The Power of Voice:

Managerial Affective States and Future Firm Performance

William J. Mayew Fuqua School of Business

Duke University 1 Towerview Drive Durham, NC 27708

Email: wmayew@duke.edu Web: http://www.fuqua.duke.edu/faculty/alpha/mayew.htm

Mohan Venkatachalam Fuqua School of Business

Duke University 1 Towerview Drive Durham, NC 27708

Email: vmohan@duke.edu Web: http://www.duke.edu/~vmohan

July 21, 2008

We appreciate the assistance of Amir Liberman and Albert De Vries of Nemesysco for helpful discussions and for customizing the LVA software for our academic use. We acknowledge helpful comments and suggestions from Dan Ariely, Jim Bettman, Patricia Dechow, Lisa Koonce, Feng Li, Mary Frances Luce, Greg Miller, Chris Moorman, Chris Parsons, Paul Tetlock, TJ Wong, and workshop participants at Chinese University of Hong Kong, University of Connecticut, Duke Finance Brown Bag, Fuqua Summer Brown Bag, Penn State University and Vanderbilt University. We thank Daniel Ames, Erin Ames, Jacob Ames, Zhenhua Chen, Sophia Li, Ankit Gupta, Mark Uh, and Yifung Zhou for excellent research assistance.

The Power of Voice: Managerial Affective States and Future Firm Performance

Abstract

In this study, we measure managerial affective states during earnings conference calls by analyzing conference call audio files using vocal emotion analysis software. We hypothesize and find that negative affect displayed by managers discussing their firms’ results and prospects is informative about the firm’s financial future. In particular, we find that managers exhibiting negative affect are less likely to meet or beat earnings expectations over each of the three subsequent quarters. However, market participants do not immediately impound these implications. Negative affect is not associated with analyst forecast revisions of next quarter’s earnings. Over the subsequent 180 trading days, cumulative abnormal returns are negatively associated with negative affect, revealing that the market eventually impounds the implications of negative affect into price. Together our findings suggest that vocal cues contain useful information about firms’ fundamentals, incremental to, and of comparable magnitude to, qualitative information conveyed by the linguistic content.

1

“It is not what you say that matters but the manner in which you say it; there lies the secret of the ages.” - William Carlos Williams

Managers disseminate an inordinate amount of quantitative and qualitative information

about their actions and firm performance on a voluntary and mandatory basis through several

avenues such as press releases, quarterly and annual reports, and earnings conference calls. Prior

literature is replete with articles that evaluate the extent that capital market participants react to

quantitative information contained in such disclosures. Only recently have researchers begun to

explore the capital market implications of qualitative verbal communication via financial news

stories (Tetlock, 2007; Tetlock, Saar-Tsechansky, and Macskassy, 2008), annual reports (Li,

2006) conference presentations (Bushee, Jung and Miller 2007), and earnings press releases

(Davis, Piger, and Sedor, 2007; Demers and Vega, 2007). In general, the findings support the

hypothesis that qualitative verbal communication by managers is incrementally useful to

quantitative information in predicting future firm fundamentals and stock returns. This paper

extends this line of inquiry by focusing on how one important type of nonverbal communication,

vocal cues from executives during conference calls, can inform about a firms’ future profitability

and stock returns.

Using a sample of conference call audio files and proprietary Layered Voice Analysis

(LVA) software, we analyze managerial vocal cues to measure positive and negative dimensions

of a manager’s affective or emotional state. Vocal cues or expressions are considered to be

important in drawing inferences about both positive affective states (e.g., happiness, excitement,

and enjoyment) and negative affective states (anger, fear, sadness, tension, and anxiety). We

hypothesize that managers’ affective states during conference calls help predict future firm

performance. Consistent with this prediction, we find that negative affect, exhibited by both

2

CEOs and CFOs during earnings conference calls, is negatively associated with future earnings

realizations. In particular, when managers exhibit high negative affect during conference calls, it

results in a lower likelihood of meeting or beating earnings expectations over the subsequent

three quarters. We find no relation between positive affect and future accounting performance.

Market participants are slow to impound the implications of negative affect for future

performance. We also fail to find an association between negative affect and analyst forecast

revisions for next-quarter’s earnings. Interestingly, we find managerial negative affect is

positively associated with abnormal stock returns over the three-day window centered on the

earnings conference call date. Investors, however, do impound the future performance

implications of negative affect into price over time. Negative affect is negatively related to

cumulative abnormal returns over the subsequent 180 trading days following the earnings

conference call. A hedge strategy based on decile ranks of negative affect generates abnormal

returns of about 9% over 180 trading days, which is incremental to and statistically equivalent to

the hedge returns generated using linguistic content.

While we cannot identify why market participants are slow to act on vocal cues, we offer

two conjectures. First, market participants may not value nonverbal communication if they hold

a belief that there is little incremental information beyond that contained in quantitative and

linguistic qualitative information provided around earnings announcements. Second, it may be

that market participants do care about nonverbal communication but cannot discern the

information contained in the vocal cues because such cues are at frequencies too difficult to

detect and/or process without the aid of computer software.

This study makes three contributions. First, to our knowledge, this is the first paper to

provide evidence on the role of nonverbal communication in a capital market setting. We apply

3

findings in social psychology research that provide unequivocal support for vocal expressions as

one particular type of nonverbal communication that is influential when communicating

messages over and above the verbal content of message (Mehrabian and Weiner, 1967; Scherer,

London, and Wolf, 1973; Scherer, 2003). Our findings confirm that important information can be

gleaned from vocal cues in the capital market setting by showing that managers’ emotional state

is associated with future firm performance, after controlling quantitative information and

qualitative verbal content.

Second, our results provide new insights into how conference calls can provide

information to financial markets. Prior research documents that conference calls provide

significant information to market participants above and beyond that contained in the earnings

press release (e.g. Frankel, Johnson, and Skinner, 1999; Tasker, 1998; Bushee, Matsumoto, and

Miller, 2004). As conference call audio broadcasts are commonplace for many firms and open to

public access subsequent to Regulation FD, our findings suggest that investors can utilize vocal

cues during such communication to learn about a manager’s affective state, and in turn about the

firm’s financial future.

Finally, we exploit a new tool that can potentially help researchers measure emotional

states. Our results provide preliminary evidence that the LVA software can produce useful and

reliable proxies for managerial affect in the capital market setting. Future research in both

economics and psychology can explore vocal cues using this software in other settings. For

example, examining information about the affective states of economic leaders can perhaps

inform about broader changes in economic fundamentals. For psychology researchers who rely

on human subjects to observe and measure affect from voice, this represents a useful and feasible

4

alternative. Certainly more empirical validation of this software’s reliability is necessary, but we

view our results as encouraging.

Our paper proceeds as follows. In Section I we review related literature and develop our

hypotheses. Section II discusses the nonverbal measures used in the study. In section III we

outline our sample selection, define our variables of interest, and provide descriptive statistics.

Sections IV and V provide our empirical results and robustness checks, and in Section VI we

offer concluding remarks.

I. Related Research and Hypotheses Development

A. Related Research

Research in social psychology has long held the view that nonverbal cues such as vocal

and facial expressions significantly impact how a message is interpreted. Communication experts

generally agree that in face-to-face conversations, only a small fraction of the message regarding

emotional state is contained in the verbal content (e.g., Mehrabian, 1971). A significant

component of the message is contained in vocal attributes such as voice intonation, accent,

speed, volume, and inflection. Kinesics, i.e., facial expressions, postures, and gestures, also plays

a large role in communication. However, we do not study them in this paper and will therefore

not elaborate further on the role of kinesics. We will focus on the vocal channel and describe

how voice can convey emotions or affective states reliably to a receiver (see for example Juslin

and Laukka, 2003).1

The expression and perception of emotional states via vocal cues are fundamental aspects

of human communication. People express emotions by yelling; using a quiet, low or monotonous

voice; and at the extreme, by being silent (e.g., Walbott et al., 1986). Such expression of

1 Although we use the terms affect and emotion interchangeably, there is a subtle but important difference between the two. Emotion refers to a feeling that occurs in response to events, while affect is viewed as valence of an emotional state (Frijda, 1993).

5

emotions through voice can be used to convey information or influence others. Several studies

have shown that the tone of a person’s voice leaks information about an affective state that is not

revealed by the verbal content or facial expressions associated with the message (Zuckerman et

al., 1982). Juslin and Scherer (2005) review fifty years of research establishing that acoustic

voice patterns provide insights into the speakers affective, or emotional, state. While the role of

nonverbal cues has been studied extensively in the social psychology literature, it is virtually

absent from the accounting and finance literatures.

Corporate financial reporting represents an important channel for managers to

communicate information to various stakeholders and much of the literature has primarily

focused on the capital market implications of quantitative information disclosed in the financial

statements. Recently, researchers have begun to explore verbal communication as an additional

mechanism through which information is conveyed and used in capital markets. For example, the

role of verbal communication has been examined in the context of financial news stories

(Tetlock, 2007; Tetlock et al., 2008) and messages in Internet chat rooms (Antweiler and Frank,

2004). Using the General Inquirer textual analysis software, Tetlock (2007) examines whether

pessimism in news media language content can predict movements in market returns. He finds

that pessimism in the Wall Street Journal “Abreast of the Market” column exerts significant

downward pressure on aggregate market valuations—but, such pessimism reverts quickly and

has no implications for aggregate fundamentals. Building on Tetlock (2007), Tetlock et al.

