle contraction-relaxation patterns, as signals about firm value. Chen et al. (2023) extract emotional information from the audio of earnings calls and find that positive emotions influence positively the analysts’ willingness to issue rating reports.

We differentiate our study from prior research examining the impact of sentiment (or tone) on stock returns. Compared with tone, emotions encompass a broader range of nuanced experiences (Gordon, 2017) and can provide deeper insights into how motivations translate into actions (Tamir, 2016). Tone of language or sentiment can be predominantly negative or positive and can be used to express a general state of polarity (certainty vs doubts, agreement vs disagreement) in relation to a subject. Hence, the nuances or spectrum of emotions do not count. Emotion, on the other hand, is an intense affective state that prevents someone from reacting reasonably and appropriately to the event that provoked it. Emotions, therefore, involve a spectrum of human emotions (i.e. anger, sad, fear, joy and surprise).

The focus of our study is to examine how manager emotion profiles affect investor reactions. Given that firm managers represent the face of their companies to outside observers, especially for large public firms, and that their unique patterns in how they speak, and act will be visible to outside actors, we propose that part of this assessment will focus on the behavioral patterns and other tendencies stemming from manager emotions. We theorize that investors will use the managers’ verbal and behavioral cues to form judgments of them, and, as such, these cues serve as proxies for the managerial underlying emotions.

We draw on the psychology literature to develop our hypotheses based on the notion of social effects of emotions. Prior research suggests that individuals derive information from others’ emotions through different processes (Elfenbein, 2007; Van Kleef et al., 2004; Hareli and Hess, 2010). These processes essentially involve the receivers’ ability to extract information from others’ emotional expressions about a particular situation. Emotions as social information (EASI) theory postulates that emotional displays help to clarify social situations and promote interpersonal coordination (Van Kleef, 2009, 2016). In turn, the observers’ behavioral responses are shaped by affective reactions and inferential processes. ^[1] Based on EASI theory, we argue that investors, as observers of the earnings calls, draw inferences from the manager emotions during the earnings calls and subsequently engage in an investment behavior, using emotions as information cues.

Emotions can be categorized into two dimensions: positive and negative. Positive and negative emotions provide different information about the firm’s earnings and as a result, the investors react differently. We posit that joy, trust, surprise and anticipation, i.e. positive emotion profiles, have a positive social effect and as a result investors react positively. In contrast, negative emotions have a negative social effect. We posit that negative emotion profiles such as sadness, fear, disgust and anger are likely to have negative reactions from investors.

EASI theory is applicable to emotions expressed verbally, in text, facial or in any form. Prior research provides evidence that observers draw inferences from others’ expression of emotions and change their behavior accordingly (Van Kleef and Côté, 2021). We take the view that manager emotions influence abnormal returns because emotions have been considered an important predictor of observers’ behavior. The psychology literature suggests the notion of a blend of multiple emotions which then make up emotion profiles (Fernando et al., 2014; Wortha et al., 2019). These emotion profiles reflect co-occurrence of multiple emotions and posit considering the occurrence of multiple emotions simultaneously. We follow this assertion of co-occurrences of multiple emotions and assert that positive emotion profile consists of co-occurrence and blend of joy, trust, surprise and anticipation, while the negative emotion profile consists of co-occurrence and blend of sadness, fear, disgust and anger.

Based on this discussion, we formulate the following hypothesis:

The core premise of our study is that emotions profiles of call participants influence investor reactions, as emotions serve as social information cues. According to EASI theory, the effect of such emotional cues on observers is context dependent. We argue that earnings surprises constitute a salient contextual factor that can moderate the relation between emotion profiles and stock returns. When earnings surprises are large and positive (or negative), investors receive a strong informational signal regarding firm performance. In such cases, reliance on emotional cues may diminish, as the market already has a clear basis for valuation. Conversely, when earnings surprises are small or ambiguous, emotional cues may play a more prominent role in shaping investor interpretation. Building on these insights, we contend that the association between managerial emotions and market reactions is conditional on the earnings news conveyed during the call. Accordingly, we propose the following hypothesis:

We obtain a sample of 17,226 conference call transcripts from EIKON Refinitiv Thomson Reuters for S&P 500 firms during the period 2010–2020. To facilitate the measurement of emotion profiles (see Section 3.2), we require that the transcripts are available in PDF format, with the following information: company name, earnings quarter, conference call date, corporate participants (e.g. CEO, CFO) and other participants (e.g. analysts), and the presentation and Q&A sections with each participant’s name, position and related text of their talk, which reduces the size of our sample to 15,512 calls. We remove 9,623 calls for which we lack data from Compustat, CRSP or IBES to compute the control variables used in our empirical analyses. Accordingly, our primary tests use 5,889 unique conference calls.

We measure emotions based on an established psychological emotion model. In this model, Plutchik (1980) posits that human emotion consists of eight primary emotions: anger, fear, sadness, disgust, joy, anticipation, trust and surprise. Each of these primary emotions has three degrees of intensities. For instance, for the emotion joy, there are three levels of intensities: neutral (“joy”), intensifier (“ecstasy”) and inhibitor (“serenity”).

Drawing on Plutchik’s (1980) emotion model ^[2], we curated word lists for each of these emotions to build a word list that can be applied to our earnings call transcripts. For this purpose, we employ the artificial-intelligence approach proposed by Gamage et al. (2024). This approach uses a variation of state-of-the-art deep learning language model, the Bidirectional Encoder Representations from Transformers (BERT) (Devlin et al., 2018), which has been shown to outperform traditional ML techniques in emotion and sentiment extraction tasks (Acheampong et al., 2021; Bochkay et al., 2023). We construct seed terms for each of the eight emotion categories.  Appendix 1Table A1 reports some examples of these constructed seed words. We use the DistilBERT base-case model (Sanh et al., 2019) for the finetuning of our emotion list. The final word count for each category totals 1,341 words ^[3]. We manually validate this emotion list by using various transcripts with financial sentences. We perform a cross-validation on the transcripts with the five highly positive and the five highly negative emotions. We consider the surrounding words of the emotion expression because it might contain modifications such as negations, intensifiers and inhibitors for granular and accurate emotion intensity measurement. For example, the expression “It was not a hostile takeover bid” contains a negative term “hostile” while the whole phrase suggests a positive emotion. In such cases, we use Plutchik’s (1980) polar opposites to correctly map emotions. Similarly, intensifiers and inhibitors can significantly contribute towards improving the accuracy of emotion intensity. For example, the positive emotion score of the sentence “There was an increase in yield,” is intensified in “There was a significant increase in yield” and inhibited in “There was limited increase in yield.” In these cases, we employ the emotion words rectification process based on the valence detection approach proposed by Hutto and Gilbert (2014).

Following Adikari et al. (2021), we calculate emotion frequency intensities for each emotion relative to the length of the segment as denoted in equation (1). For each emotion (e), the intensity (θe), is calculated as follows:

where k is the emotion expression; n is the number of emotion expressions; α is the intensity factor, equals 2 if there is a boosting/intensifier intensity, 0.5 if there is an inhibiting intensity, and 1 otherwise; l is the length of the segment. We construct emotion scores for each manager by considering the occurrence and intensity of emotional terms from the list. We provide some examples of emotions extracted from CEO statements during the Q&A section of the conference call in  Appendix 2Table A2. Human annotation of transcripts would provide a complementary benchmark for our automated emotion scoring. While we acknowledge this as a limitation of the study, future research could use targeted human coding on smaller sets of transcripts to further triangulate automated measures.