(2008) measure the pessimism in financial news stories for a sample of S&P 500 firms and find

that pessimism predicts lower future firm earnings and stock returns.

Antweiler and Frank (2004) examine whether messages communicated through Internet

message boards such as Yahoo! Finance have predictive ability for subsequent stock returns,

6

trading volume, and volatility. They report that an increase in message board activity results in

subsequent negative returns, greater trading volume, and return volatility. This evidence is

consistent with Internet message boards conveying financially relevant information.

In the accounting literature, Davis et al. (2007) and Demers and Vega (2007) analyze the

linguistic style of managers’ language use in earnings press releases. They employ computerized

textual-analysis software, DICTION 5.0, to measure the extent that managers exhibit optimism

or pessimism in their language use. Collectively, they find that the levels of optimistic and

pessimistic language in earnings press releases predict future performance and correlate

significantly with contemporaneous market reaction surrounding the press release. Li (2006)

examines the information content of textual information in annual reports.2 He develops a

linguistic-based risk-sentiment measure by using words in the annual report and demonstrates

that this risk-sentiment measure predicts both future earnings and returns.

Another avenue for managers to communicate information is via presentations at various

conferences. Research by Bushee et al. (2007) shows that conference presentations are

economically important information events in that they find significant short-window market

reaction surrounding the conference date. In addition, they find that such conference

presentations increase analyst and investor following consistent with the notion that firms use

these venues to improve firm visibility and generate investor following.

While a growing body of literature explores the role of verbal communication in the

financial markets arena, the implications of nonverbal communication represent a fairly nascent

and unchartered territory. One exception is Coval and Shumway (2001) who examine the role of

ambient noise level in the Chicago Board of Trade’s bond future trading pit. They find that

2 In the management literature, Bettman and Weitz (1983) explore the linguistic content in annual reports to examine how managers explain firm outcomes. However, they do not examine investor reactions to managers’ attributions. They find that managers attribute unfavorable outcomes to external and uncontrollable causes.

7

ambient sound level conveys economically and statistically meaningful information and that

traders process subtle and complex nontransaction signals in determining equilibrium prices.

While this suggests that decibel levels in trading pits have information content for equilibrium

supply-and-demand conditions in the futures market, it does not speak directly to the specific

attributes of nonverbal communication between managers and market participants that we

address in our study.

B. Hypotheses

The affective or emotional state of an individual enables us to draw inferences about the

events or type of events that caused an individual to be in such an emotional state. This is based

on the appraisal theory of emotion that is founded on the notion that emotions arise or are

elicited by evaluations or appraisals of events and situations (Arnold 1960; Roseman 1984;

Lazarus 1991). For example, a positive state is elicited by a successful outcome such as winning

a basketball game, passing an exam, or being selected to a University. In contrast, a negative

state is elicited by personal loss, frustration, cognitive dissonance or simply a bad outcome.

Frijda (1988) calls it the Laws of Situational Meaning and Concern and states that “emotions

arise in response to the meaning structures of given situations; . . . arise in response to events that

are important to the individual’s goals, motives, and concerns.” In other words, it is the

interpretation of events and situations rather than the events and situations themselves that result

in a particular emotional outcome or affective state.

In the context of financial markets, where managers communicate information to

investors about both past and future performance, it is likely that managers exhibit different

affective states depending on their interpretation of events and situations pertaining to the firm.

The determination of managerial affective states should enable investors infer the managers’

8

implicit assessment of firm performance, both past and future. For example, a manager is likely

to exhibit positive affective state if the manager expects positive future firm performance due to

private information regarding current outcomes (e.g., persistence of current period earnings)

and/or future outcomes (e.g., prospective drug approval, anticipated orders, successful outcome

of strategic initiatives such as restructuring). In such instances, a manager is more likely to be

excited or exhibit positive psychological arousal in communication with investors. Therefore,

we posit that positive affective state will be positively related to future firm performance.3

In contrast, a manager may exhibit negative affective state when he/she has private

information that is not indicative of positive future performance. Examples include information

about the transitory nature of accounting earnings, impending law suits, product failures or order

cancellations. Negative affect may also stem from managers’ psychological discomfort due to

cognitive dissonance. The theory of cognitive dissonance, developed by Festinger (1957), is

based on the notion that inconsistency between an individual’s beliefs and their actions creates a

feeling of discomfort and anxiety. In experiments conducted by Elliot and Devine (1994)

counterattitudinal behavior evoked psychological discomfort arousing a negatively valenced

state (see also Harmon-Jones 2000).

To apply cognitive dissonance in the economic setting we explore here, consider a

manager who believes that she is competent and in control of the firm she operates. Information

about firm performance would reflect her actions taken while running the firm. If the manager

has private information that is inconsistent with her own beliefs regarding her competence, an

uncomfortable emotional state will arise from this dissonance. As such, we posit that cognitive 3 We assume that a manager’s affective state is not an innate characteristic of the manager per se. Rather, it is time-dependent and is a function of private information about the firm that managers possess during the conference call. It is plausible that a manager could exhibit both positive and negative affective states during the conference call if the manager has both good news and bad news about specific issues discussed during the conference call. For example, a manager may discuss poor past performance in the form of negative earnings surprise and at the same time discuss better expected future performance as a result of increasing backlog of orders.

9

dissonance induced negative affect should be indicative of potential bad news or uncertainty

about good news. Therefore, if we observe a negative affective state in a manager it is more

likely that events and circumstances are unfavorable and/or that the manger is psychologically

uncomfortable due to cognitive conflicts in elements of information that the manager has.4 Thus,

we posit that negative affect will be negatively related to future firm performance.

We next develop hypotheses for the stock market implications of managerial affective

states inferred from management communication. If managerial affect has new information

about a firm’s future performance we should observe a market reaction surrounding the

communication date in a manner consistent with the signal contained in each of those cues. We

therefore predict a positive (negative) contemporaneous stock market response to positive

(negative) affect.

Observing such a market reaction is contingent on the efficiency with which market

participants observe and act on the information contained in the affective state. Research in

social psychology suggests that vocal indicators of various emotions are accurately detected and

are often as good as or better than those of facial cues and expressions (Kappas, Hess, and

Scherer, 1991). It is also widely accepted that one’s voice is not easily controlled and that the

voice channel “leaks” more information than facial cues (Ekman and Friesen, 1974). Evidence

in Ambady and Rosenthal (1993) suggests that human beings can form impressions and

judgments from even “thin slices” of nonverbal behavior. However, the strength of any relation

4 Our study attempts to measure the negative affect resulting from cognitive dissonance. Other research has investigated how individuals in economic settings take action to avoid cognitive dissonance ex ante. Akerlof and Dickens (1982) formally model cognitive dissonance costs as part of expected utility maximization and discuss how workers in dangerous occupations take actions to avoid reminders of how dangerous their work is. Argentesi, Lutkepohl and Motta (2007) show that individual investors (who believe they are competent investors) are more likely to avoid buying newspapers when their holdings have likely experienced losses because reading the paper would confirm unfavorable news would cause dissonance. Prentice (2003) notes that Enron repeatedly rebuked critics, making it difficult for its employees to process any negative information about the firm. In relying on management, employee beliefs were that the firm was in fine financial shape, and employees in turn would not process information to the contrary to avoid dissonance costs.

10

between affect and contemporaneous stock returns is also subject to the information processing

efficiency and ability of the analysts and investors as well as measurement error in the nonverbal

measures.

II. Measuring Nonverbal Communication

The main challenge in this study is to construct useful and reliable measures of affective

states from nonverbal communication. We utilize CEO and CFO voice recordings from earnings

conference calls to develop measures of managers’ emotive states when communicating

information to analysts and investors.5

There are several advantages to using the audio content in earnings conference calls.

First, conference calls represent a common and important disclosure mechanism for U.S. firms.

They are intended to supplement mandated disclosure of earnings information and, in particular,

help analysts understand the implications of specific events or items reported in the earnings

press release. Early research in the area (Frankel et al., 1999; Tasker, 1998) demonstrated that

conference calls provide material information to analysts in particular, and to investors, more

generally. The passage of Regulation FD has not diminished the role of conference calls

(Bushee, Matsumoto, and Miller, 2004), and in fact most public firms regularly host quarterly

earnings conference calls to comply with Regulation FD (Skinner, 2003). Recent research

explores the implications of differences in the nature of conference calls, in terms of the amount

of time allocated by firms for the conference call, the linguistic tone (Matsumoto et al., 2007;

Frankel, Mayew, and Sun, 2007), the interactions between the presentation and discussion

portions of the call (Matsumoto et al., 2007), and management discrimination in allowing

5 In the psychology literature, nonverbal cues are often generated by using actors to produce vocal emotion expressions, and human judges are used as “decoders” to determine whether such vocal patterns are recognized. While professional actors can provide strong vocal cues and it is easy to get consistent audio recordings, their emotional portrayals may differ from vocal expressions that occur in real life.

11

analysts to ask questions (Mayew, 2008). Together, these studies suggest conference calls are

important venues for managerial communication with stakeholders of the firm. We extend this

literature by examining whether vocal cues from managers during conference calls add

incrementally to the quantitative and qualitative linguistic information already provided in the

earnings conference call, as suggested by Block (2000).6

A second advantage of using conference calls is that, unlike annual meetings where

managers appear face-to-face to meet with current investors, conference calls offer one of the

few opportunities for firms to communicate directly with current and potential investors as well

as other stakeholders. Third, because conference calls are rarely broadcast over video, other

channels of nonverbal communication such as facial expressions and gestures (kinesics) do not

contaminate the signal in the voice channel. In other words, we are able to isolate the vocal

channel of the nonverbal communication.7 Additionally, vocal expressions have been shown to

be more effective than facial expressions over long distances (Marler, 1977), which is certainly

the case with conference calls where analysts and investors are often many miles away from the

managers. Lastly, as a practical matter, we were able to obtain audio files of the conference calls

from the Thomas Financial StreetEvents database.

We first considered using human judges to interpret emotive states in the managerial

communication during conference calls. The use of human judges to interpret emotion

expressions is a difficult and complex procedure, mainly due to the difficulties with an

6 An inherent assumption in prior research (e.g., Davis et al. 2007) that examines linguistic content in press releases is that managers use linguistic style in press releases and conference calls truthfully. It is unclear what the relative cost-benefits tradeoffs are for managers to engage in truthful revelation versus opportunistically manipulating the market via optimistic and pessimistic tone in earnings press releases. In contrast, our paper does not rely on the assumption of truthfulness on the part of managers. Rather, the vocal measures that we use capture both voluntary and involuntary aspects of communication and hence are less likely to be influenced by opportunistic manipulation of information. To the extent that managers are able to portray emotions to obfuscate information, we would bias our tests against finding results in support of the predictions. 7 Note that the message recipient may react to the verbal or linguistic aspect of the communication in addition to or in lieu of the nonverbal content. We control for this possibility by including linguistic tone in the empirical analysis.

12

appropriate judge selection procedure and the lack of reliable inferences on judgments made by

them. The use of sophisticated computer-aided analysis of emotional expressions has been

steadily increasing in recent years. However, acoustic measurement brings with it its own set of

problems. The sound recording quality is very crucial because the success of the acoustic

measurement depends critically on the recording quality. Furthermore, translating the vocal cues,

such as the fundamental frequency or voice pitch, voice intensity, voice quality, and temporal

aspects of speech into specific measures that capture emotive states is nontrivial.

We construct measures of affective states with the help of a computer software program

that uses Layered Voice Analysis Technology (LVA) developed by Nemesysco Ltd. LVA was

invented in Israel in 1997. Lie detection was the main purpose for which this software was

originally developed, but the technology now serves a broader set of constituents by providing

voice analysis solutions through the detection of emotion. Commercial applications utilizing

LVA emotion-detection technology include call center telephone monitors to detect emotional

states of customers and customer care specialists, healthcare applications to detect emotional

stress of patients, insurance fraud detection and law enforcement interrogations. LVA is

comprised of a set of unique proprietary signal processing algorithms that identify different types

of stress, cognitive processes, and emotional reactions. The algorithms use 129 vocal parameters

to detect and measure minute, involuntary changes in the speech waveform and create the

foundation for indentifying the speakers’ emotional profile. As such, LVA technology enables a

better understanding of a subject’s mental state and emotional makeup during the time that he or

she is speaking.

In its fundamental layers, the LVA technology identifies several key indications of the

waveform, each capturing a different type of brain activity or a cognitive process. There are

13

primarily two main components measured in the human waveforms by the LVA technology that

are pertinent to our research—the Relative High Frequency Range (RHFR) and the Relative Low

Frequency Range (RLFR), each indicative of different aspects of emotions and cognitive

activity. The RHFR and the RLFR are measured using a proprietary mathematical formula and

are indicative of the presence of certain frequencies, rather than the amplitude of those

frequencies.8 While the RHFR captures high states of excitement and emotional arousal, RLFR

captures states of thinking and other cognitive processes like imagery. The technology classifies

the multitude of permutations between these two ranges into fundamental layers that in turn help

identify key indications on the waveform that are suggestive of a particular type of brain activity

or a cognitive process. These fundamental indications are then categorized into emotional,

logical, and stress-related groups which are then parameterized using the proprietary algorithm

developed by Nemesysco.

While the LVA software identifies several different parameters that capture emotional,

logical, and stress characteristics, we focus on two measures that are relevant for

operationalizing positive and negative managerial affect.9 The first measure, cognition level,

measures the level of cognitive dissonance (Festinger, 1957), which is a determinant of negative

affective state (Forgas, 2001). Cognitive dissonance is the uncomfortable, anxious feeling an

individual experiences when beliefs and actions are contradictory. Cognition level is a number

based on low vocal frequency parameters that capture cognitive conflicts and general cognitive

8 There are other vocal properties captured by the LVA technology for various other uses but we consider those attributes as outside the scope of this research. 9 There are several parameters that are generated by the software. The commercial versions of the software (e.g., eX-Sense Pro) combine these parameters to capture and portray different attributes of vocal expression. However, the commercial versions do not produce the individual parameters that are needed for our academic purposes. Rather, they merely output the average parameter and graphically display the levels of the different vocal attributes (such as cognition and excitement levels). Thus, the commercial versions are more relevant and useful for context-specific and real time analysis of a subject. For our purposes, we require detailed vocal parameter values that are generated throughout the conference call. We thank Nemesysco for accommodating our request to modify the software so that we can obtain detailed parameter values for the two vocal attributes we study.

14

activity. This indicator takes on values ranging from 30% to 300%. Values above 120% indicate

increasing doubt in the mind of the subject when speaking. Thus, the higher the cognition level,

the greater the subject’s cognitive conflict, and hence, the larger the negative affect (hereafter,

NAFF).

The second measure, emotion level, measures the level of excitement exhibited by the

subject. Excitement is one of the biological expressions that accompanies a positive affective

state (Tomkins, 1962). As with cognition level, emotion level values range from 30% to 300%.

Emotional levels greater than 110% indicate rising levels of excitement. Thus, the higher the

emotion level, the larger the positive affect (hereafter, PAFF).10

While the LVA technology underpins various software products for commercial and

entertainment purposes (see www.nemesysco.com/solutions for a complete list), a reader might

question the internal validity and reliability of LVA measures. We are not aware of a systematic

academic examination of the validity of the measures generated by commercial versions of LVA

based software. However, there exists some evidence on the reliability of an early generation

LVA based lie detection software product called Vericator. A study conducted by the U.S. Air

Force Research Laboratory (Heddad and Ratley, 2002) with funding from National Institute of

Justice examined several voice stress analysis software including Vericator. Using a sample of

48 utterances by two suspects who confessed to committing murder, they concluded that the

Vericator system detected all the instances where the subject was in some form of stress and/or

displayed deceit.11 In contrast, a study by Sommers (2006) concluded that this Vericator software

10 In the limit, a high emotion level is likely when the context is either deceptive or traumatic in nature. To the extent that such situations dominate in the determination of the PAFF, it will bias against finding the predicted relation. 11 The same report reported a field test by a criminal investigator from the New York Police Department who strongly supports the use of this technology as a “valuable investigative tool for the law enforcement officer on the streets of our cities, towns, and villages across the nation.” (Heddad and Ratley (2002)).

15

is fairly effective in identifying certain variations in stress levels, but at higher levels of stress the

ability to detect deception is no better than that obtained by chance.

Based on the scant empirical evidence, it is difficult to determine ex ante the reliability of

measures constructed from the LVA technology. Further, the focus of these studies was to

validate an LVA based software engineered for identifying deception. The primary goal of the

LVA technology is not lie detection, but rather to capture a broader set of vocal and emotion

attributes. The objective of our work is to utilize this technology to measure affective states of

managers of publicly traded corporations.

Regardless, we acknowledge that our results are going to be influenced by the reliability

of the nonverbal measures. To the extent that these measures have considerable measurement

error, it would bias against finding any relation between these measures and future firm

performance. In addition, the vocal parameters captured by the LVA technology measure cues

that are often subtle and sometimes inaudible to the human ears, and hence analysts and investors

listening in on the conference call may not detect and/or process these cues without the aid of

computer software. Hence, we may not observe a relation between these measures and

contemporaneous stock returns or analyst forecast revisions.

III. Sample Selection, Variable Measurement, and Descriptive Statistics

We derive our sample of audio files from all conference calls held between January 1 and

March 31 of 2007 available on the Thomson StreetEvents database. We chose this time period

for two reasons. First, Thomson Financial does not retain the audio files indefinitely. They

archive the audio files for approximately one year following the conference call date after which

they are deleted. We wanted to ensure that the audio files we consider were available for a

sufficient time period to facilitate completing the empirical analyses. Second, each of the audio

16

files requires a processing time of approximately two hours to construct the LVA measures.