Table 1 presents the summary statistics and t-test estimates comparing the means of the emotion scores. As shown in Panel A, we find that the mean values of the two emotions, joy and anticipation are significantly higher in the presentation section compared to the Q&A section for the CEO. The average values of surprise, anger, sadness, fear and disgust emotions are significantly lower in the presentation section than in the Q&A section for the CEO while the average value of trust is not significantly different between the two sections of the conference call. In Panel B, we present t-test estimates for the average differences of emotions between conference call participants. The average values of the CEO emotions, joy and anticipation are significantly higher than the Other Managers’ respective emotions in the presentation section. In the Q&A, the average values of all emotions of the CEO are significantly lower than the emotions of the Other Managers. When we compare the averages between the CEO and Analysts, we find that the CEO displays lower average values for all negative emotions while they display higher averages in terms of positive emotions than the Analysts’ emotions.

We use principal component analysis (PCA) to extract factors to explain positive and negative emotion profiles. PCA is a statistical technique that uses an orthogonal transformation to transform a set of raw variables, which are highly correlated into a set of values of linearly uncorrelated variables called principal components (Jolliffe, 2002). By using PCA, we expect to estimate parsimonious regression models rather than using raw emotion variables which would force us to use numerous emotion variables for all participants in one regression estimate. We extract four factors to explain the variances of all emotion variables in the presentation section of the conference call for the CEO and Other Managers. We extract CEO_PRES_NEGT and CEO_Q&A_NEGT to explain the variations of CEO anger, sadness, fear and disgust in the presentation and Q&A section, respectively. CEO_PRES_POST and CEO_Q&A_POST explain all the positive emotions (trust, joy, anticipation and surprise) of the CEO in the presentation and Q&A section, respectively. To evaluate the reliability of the four emotion constructs derived from PCA, we examine the internal consistency of the underlying emotion components. For the presentation section of the earnings calls, Cronbach’s α is 0.66 for positive emotions (joy, trust, anticipation and surprise) and 0.91 for negative emotions (anger, fear, disgust and sadness), indicating moderate and strong internal reliability, respectively. For the Q&A section, Cronbach’s α is 0.67 for positive emotions and 0.89 for negative emotions, again demonstrating moderate to strong internal consistency. We further assess the temporal stability of these constructs by computing Cronbach’s α by year. The presentation positive factor ranges from 0.46 to 0.68 across years and the presentation negative factor from 0.88 to 0.91, while the Q&A positive and negative factors range from 0.50 to 0.67 and from 0.88 to 0.90, respectively. These patterns indicate that the negative emotion constructs are highly stable over time and that the positive emotion constructs exhibit moderate stability while remaining within theoretically acceptable levels for multi-dimensional affect constructs. Overall, the findings confirm that the emotion measures used to generate the positive and negative factors are internally consistent and stable across time.

We use a similar approach to extract emotion profiles for the Other Managers and for the Analysts participating at the conference call. Table 2 reports the factor loadings, and the variance explained for each emotion profile across the presentation (Panel A) and the Q&A (Panel B) sections.

To examine the market reaction on manager emotions, we use the regression model (2):