Because the costs of data collection and processing are exorbitant, we restrict our sample to

conference calls that occur in the first calendar quarter of 2007.

During this time period, we identify 2,650 conference calls of fiscal year 2006 4th quarter

earnings where company identifiers are available on the CRSP, Compustat and I/B/E/S

databases. We focus on fiscal year-end conference calls because they have different

characteristics than non-fiscal year-end conference calls (Frankel et al., 2007). From this initial

sample we remove 1,569 observations without archived and indexed conference call audio files,

as indexed audio files are required for utilizing the LVA software. We then remove 466

observations where missing data on either CRSP, Compustat or I/B/E/S prevents the construction

of variables needed in the empirical analysis. The final sample consists of 615 firm conference

call observations that we analyze using the LVA software.

We obtain stock return data from the CRSP database and www.yahoo.com as necessary.

We obtain financial data from the Compustat database to the extent it is available. For financial

data relating to the most recent periods, we hand-collect it from the Edgar database at

www.sec.gov. We obtain analyst expectations of earnings and earnings forecast revision data

from I/B/E/S.

To construct our measures of managerial affect during conference calls, we run the

conference call audio files through LVA. The software requires a calibration period over which

“normal” voice characteristics of the speaker are measured. Subsequent to calibration, LVA

analyzes audio output at constant intervals relative to the calibration benchmark and produces

various measures, including our variables of interest, cognition level and emotion level, which

17

serve as the basis for negative and positive affect. LVA measurement continues until the

researcher manually ends the test.

The audio files provided by StreetEvents are uniquely suited for LVA analysis for two

reasons. First, firm executives commonly begin the conference call with mundane introductions

of the conference call participants and Safe Harbor statements. These “boilerplate” opening

statements are ideal for calibrating the voice of each executive because they require little

cognitive investment. Second, StreetEvents has a proprietary technology called “indexed audio”

that maps audio files onto the conference call transcripts. With indexed audio, a researcher can

point and click to specific locations of the entire conference call where a given executive speaks.

Since voice analysis is speaker dependent, the use of audio indexing allows us to seamlessly

capture the vocal content for a given executive throughout a conference call dialog without the

confounding effects of other speakers.

For each conference call, we separately measure positive and negative affect for the CEO

and CFO because each individual speaker has a different vocal pattern that requires separate

calibration. We calibrate each executive based on their introductory remarks in the presentation.

If an executive does not provide introductory remarks in the conference call we calibrate his

vocal pattern using the first few minutes of his speech. We aggregate the affect measures

obtained for both executives present in a call to obtain a firm level NAFF and PAFF measures.12

LVA measures each parameter approximately 35 times per minute. This implies for a 10 minute

CEO speech, LVA will generate 350 parameter readings.

12 We do not analyze CEO and CFO affective states separately because a priori we do not anticipate the two executives to have different information sets or different appraisals of the same information set and consequently, do not expect them to be in different affective states. The pearson correlation coefficients between CEO and CFO PAFF and NAFF measures separately are positive and statistically significant (ρ= 0.28 and 0.59; p = 0.00).

18

To generate conference call level measures of NAFF and PAFF, we measure how many

individual cognition level and emotion level readings from each executive were above the

“critical” level as defined by the developers of LVA. We count the number of “critical” instances

and scale it by the total number of individual readings. With respect to the cognition level,

readings above 120% are indicative of severe cognitive dissonance by the subject. Hence, we use

the proportion of readings that have cognition levels above 120% to construct the NAFF

measure. For PAFF, we measure the proportion of readings with emotion levels greater than the

critical 110% level. Figure 1 reproduces a screen shot from LVA for one CEO in our sample and

depicts a graph that contains the various values of specific parameters LVA measures. The actual

values of cognition level and emotion level graphically depicted are exported to a data file which

is subsequently processed to obtain the NAFF and PAFF values.

Panel A of Table 1 presents the descriptive statistics for the emotion measures as well as

financial and accounting variables used in the study. The mean (median) for PAFF is .1028

(0.1039) indicating that on average, managers exhibit positive affect 10% of the time during a

conference call. In contrast, managers exhibit negative affect about 17% of the time (mean NAFF

= .1663). The sample firms have an average (median) quarterly return on assets (ROA) of 0.37%

(0.92%) and assets of $9.1 ($1.2) billion. The mean (median) firm has revenues of $911 million

($196 million) and market value of equity of $5.4 billion ($1.3 billion) (results not tabled). Thus,

our sample has predominantly large firms. In Panel B we provide the industry composition for

our sample firms. While we do not observe significant industry clustering, the sample contains a

relatively greater number of firms from the computer, financial, and services industries.

The pearson correlation matrix of all the financial variables and the two affect measures

are presented in Panel C of Table 1. Several observations are worth noting. First, we observe

19

that NAFF is negatively related to size (LNMVE) (ρ = -0.16, p = 0.00), negatively related to

current ROA (ρ = -0.08, p = 0.06), and positively related with both STDROA (ρ = 0.09, p = 0.03)

and VOL (ρ = 0.14, p = 0.00). These correlations serve as internal validity checks for the affect

variables computed by the LVA software. Recall that NAFF is based on measures of cognitive

dissonance. For managers who believe in they are competent and in control of their firms,

negative current accounting performance will cause cognitive dissonance because it undermines

the manager’s belief about competency. Additionally, if small firms and firms with high

volatility capture settings that are more uncertain, it is likely that managers who believe they are

in control of the firm will experience cognitive dissonance.

Second, we do not observe a systematic relation between the two affect measures (ρ = -

0.03, p = 0.51). This is not surprising because managers discuss many issues during a conference

call, each of which may induce a positive or negative affect on the manager.13 Further, research

suggests that positive and negative affect need not be negatively correlated (Diener and Emmons

1985, Cacioppo and Bernston 1994). The lack of relation also suggests that neither affect

measure subsumes the other in a univariate sense. Third, we find that NAFF is negatively related

to one quarter ahead analyst based earnings news ( AtUE 1+ ) (ρ = -0.13, p = 0.01). This provides

some of the first clues about potential information contained in these affect measures for

predicting future performance. Finally, NAFF is negatively related to future abnormal returns

(FFCAR(2,180)) (ρ = -0.09, p = 0.03).14 However, we do not find a significant positive relation

between positive managerial affect (PAFF) and future performance in the univariate setting.

13 Some firms explicitly attempt to provide a balanced view of the firm such that a portion of their conference call presentation is dedicated to positive aspects of the firm and another portion to negative aspects. For example, Cisco Systems noted the following in its Q1 2004 earnings conference call: “Reminding those who have limited exposure to our prior conference calls, we try to give equal balance to both what went well and our concerns.” 14 Johnson (2004) argues that firms with high idiosyncratic uncertainty have increased option values that expire over time and yield negative future stock returns. Since NAFF is positively correlated with idiosyncratic return volatility

20

IV. Results

A. Does Managerial Affect Predict Future Firm Performance?

In this section we test whether vocal cues provide insights into managerial affective states

that in turn inform about future firm performance. In particular, we test the hypothesis that

positive affect (negative affect) is positively (negatively) related to future performance. To test

this relation we estimate a logit model that determines the probability that a firm will meet or

beat earnings thresholds. We focus on the sign of future unexpected earnings as opposed to the

magnitude for two reasons. First, recent research finds that meeting or beating analysts’

forecasts or prior period earnings benchmarks results in a significant stock price premium

(Bartov, Givoly and Hayn 2002; Kasznik and McNichols 2002). Such premium is attributed to

information about firm fundamentals in that it predicts future firm performance. Furthermore,

prior research (Ou and Penman 1989) finds that predicting the direction of future earnings

changes helps generate significant abnormal returns. Second, we can not specify the exact

functional form relating the affect measures and the magnitude of future earnings changes.15

To determine whether a firm has met or exceeded earnings expectations we consider two

benchmarks: seasonally lagged quarterly earnings and consensus analyst earnings forecasts.

Graham, Harvey and Rajgopal (2005) report that CFOs care about both four quarters lagged

earnings and consensus forecasts and that management’s inability to meet such benchmarks

would be viewed by the market as the firm having poor future prospects. In addition, Rees

(2005) reports that a trading strategy based on the prediction of firms reporting a positive

earnings change and meeting analysts’ earnings forecasts yield significant excess returns.

(VOL), an alternative explanation for the negative relation between NAFF and future stock returns is that NAFF simply captures firm specific idiosyncratic uncertainty. In our multivariate analysis, we control for idiosyncratic return volatility. 15 Estimation of model (1) using the magnitude of earnings surprises results in inferences consistent with those reported Table 2 Panel A.

21

We estimate the following empirical specification:

Pr ( jtBEAT 1+ ) = β0 + β1 PAFF + β2 NAFF + β3

jtBEAT + β4 FREV + β5 FDISP

+ β6 LNMVE + β7 MOM + β8 BM + ε (1) where j

tBEAT 1+ is an indicator variable that equals 1 if the firm equals or exceeds the earnings

benchmark j (where ( )EAj ,∈ , A = quarterly consensus analyst earnings estimate, and E = four

quarters lagged earnings) in period t+1. PAFF and NAFF represent the percentage of time the

emotion level and cognition level scores computed by LVA were above the critical levels during

the conference call. We predict a positive (negative) coefficient on PAFF and NAFF. We also

predict a positive coefficient on jtBEAT as managers have incentives to establish strings of

positive earnings surprises (Bartov et al. 2002). The model also includes several other control

variables for analyst forecast revisions, forecast dispersion, firm size, return momentum, book-

to-market ratio and return volatility.

We measure analyst forecast revisions (FREV) as the one quarter ahead forecast revision

representing the difference between the median one quarter ahead forecast issued after and

before the current period earnings announcement, scaled by stock price two days preceding the

conference call. The median forecast before (after) the current period earnings announcement is

determined using the last (first) forecast of all individual analysts issuing forecasts during the 90

day period before (after) the current quarter earnings announcement date. We measure forecast

dispersion (FDISP) as the standard deviation of analysts’ earnings per share forecasts derived

from the distribution of I/B/E/S consensus earnings per share forecasts immediately prior to the

earnings announcement.16 Return momentum (MOM) is measured as the cumulative daily

abnormal return over the 125 day trading window [-127,-2] prior to the earnings announcement.

16 Inferences are unchanged when we scale the dispersion of analyst forecasts with either the absolute value of actual earnings per share or the standard deviation of earnings per share.

22

To estimate abnormal returns we use the daily returns on the size-BE/ME portfolio in which the

firm belongs (Fama and French, 1993) as the benchmark returns. We measure firm size as the

natural logarithm of market value of equity (LNMVE) and book to market ratio (BM) at the end

of fiscal quarter t.

Panel A of Table 2 presents the regression results from estimating equation (1). Results

presented in column (1) indicate that higher levels of excitement exhibited by CFOs and CEOs

(PAFF) do not help predict future earnings news. However, the coefficient on NAFF is -2.8316

(one-tailed p-value = 0.023), consistent with expectations that negative affective state is

negatively related to the direction of future earnings news. The coefficient on the lagged BEAT

variable is positive and statistically significant, indicating that a firm is more likely to meet or

beat next analysts’ forecasts if it had done so in the prior quarter. Our results are similar when

we use seasonally lagged quarterly earnings as the earnings benchmark instead of the consensus

analysts’ forecast (see column 3).

We next consider whether the vocal cue–based measures capture information incremental

to that contained in the linguistic tone documented in prior research. As in Tetlock (2007) and

Tetlock et al. (2008) we include the unexpected percentage of negative words (NEGWORDS)

used by the CEOs and CFOs in the conference call as our linguistic measure and expect a

negative coefficient on the percentage of negative words. We use the General Inquirer text

analysis program to compute the linguistic measure and base the expected percentage of negative

words on the percentage of negative words spoken during the entire dialog of the firm’s prior

quarter earnings conference call. We focus on negative words in particular, as prior research by

Tetlock (2007) finds that negative words capture the common variation in General Inquirer word

23

categories more than any other. Our inferences are unaltered if we include the proportion of

positive words as an additional variable in the empirical analyses.

Results of estimating equation (1) after including the linguistic measure NEGWORDS are

presented in columns (2) and (4) of Table 2, Panel A. The inclusion of NEGWORDS in the

empirical specification does not affect the results we previously obtained for the vocal cue

measures or for the control variables. We find a statistically negative relation between negative

words spoken during the conference call and the likelihood of the firm beating analyst earnings

forecast benchmark. However, when we use the lagged seasonal earnings as the benchmark the

coefficient on NEGWORDS is not statistically significant. Overall, our findings suggest that

NAFF possesses incremental information to NEGWORDS in influencing the likelihood that a

firm will meet or beat seasonal earnings benchmark or analysts’ earnings expectation one quarter

ahead.

It is difficult to compare the economic impact of the linguistic and vocal cues measures

based on the coefficients from the logistic regressions because of the different distributional

properties of the two measures. If we transform the independent variables in column 2 (4) to

have zero mean and unit variance, we find that the coefficient on NAFF is -0.18 (-.17) whereas

the coefficient on NEGWORDS is a statistically equivalent -0.14 (-.11). Thus, the predictive

ability of vocal cues appears to be on the same order of magnitude as linguistic measures.

To examine whether vocal cues provide information about future performance for more

than one period, we examine whether PAFF and NAFF predict the likelihood of firms meeting or

beating earnings benchmarks two and three quarters ahead. Results are presented in Panel B,

Table 2. Our results are similar to that reported earlier with one exception – NAFF does not

24

influence the likelihood of meeting or beating earnings expectations two quarters ahead using the

seasonally lagged earnings as benchmark (see column 3).

We report no statistical association between PAFF and the likelihood of meeting or

beating earnings benchmarks in any of the specifications. Collectively, our lack of evidence

supporting a positive association between PAFF and future firm performance may be due to the

lack of variation in PAFF relative to NAFF. For example, in Panel A of Table 1, the standard

deviation of PAFF is 0.0174, which is much smaller than the variation in NAFF of 0.0644. This

lack of variation could be due to all managers exhibiting a constant degree of optimism about the

firm’s prospects or due to the software’s inability to powerfully measure positive affect.17 We do

note that the relative inability of PAFF versus NAFF to predict future accounting performance is

consistent with results in prior research that examined the implications of positive words for

fundamentals. In particular, Tetlock et al. (2008) find that positive words are weaker predictors

of a firm’s future financial performance relative to negative words.

B. Do Market Participants Respond to Managerial Affect?

In this section, we test whether capital market participants, i.e. investors and analysts,

respond to the information in managerial affect. We use contemporaneous market returns and

analyst quarterly earnings forecast revisions to measure investor and analyst responses,

respectively. First, we consider contemporaneous market reactions to vocal cues. We estimate

daily abnormal returns using the returns on the size-BE/ME portfolio in which the firm resides

(Fama and French, 1993) as the benchmark return, and then regress the three-day cumulative

abnormal returns measured around the conference call date (FFCAR(-1,1)) on the vocal cue

measures, NAFF and PAFF. We control for quantitative accounting news contained in the

17 Kothari et al. (2008) provide evidence consistent with the notion that managers leak good news and withhold bad news. If this is the case, it is less likely that a manager will possess positive private information at the time of the conference call that might elicit a positive emotional state.

25

earnings conference call such as the magnitude and sign of unexpected earnings (UEA and

BEATA), determined relative to the last consensus analyst forecast prior to the earnings

conference call. We expect positive coefficients on the quantitative accounting news variables.

We also include NEGWORDS as a control variable for the qualitative linguistic content of the

words spoken by executives, and expect a negative coefficient. We control for size, growth, and

risk that has been shown to be related to market returns (Atiase et al., 2005). We use the natural

logarithm of market value of equity at the end of the current quarter (LNMVE), book to market

ratio (BM) calculated as the book value of shareholders equity at the end of the current quarter

scaled by the market value of equity, and return volatility (VOL) measured as the standard

deviation of daily stock returns over 125 trading days prior to the earnings announcement date as

the empirical proxies for size, growth, and risk respectively. We estimate the following

specification:

FFCAR(-1,1) = λ0 + λ1 PAFF + λ2 NAFF + λ3 A

tUE + λ4A

tBEAT + λ5 LNMVE + λ6 BM

+ λ7 VOL + λ8 NEGWORDS + φ (2) Table 3 presents the results of estimating equation (2). As expected, the coefficient on

unexpected earnings ( AtUE ) is positive and statistically significant suggesting that the market

responds significantly to the extent of news contained in the earnings announcement. The

coefficient on AtBEAT is also incrementally positive suggesting that the market places a

significant premium (about 3.2%) to meeting/beating analysts’ earnings forecasts (Bartov et al.

2002). Consistent with prior evidence in Davis et al. (2007) and Tetlock et al. (2008) we find a

statistically significant negative relation between the percentage of negative words used during

conference calls (NEGWORDS) and contemporaneous returns.

26

With respect to our variables of interest, we do not observe any statistical relation

between positive affect (PAFF) and returns. However, we do observe an unexpectedly positive

coefficient on NAFF. This indicates that investors react positively to more negative affect as

measured by the LVA software. We offer one plausible explanation for this finding. While the

executives in the conference call may exhibit psychological discomfort due to cognitive

dissonance that is captured by the LVA software, it is plausible that the executive may attempt to

counter this dissonance by making attempts to persuade the listeners in the conference call to

alter their opinions (Adams 1961). Dissonance can be countered by eliminating dissonant

cognitions or by adding consonant cognitions. For example, executives may focus on or seek out

information that is consistent with their consonant beliefs or at the extreme trivialize the issue to

eliminate the dissonant cognition that creates discomfort (Simon, Greenberg and Brehm 1995).

To the extent that the NAFF is correlated with the managers’ effort to counter this dissonance

and the investors are persuaded to change their opinion we would observe a positive relation

between NAFF and contemporaneous returns.