CAR is the cumulative abnormal size-decile adjusted return during two trading day window surrounding firm i’s conference call in quarter q. CAR (0, +1) starts on the conference call date and ends the day following it. We use PCA to determine emotion profiles (EMOTION). We include the net tone (⁠$TONEi,t$⁠) of the presentation (TONE_PRES_i,t) and the Q&A (TONE_Q&A_i,t) sections as a control variable. We calculate net tone as the total number of positive words less the total number of negative words scaled by the sum of the total positive and negative words in the presentation and the Q&A sections, respectively. We follow the Loughran and McDonald (2011) dictionary in determining the positive and negative words. We control for size, growth, profitability and risk. Firm size (SIZE) is the natural logarithm of market value of equity at the end of the current quarter; firm growth (BTM) is the natural logarithm of the book-to-market ratio; firm profitability (ROA) is the return on assets, calculated as net income divided by total assets; and firm risk is the return volatility (RETVOL), measured as the standard deviation of daily stock returns over the 125 trading days prior to the conference call date. Leverage (LEV), measured by total liabilities divided by total assets, controls for increased information demand when firms are experiencing financial distress. We capture the uncertainty of firm operations (FORE_DISP) using the standard deviation of analysts’ earnings forecasts scaled by the most recent stock price. The model also includes the reporting lag (LAG); the number of annual earnings forecasts (FORE_NUM); and an indicator variable equal to one if firm i reported negative income before taxes in at least one of the most recent four quarters preceding firm i’s conference call at time t, and zero otherwise (LOSS). Following prior literature (Call et al., 2024), we control for POS_EARNSURP_, the absolute value of unexpected earnings if unexpected earnings is greater than or equal to zero and zero otherwise and NEG_EARNSURP, the absolute value of unexpected earnings if unexpected earnings is less than 0 and equal to 0 otherwise. ^[4] Appendix 3Table A3 reports the variables definitions.

Table 3 Panel A contains the descriptive statistics for the dependent and control variables. The sample firms have an average (median) cumulative abnormal return (CAR) of 0.001 (0.002). TONE_PRES (TONE_Q&A) has an average value of 0.19 (0.17). Firm SIZE has an average (median) of 9.70 (9.57) and log of book-to-market ratio of −1.34 (−1.28). The average of return volatility is 0.018. Firm leverage (LEV) is 61% on average. Averages for FORE_NUM and FORE_DISP are 18.24 and 0.12, respectively. 3% of the observations have reported an earnings loss. The average value for POS_EARNSURP (NEG_EARNSURP) is 0.002 (−0.004). Panel B presents the Pearson correlation coefficients between CAR and all the emotion profiles for all participants across the presentation and the Q&A. The negative emotion profile variables are negatively correlated with CAR while the positive emotion profile variables are positively correlated with CAR. Panel C presents the Pearson correlation coefficients between the dependent variable and all other control variables. To assess multicollinearity, we compute variance inflation factors (VIF). The mean VIF across all regressors is 4.10, below conventional thresholds for concern, and the VIFs for the focal emotion variables range between 1.06 and 1.19, indicating no multicollinearity issues that could bias inference.

Table 4 Panel A presents the results for the presentation section. The quarterly abnormal stock returns (CAR) is significantly negatively associated with CEO_PRES_NEGT (β = −0.0027, p < 0.01). CAR is also significantly negatively associated with OM_PRES_NEGT, (β = −0.0014, p < 0.05). Economically, a one–standard-deviation increase in CEO_PRES_NEGT reduces CAR by approximately 52 basis points, while a one-standard-deviation increase in OM_PRES_NEGT reduces CAR by approximately 26 basis points. These results hold even when we include the CEO and Other Managers’ emotions in a single regression model. These results provide some indication that investors react negatively to the managers’ negative emotions portrayed during the presentation section of the conference call. For the positive emotions, we find that quarterly abnormal stock returns respond positively to CEO emotions. Specifically, CEO_PRES_POST is significantly associated with CAR (β = 0.0015, p < 0.05) with a one-standard-deviation increase in this profile associated with an increase of approximately 27 basis points in CAR. Overall, one could argue that market participants react mostly to negative emotion profiles from the firm managers (both CEO and Other Managers) exhibited on the conference call, while for the positive emotion profiles, they only react to the emotion profiles exhibited by CEOs. The results on the control variables suggest that net tone, and those with higher return volatility and higher positive earnings surprises have higher immediate abnormal returns, consistent with previous research.