To assess the sensitivity of our results and explore whether extreme outliers could be

responsible for the positive relation between NAFF and returns we standardize all the

independent variables by converting them to scaled decile ranks as follows. We rank each

variable into deciles (0 to 9) and then scale the decile values by nine such that each variable will

have values between 0 and 1. We do not convert the BEAT variable into decile ranks because it

is an indicator variable. The results of the return regression (2) using the decile ranks of the

variables are provided in column 2 of Table 3. The inferences are unchanged. A hedge portfolio

of taking long position in stocks within the high NAFF decile and shorting stocks in the low

NAFF decile would approximately return 1.78% surrounding the earnings announcement

27

window. The coefficient of NEGWORDS (-0.0195) indicates that a hedge portfolio of shorting

the firms in the high NEGWORDS decile and taking long positions in the low NEGWORDS

decile would approximately return 1.95% surrounding the earnings announcement window.

We now turn to the question of whether analysts incorporate the signals in the vocal cue–

based measures when revising their earnings forecasts. To test this hypothesis, we use analyst

forecast revision (FREV) as the dependent variable instead of FFCAR(-1,1) in equation (2). Our

expectations on explanatory variables are identical to equation (2). FREV is measured as

previously defined.18

We also add the contemporaneous market reaction FFCAR(-1,1) as an explanatory

variable to control for news in the earnings release that is not quantifiable by our other control

variables. We expect a positive coefficient on FFCAR(-1,1) consistent with our other news

proxies. The results are reported in Table 4. The coefficients on the managerial affect measures,

PAFF and NAFF, are statistically insignificant. This implies that the analysts do not take into

account the information contained in the affect measures when revising their earnings

expectations. In addition, we do not find evidence to support that linguistic tone, i.e., the

percentage of negative words (NEGWORDS) is related to earnings forecast revision surrounding

the earnings announcement. Combining this result with the finding in Tetlock et al. (2008) that

linguistic tone predicts future returns, one can conclude that part of the reason why linguistic

tone predicts future returns is that analysts do not alert investors regarding the future

performance implications of linguistic tone.

To summarize, the contemporaneous reactions by market participants and forecast

revisions surrounding the earnings release provide no support for the hypothesis that investors

18 Observations (128 firms) with no individual analyst forecast revisions during the period are set to zero values. We re-estimated the regression after eliminating the 128 firms and our inferences remain unchanged.

28

and analysts incorporate the information contained in vocal cue measures when revising their

expectations of future earnings and cash flows. In fact, if anything, investors appear to react in a

manner inconsistent with the information contained in vocal cues that capture negative emotional

states.

C. Does Managerial Affect Predict Future Stock Returns?

Based on the evidence thus far, we can conclude that nonverbal vocal cues, in particular

the negative affect measure, has significant information content for future firm performance.

Analysts and investors, however, fail to incorporate the information contained in these vocal

cues. Past accounting literature has documented price drift with respect to quantitative earnings

information (Bernard and Thomas, 1989; Doyle et al. 2006) and with respect to optimistic and

pessimistic language (Demers and Vega, 2007; Engelberg, 2008). To the extent that the market

fails to fully appreciate the implications of the nonverbal signals we investigate here, we should

find a systematic relation between the vocal cues and future stock returns. Therefore, in this

section, we test whether this information subsequently gets incorporated into stock price.

Alternatively stated, we examine whether the affect measures can help predict future abnormal

returns. Because stock returns reflect revisions in future expectation of cash flows and earnings,

we expect that the information contained in the affect measures, although not correctly

incorporated in contemporaneous market returns, will likely be incorporated in future stock

returns when the implications of these measures for future fundamentals subsequently realize.

We test this prediction by using the cumulative abnormal returns over the 180 trading

days following two days after the earnings conference call (FFCAR(2,180)) . We restrict our

analysis to 180 days because we hand-collect returns data from www.yahoo.com and we stopped

data collection on December 14, 2007. Our univariate evidence presented in Panel C of Table 1

29

reveals a negative statistical association between negative affect (NAFF) and future abnormal

returns (FFCAR(2,180)). However, there is no statistically significant correlation between PAFF

and future abnormal returns.

To demonstrate the economic significance of this univariate result, we examine how

stock returns evolve for stocks in various NAFF deciles by plotting the cumulative abnormal

return each day for stocks in NAFF deciles 1, 5 and 10. Figure 2 reveals that NAFF does not

begin separating stock returns until approximately day 55, at which point stocks in NAFF decile

10 begin to decline. By 180 days, the lowest NAFF decile experiences cumulative abnormal

returns of 3.75%, followed by -0.09% for NAFF decile 5, and finally -11.87% for NAFF decile

10. The difference between NAFF deciles 1 and 10 is 15.62%, implying economically

significant differences in returns for future stock returns partitioned by NAFF levels. This

univariate result, however, does not account for other factors known to predict long run returns.

Therefore, we estimate the following multivariate model that controls further for risk and other

factors shown to determine future stock returns:

FFCAR(2,180) = ω0 + ω1 PAFF + ω2 NAFF + ω3 UEA + ω4 BEATA + ω5 LNMVE + ω6 MOM + ω7 VOL + ω8 BM + ω9 NEGWORDS + ξ (3)

We estimate equation (3) using raw independent variables as well as scaled decile ranks of the

independent variables. The advantage of the scaled decile ranks is that it standardizes the

variables to facilitate comparison across coefficients as well as allows for the coefficients to be

interpreted as the hedge abnormal return to a trading strategy of going long on the highest decile

portfolio and short on the lowest portfolio. As in the estimation of contemporaneous returns, the

scaled-decile rank for each variable is obtained by ranking each variable into deciles (0 to 9) and

then dividing the decile values by nine such that each variable will take on values between 0 and

1 (see Abarbanell and Bushee, 1998; Pincus, Rajgopal, and Venkatachalam 2007).

30

The independent variables in model (3) are identical to model (2) used to estimate effects

on contemporaneous returns, except that we also include a momentum variable (MOM) to

control for return momentum (Carhart, 1997). MOM is measured as the cumulative abnormal

returns over a 125 trading-day return window ranging from -2 to -127 days relative to the

earnings announcement date.

Results of estimating equation (3) using raw variables (scaled decile ranks) are presented

in column 1 (column 2) of Table 5. The coefficient on AtUE is positive and statistically

significant in the scaled rank regression (coefficient = 0.096; p < 0.10) capturing the premium to

meeting or beating analyst expectations (Bartov et al., 2002) and/or post–earnings announcement

drift (Bernard and Thomas, 1989; Doyle et al. 2006). In other words, firms that have positive

earnings surprises are likely to enjoy significant post–earnings announcement returns. Consistent

with prior work (Engelberg 2008; Demers and Vega, 2007), we find that the coefficient on

linguistic tone measure (NEGWORDS) is negative (coefficient = -0.14) and statistically

significant, suggesting that the percentage of negative words predict future returns.

Pertinent to this study, we find that the coefficient on NAFF is negative and statistically

significant at the 5% level, whether considered as raw variables or in scaled decile rank form.

The coefficient of -0.090 in column 2 suggests that the hedge returns are 9.0% over a 180 day

trading period for a portfolio strategy that is long on stocks with low negative affect levels and

short on stocks with high negative affect levels.19 Note that long run hedge returns of 9.0% are

far greater than the abnormal return of 1.78% documented for contemporaneous returns,

suggesting the long run returns are not simply reversals of mispricing at the conference call date.

19 The returns do not necessarily imply returns to an executable trading strategy because the positions are taken at different points in time based on the date of the earnings conference call.

31

We find no significant returns for the vocal cue measure that captures positive affect. That is, the

coefficient on PAFF is positive (0.03) but statistically insignificant.

The affect based hedge returns are incremental to, but not statistically different from, the

linguistic based hedge return. We view these results as adding to the evidence in Tetlock et al.

(2008), Demers and Vega (2007) and Engelberg (2008) who show that negative words in various

press reports and earnings press releases predict future returns. However, we caution the reader

that although we document a relationship between negative affect and future returns this does not

establish that we have a profitable trading strategy especially because our sample is restricted to

a cross-section of firms for a single quarter in 2007.

V. Additional Analyses and Robustness Checks

Thus far, we have documented that negative affect exhibited by managers help predict (i)

the likelihood of meeting/beating future earnings expectations and (ii) future returns. In this

section, we explore whether the predictive ability of negative affect varies predictably based on

the firm’s information environment. In particular, we posit that the importance of negative affect

in predicting future firm performance depends on the extent of private information that managers

possess. To the extent that a firm’s information environment is strong, managers are less likely

to possess private information at the time of the earnings release that could underpin an affective

state. That is, stock prices of firms with stronger information environments will promptly reflect

firm-specific information and hence, the incentives for information production and dissemination

as well as the value implication to new information is significantly lower (e.g., Brennan,

Jegadeesh and Swaminathan 1993; Collins, Kothari and Rayburn 1987). To test whether the

predictive ability of negative affect exhibited by managers systematically varies as a function of

the information environment we interact proxies for information environment with NAFF and

32

include them in the empirical specification.20 We consider three proxies for a firm’s information

environment: size, institutional ownership and analyst following. We obtain institutional

ownership from 13(f) filings during the first calendar quarter of 2007 provided the Thomson

Reuters database. For analyst following, we use the number of analysts from the I/B/E/S

database contributing to the consensus earnings estimate.

The results are presented in Table 6. Our main variables of interest are the interaction

terms NAFF*Hsize, NAFF*Hinst, NAFF*Hanal. Hsize, Hinst, Hanal are indicator variables that

take the value of 1 if the value of size, level of institutional ownership and number of analysts is

above the median of their respective in-sample distributions. We predict the coefficient on the

interaction terms to be positive, consistent with the notion that for firms with richer information

environments (i.e., firms with greater size, institutional and analyst following) the negative affect

variable NAFF will have lower predictive content for both future earnings and returns. Our

results are consistent with predictions when we consider the likelihood of firms meeting/beating

analysts earnings expectations. The first column of estimates in Table 6 reveal the coefficients

on NAFF*Hinst and NAFF*Hanal are both positive and statistically significant. In the future

return regression reported in the second column of estimates, none of the coefficients on the

interaction terms in the future returns regressions is statistically significant. Notice that the main

effect is not significant once we include the interaction terms. One plausible explanation for the

insignificant main effects in the returns regressions is collinearity among the interaction terms.

Next, we investigate the robustness of our findings by performing several supplementary

analyses. First, we examine whether the influence of affect for future firm performance is likely

to be greater and easier to detect during the question and answer session than the presentation

20 We restrict our attention to NAFF because we do not find any significant relations for PAFF. Our inferences are unaffected if we consider the interaction terms for PAFF as well.

33

session. Unlike the presentation portion of the conference call where the CEO and/or the CFO

reads a prepared script, the question and answer session is relatively more unstructured in that

the questions from the analysts and investors can be quite probing. Consequently, it may be more

likely that the executive will have to process quantitative and qualitative information before

answering a question, thereby inducing variation in his/her affective state. In unreported results

we estimate equation (1) using vocal cue measures captured during the question and answer

session as opposed to the entire call, after controlling for the predictive ability of the linguistic

tone measure. Untabulated results reveal a coefficient on negative affect variable (NAFF) similar

to the magnitudes reported in Table 2. The coefficient on PAFF, however, is still insignificant.

Our future return results are also very similar to that reported in Table 5.

Second, we examine whether our results are driven by small stocks. Specifically, we

conduct the analyses after removing 35 observations where the stock price is less than $5 per

share on the day prior to the conference call. Consistent with the results previously reported, we

find NAFF is a reliable predictor of future earnings news. Neither investors nor analysts impound

contemporaneously the information contained in the conference call voice signal, but future

returns are associated with NAFF. Hedge returns are approximately 9.1% which is consistent

with the magnitude reported in column 2 of Table 5 and suggests that small stock prices do not

drive the hedge returns we document.

Third, we construct an alternative measure of managerial affect that combines positive

and negative affect into one measure. Specifically, we use the natural logarithm of the ratio of

NAFF to PAFF as the explanatory variable in each model instead of allowing NAFF and PAFF

to enter in separately. All of the inferences remain unchanged from those reported here.

34

Fourth, we incorporate more refined controls for the words spoken during the conference

call. Our current control for linguistic content is the variable NEGWORDS, based on prior work

by Tetlock (2007) and Tetlock et al. (2008). NEGWORDS is derived from words listed in the

“Negativ” category, one of 77 word categories of the General Inquirer’s Harvard-IV-4

classification dictionary. The General Inquirer also contains dictionaries for positive affect (the

“PosAff” category), negative affect (the “NegAff” category), as well as uncertainty (the “If”

category). To control for affective states and managerial uncertainty that might manifest

linguistically we measure the unexpected percentage of total words in each of these additional

categories in a manner identical to NEGWORDS, and include them in addition to NEGWORDS

in each of our regressions (results not tabled). The inclusion of these additional variables does

not alter any of our inferences qualitatively or quantitatively. This is not surprising, as each

additional category is statistically correlated with NEGWORDS, suggesting the presence of

NEGWORDS provides sufficient control for variation in each of these additional categories.

VI. Conclusions

Our study is the first to provide evidence on the role of vocal cues in predicting future

firm performance and stock returns. We posit that vocal cues from conversations with executives

during earnings conference calls convey information about their affective states that, in turn, help

predict future profitability and returns. We find that higher levels of negative affect, as

operationalized via higher levels of cognitive dissonance determined by the proprietary LVA

software, conveys bad news about future firm performance. More negative affect is associated

with lower likelihood of meeting or beating earnings expectations and correspondingly, lower

future returns. This relation obtains even after controlling for quantitative information and

managers’ linguistic style or word usage during conference calls.

35

We do not observe, however, a contemporaneous negative relation between negative

affect and analyst forecast revisions surrounding the conference call. This result, coupled with

the finding that the cognitive dissonance measure is related to future performance, suggests that

analysts under react to the negative information that can be gleaned from vocal cues. A puzzling

finding in the paper is a positive relation between negative affect and contemporaneous stock

market returns.

An important implication of our paper is that information gleaned from nonverbal cues

conveyed during communications between managers and shareholders may be quite useful in

resource allocation and portfolio decisions. Thus our evidence adds to the body of research in

social psychology that finds an incrementally important role for nonverbal cues in

communication. Although our evidence is somewhat preliminary as we base our findings on a

small sample of firms, the results offer encouragement to extend our inquiry in various ways.

First, identifying which particular business transactions or events (such as restructurings, new

customer agreements, restatements etc.) elicit positive or negative managerial affect can

potentially lead to more powerful tests and further our understanding about how vocal cues can

inform investors in a capital market setting. Second, our analysis could be utilized in other

settings, such as how vocal cues from depositions and communications by the Federal Reserve

chairman might assist in forecasting interest rates. Such an analysis would complement current

work analyzing the predictive ability of the chairman’s linguistic style (Piger, 2006; Bligh and

Hess 2007). Finally, technological advances have increased the availability of video in addition

to audio. We plan to explore facial expression as yet another channel of nonverbal managerial

communication, as even thin slices of nonverbal facial cues are known to improve judgments

about affective states (Ambady and Rosenthal, 1992).

36

Figure 1 – Screen Shot of LVA analysis of audio files

37

Figure 2 – Cumulative Abnormal Returns by NAFF deciles

-14.00%

-12.00%

-10.00%

-8.00%

-6.00%

-4.00%

-2.00%

0.00%

2.00%

4.00%

6.00%

2 7 12 17 22 27 32 37 42 47 52 57 62 67 72 77 82 87 92 97 102 107 112 117 122 127 132 137 142 147 152 157 162 167 172 177

DAY RELATIVE TO EVENT DATE

CU

MU

LA

TIV

E A

BN

OR

MA

L R

ET

UR

N

Lowest NAFF (DECILE 1) Medium NAFF (DECILE 5) Highest NAFF (DECILE 10)

38

References Abarbanell, J., and B. Bushee. 1998. Abnormal Stock Returns to a Fundamental Analysis Strategy. The

Accounting Review 73: 19-45.

Adams, J.S. 1961. Reduction of Cognitive Dissonance by Seeking Consonant Information. Journal of Abnormal Social Psychology 62: 74-78.

Akerlof, G. and William Dickens. 1982. The Economic Consequences of Cognitive Dissonance. The

American Economic Review 72 (3)307-319.

Ambady, N., and R. Rosenthal. 1993. Thin Slices of Expressive Behavior As Predictors of Interpersonal Consequences: A Meta-analysis. Psychological Bulletin 111 (2):256-274.

Antweiler, W., and M.Z. Frank. 2004. Is All That Talk Just Noise? The Information Content of

Internet Stock Message Boards? The Journal of Finance 59 (3): 1259-1294. Argentesi, E. , H. Lutkepohl, and M. Motta. 2007. Acquisition of information and share prices: An

empirical examination of cognitive dissonance. Working paper, University of Bologna and European University Institute, Florence.

Arnold, M.B. 1960. Emotion and Personality. New York: Columbia University Press.

Atiase, R., Li, H., Suppattarakul, S., and Tse, S. (2005. Market Reaction to Multiple Contemporaneous

Earnings Signals: Earnings Announcements and Future Earnings Guidance. Review of Accounting Studies 10: 497–525.

Bartov, E., Givoly, D., and C. Hayn. 2002. The Rewards to Meeting or Beating Analysts’ Earnings

Forecasts. Journal of Accounting and Economics 33 2): 173–204. Bernard, V. and Thomas, J. 1989. Post Earnings Announcement Drift: Delayed Price Response or Risk

Premium? Journal of Accounting Research (Supplement): 1–36.

Bettman, J.R. and B.A. Weitz. 1983. Attributions in the Board Room: Causal Reasoning in Corporate Annual Reports. Administrative Science Quarterly 28 (2): 165-183.