Table 4 Panel B presents the results for the Q&A section. We find a negative and significant association between Other Managers’ negative emotions and stock returns (β = −0.0028, p < 0.01). We also find a negative and significant association between Other Managers’ positive emotions and CAR (β = −0.0045, p < 0.01). In economic terms, a one–standard-deviation increase in Other Managers’ negative emotions reduces CAR by approximately 52 basis points, and a one–standard-deviation increase in Other Managers’ positive emotions reduces CAR by approximately 78 basis points. As this finding is inconsistent with the theory, we decompose positive emotions expressed by other managers during the Q&A into two sub-dimensions: joy/trust and anticipation/surprise. We construct two new variables reflecting these emotional sub-groups. Re-estimating the main model (results untabulated) reveals that joy/trust is positively and significantly associated with CAR, whereas anticipation/surprise is negatively and significantly associated with CAR. This pattern indicates that the negative market reaction to the original OM_Q&A_POST measure is not driven by positive affect per se, but rather by the anticipation/surprise component, which appears to convey heightened uncertainty to investors. Overall, these results clarify that while supportive emotions of other managers is received favorably by the capital market, unexpected emotions are interpreted as a signal of uncertainty and risk, generating adverse price reactions.

Furthermore, we find that Analysts’ negative emotions (ANALYSTS_Q&A_NEGT) are marginally significant (β = −0.0015 p < 0.10). Economically, a one-standard-deviation increase in analysts’ negative emotions reduces CAR by approximately 31 basis points. We provide evidence that investors react mostly to the negative emotions of Other Managers and Analysts. When we include all emotions (from the CEO, Other Managers and Analysts) into one regression model, we confirm the effect of Other Managers emotions on stock returns.

Taken together, the results reveal that the capital markets respond differently to the emotions expressed by different call participants differently. In the presentation section, negative emotions from CEOs and Other Managers reduce CAR, whereas positive CEO emotions increase CAR. In contrast, during the Q&A section, the strongest negative return effects are associated with Other Managers’ negative emotions and Analysts’ negative emotions, while CEO emotions are largely muted. Positive emotions from Other Managers during Q&A also reduce CAR, suggesting that markets treat spontaneous displays of positivity under uncertainty as overly optimistic or uninformative, consistent with emotional contagion and heightened risk perception.

Our results should be interpreted with caution. Call participants, especially CEOs, are trained communicators who are aware of the capital market implications of their tone and language. The presentation section of earnings calls is usually scripted, rehearsed and strategically constructed to convey confidence and control, minimizing the display of raw or spontaneous emotional cues. Consequently, the emotional expressions observed may not reflect unfiltered or ground-level emotional states, but rather carefully moderated signals designed to manage stakeholder perceptions. This raises concerns about the authenticity of the emotional content captured and whether investors are reacting to actual affective states or to calculated impression management. While our evidence lends some support to the premises of the EASI theory, it also points to the potential limitations of applying this framework uncritically to high-stakes, performative settings such as earnings calls. Future research may benefit from disentangling genuine affective displays from strategic emotional signaling in executive communications.

We examine the impact of emotion profile on CAR subject to either negative or positive earnings news, and we create eight variables that interact each one of the manager emotion profiles with the positive and negative earnings surprise, respectively. Table 5 Panel A (presentation section) reports that the CEO positive emotion profile is negatively associated with stock returns when there is negative earnings surprise (β = −0.0277, p < 0.05), which suggest that market is skeptical about the emotions displayed by the managers during the conference call. However, we do not find significant evidence on other moderating variables. Panel B (Q&A section) shows that the CEO negative emotion profile is negatively and significantly associated with CAR when there is positive earnings surprise (β = −0.1896, p < 0.05), suggesting that investors react to the emotions despite the positive earnings news. Investors respond negatively to the positive and negative emotions displayed by the Analysts when there is negative earnings surprise. These results suggest that investors react negatively in the presence of either negative earnings news and/or negative CEO emotion profiles even when the signals are inconsistent (e.g. bad news about earnings disclosed with positive emotions). Overall, our results provide partial support for H2.

This moderation analyses further support the view that investors react strongly when emotions and economic performance are incongruent. In the presentation section, positive CEO emotions significantly decrease CAR when earnings news is negative (CEO_PRES_POST × NEG_EARNSURP), indicating that markets discount optimistic emotions under adverse economic performance. In the Q&A section, negative CEO emotions decrease CAR when earnings results are positive (CEO_Q&A_NEGT × POS_EARNSURP), and several analyst interactions with negative earnings surprises further reinforce the adverse effect of emotional inconsistency.

We conduct several robustness tests to validate our results. We also address concerns about return leakage by incorporating pre-call return variation. We re-run our regression model and include different event windows to calculate cumulative abnormal returns. We use the day of the conference call CAR (0, +0), and a three-day event window CAR (−1, +1). We report the results in Table 6. On the day of the conference call, we find that CEO negative emotions and Other Managers’ negative emotions, exhibited during the presentation section are negatively associated with CAR at the 10% and 1% significance level, respectively. These results also hold for the other event window (−1, +1). For the Q&A section, we find that during the conference date, stock market participants react negatively to Analysts’ emotions, regardless of Analysts showing positive or negative emotions. For CAR (−1, +1), we find significant results for Other Manager’s emotions during the Q&A section of the conference call. Investors react negatively to either positive or negative emotions from Other Managers suggesting that our findings are not driven by price movements prior to the call. .

Furthermore, we run our main model in equation (2) using alternative emotion profile variables. We calculated positive (negative) emotion for all participants in the call by taking the average values ^[5] of raw emotion variables in the presentation and Q&A sections. The untabulated results are qualitatively similar to the main results.

In addition, to contrast the market response to emotions expressed during the scripted presentation versus the spontaneous Q&A within the same call, we estimate a pooled model including both segments simultaneously. Results (untabulated) reveal a distinct pattern. Investors respond negatively to CEO negative emotions when expressed during the presentation portion of the call (β = −0.00329, p < 0.01). In contrast, the strongest return effects arise during the Q&A: negative emotions from Other Managers (β = −0.00357, p < 0.01) and negative emotions from Analysts (β = −0.00188, p < 0.05) significantly reduce CAR when expressed spontaneously. These findings suggest that while the presentation does shape credibility in the capital markets, spontaneous emotions in Q&A produce an contagion response.

We use the earnings conference call setting to examine the impact of managers’ emotions on stock returns. We use a validated psychology dictionary, the Plutchik’s wheel of emotions (1980), to extract eight primary emotions portrayed during earnings calls. We label these emotions as positive (joy, anticipation, trust, surprise) or negative (anger, fear, sadness, disgust) emotion profiles. We use modern NLP techniques, such as large language models along with lexicons to create a hybrid artificial intelligence framework, with granular levels of detection by examining words in their wider context and by including three intensity levels (neutral, intensifier, inhibitor). Given the prominent role of managers in a firm, how their emotions come across to investors, should not only affect their perceptions of the managers, but should also inform the market’s view of the firm.

Investors react to both positive and negative emotions expressed during the conference call. Specifically, we find that manager emotions are significantly associated with immediate future abnormal stock returns, supporting the view of social effect of emotions. We do not find strong evidence suggesting that investor reaction to emotion profiles is conditional on earnings news. However, we provide some evidence that investors react negatively in the presence of either negative earnings news and/or negative CEO emotion profiles even when the signals are inconsistent (e.g. bad news about earnings disclosed with positive emotions).

An important implication of our study is that information gleaned from emotions during communications between managers and shareholders have economic consequences thus is of interest to corporate executives, analysts and investors who participate in earnings calls and/or use call information to make decisions. Specifically, investor relations teams might actively monitor emotional cues during conference calls and incorporate these insights into their review of conference call performance. Besides, because firms often rely on multiple speakers during conference calls, they might consider guidelines for managing emotional variance to avoid sending mixed signals to investors. Finally, market participants should consider employing tools that explicitly adjust for emotion-driven mispricing.