Bligh, M. and G. Hess. 2007. The Power of Leading Subtly: Alan Greenspan, Rhetorical Leadership,

and Monetary Policy. The Leadership Quarterly 18: 87–104. Block, S. 2000. Pay close attention to conference calls for hints about company’s future. USA Today,

Money, 3B, September 15. Brennan, M., N. Jagadeesh, and B. Swaminathan. 1993. Investment analysis and the adjustment of

stock prices to common information. Review of Financial Studies 6: 799-824. Bushee, B., M. Jung and G. Miller. 2007. Capital Market Consequences of Conference Presentations.

Working paper, University of Pennsylvania and Harvard University.

39

Bushee, B. D. Matsumoto and Miller, G. 2004. Managerial and Investor Responses to Disclosure and Regulation: The Case of Reg FD and Conference Calls. The Accounting Review 79 (3): 617–643.

Cacioppo, J. and G. Bernston. 1994. Relationship Between Attitudes and Evaluative Space: A

Critical Review, with Emphasis on the Separability of Positive and Negative Substrates. Psychological Bulletin 115 (3): 401-423.

Carhart, M. 1997. On Persistence in Mutual Fund Performance. Journal of Finance 52: 57–82. Collins, D., S. Kothari and J. Rayburn. 1987. Firm size and information content of prices with respect

to earnings. Journal of Accounting and Economics 9: 111-138. Coval, J.D. and T. Shumway. 2001. Is sound just noise? Journal of Finance 56 (5): 1887-1910. Davis, A., Piger, J., and Sedor, L. 2007. Beyond the Numbers: An Analysis of Optimistic and

Pessimistic Language in Earnings Press Releases. Working Paper. University of Oregon and University of Washington.

Demers, E. and Vega, C. 2007. Soft Information in Earnings Announcements: News or Noise?

Working paper. INSEAD. Diener, E., and Emmons, R.A. 1985. The independence of positive and negative affect. Journal of

Personality and Social Psychology 47:1105-1117. Doyle, J., R. Lundholm, and M. Soliman. 2006. The Extreme Future Stock Returns Following I/B/E/S

Earnings surprise. Journal of Accounting Research 44 (3): 849-887. Ekman, P., and W.V. Friesen. 1974. Detecting deception from body or face. Journal of Personality

and Social Psychology 29(3): 288–298. Elliot, A. and Devine, P. 1994. On the Motivational Nature of Cognitive Dissonance: Dissonance as

Psychological Discomfort. Journal of Personality and Social Psychology 67 (3): 382-394. Engelberg, J. 2008. Costly Information Processing: Evidence from Earnings Announcements.

Working paper, Northwestern University. Fama, E. and French, K. 1993. Common Risk Factors in the Returns on Stocks and Bonds. Journal of

Financial Economics 33: 3–56. Festinger, L. 1957. A Theory of Cognitive Dissonance. Stanford, CA: Stanford University Press. Forgas, J.P. 2001. The handbook of affect and social cognition. Mahwah, NJ: Eribaum. Frankel, R., Johnson, M., and Skinner, D. 1999. An Empirical Examination of Conference Calls as a

Voluntary Disclosure Medium, Journal of Accounting Research 37:133–150.

40

Frankel, R., Mayew, W., and Sun, Y. 2007. Do Pennies Matter? Investor Relations Consequences of Small Negative Earnings Surprises. Working paper. Duke University and Washington University – St. Louis.

Frijda, N. 1988. The Laws of Emotion. American Psychologist 43 (5): 349–358. Frijda, N. 1993. Mood, emotion episodes, and emotions. In M. Lewis and J.. Haviland (Ed.),

Handbook of emotions. New York: Guilford Press. Graham, J., Harvey, C., and Rajgopal, S., 2005, The economic implications of corporate financial

reporting, Journal of Accounting and Economics 40, 3-73. Harmon-Jones, E. 2000. Cognitive Dissonance and Experienced Negative Affect: Evidence that

Dissonance Increases Experienced Negative Affect Even in the Absence of Aversive Consequences. Personality and Social Psychology Bulletin 26(12): 1490-1501.

Heddad, D. and Ratley, R. 2002. Investigation and Evaluation of Voice Stress Analysis Technology.

Final Report Submitted to the U.S. Department of Justice. Johnson, T. 2004. Forecast dispersion and the cross-section of expected returns. Journal of Finance

59: 1957-1978. Juslin, P., and Luakka, P. 2003. Communication of Emotions in Vocal Expression and Music

Performance: Different Channels, Same Code? Psychological Bulletin 129 (5): 770–814. Juslin, P. and Scherer, K. 2005. Vocal Expression of Affect. Harrigan, J., Rosenthall, R. and Scherer,

K. The New Handbook of Methods in Nonverbal Behavior Research, Oxford: Oxford Press, 65–135.

Kappas, A., Hess, U., and K.R. Scherer. 1991. Voice and emotion. In: B. Rimé and R.S. Feldman,

Editors, Fundamentals of Nonverbal Behavior. Cambridge and New York: Cambridge University Press.

Kasznik, R. and M.F. McNichols. 2002. Does Meeting Earnings Expectations Matter? Evidence from Analyst Forecast Revisions and Share Prices. Journal of Accounting Research 40 (3): 727-759.

Kothari, S.P, Shu, S. And Wysocki, P. 2008. Do Managers Withhold Bad News? Working paper,

MIT and Boston College. Lazarus, R. 1991. Progress on a Cognitive-Motivational-Relational Theory of Emotion. American

Psychologist 46(8): 819-834. Li, F. 2006. Do Stock Market Investors Understand the Risk Sentiment of Corporate Annual Reports?

Working paper. University of Michigan. Marler, P. 1977. The evolution of communication. In How animals communicate. (ed. T.A. Sebeok),

Bloomington, IN: Indiana University Press.

41

Matsumoto, D., Pronk, M., and Roelofsen, E. 2007. Conference Call Disclosures and Firm Performance: An Empirical Analysis of the Length and Content of Earnings-Related Conference Calls. Working paper. University of Washington.

Mayew, W. 2008. Evidence of Management Discrimination Among Analysts During Earnings

Conference Calls. Journal of Accounting Research 46(3): 627-659. Mehrabian, A.1971. Silent Messages. Belmont, CA: Wadsworth. Mehrabian, A., and M. Weiner 1967. Decoding of inconsistent communications. Journal of personality

and social psychology 6(1):109-114: Ou, J.A., and S.H. Penman. 1989. Financial Statement Analysis and the Prediction of Stock Returns.

Journal of Accounting and Economics, 11(4): 295-329. Piger, J. 2006. Is All That Talk Just Noise? Monetary Trends August 2006, Federal Reserve Bank of

St. Louis. Pincus, M., S. Rajgopal, and M. Venkatachalam. 2007. The accrual anomaly: The international

evidence. The Accounting Review 82 (1):169-203. Prentice, R. 2003. Enron: A Brief Behavioral Autopsy. American Business Law Journal, 40 Am.

Bus. L.J. 417. Rees, L. 2005. Abnormal Returns from Predicting Earnings Thresholds.oeseman. Review of

Accounting Studies 10(4): 465-496. Roseman, I.J. 1984. Cognitive determinants of emotions: A structural theory. In P. Shaver (Ed.),

Review of Personality and Social Psychology 5:11-36. Newbury Park, CA: Sage. Scherer, K.R. 2003. Vocal communication of emotion: A review of research paradigms. Speech

Communication 40(1-2): 227-256 Scherer, K.R., London, H., and J. Wolf. 1973. The voice of confidence: Paralinguistic cues and

audience evaluation. Journal of Research in Personality 7: 31–44. Simon, L., J. Greenberg, and J. Brehm. 1995. Trivialization: The Forgotten Mode of Dissonance

Reduction. Journal of Personality and Social Psychology 68(2): 247-260. Skinner, D. 2003. Should Firms Disclose Everything to Everybody? A Discussion of “Open vs. Closed

Conference Calls: The Determinants and Effects of Broadening Access to Disclosure.” Journal of Accounting and Economics 34: 181–187.

Sommers, M. 2006. Evaluating Voice-Based Measures for Detecting Deception. The Journal of

Credibility Assessment in Witness Psychology 7 (2): 99–107. Tasker, S. 1998. Bridging the Information Gap: Quarterly Conference Calls as a Medium for

Voluntary Disclosure. Review of Accounting Studies 3 (1 & 2): 137–167.

42

Tetlock, P. 2007. Giving Content to Investor Sentiment: The Role of Media in the Stock Market.

Journal of Finance 62 (3), 1139-1168. Tetlock, P., Saar-Tsechansky, M., and Macskassy, S. 2008. More Than Words: Quantifying Language

to Measure Firm’s Fundamentals. Journal of Finance, 63 (3): 1437-1467. Tomkins, S. 1962. Affect Imagery Consciousness: The Positive Affects, Volume 1. New York:

Springer. Walbott, H.G., Ricci-Bitti, p., and E. Banninger-Huber. 1986. Non-verbal reactions to emotional

experiences. In Experiencing emotion: a cross-cultural study (ed. K.R. Scherer, H.G. Walbott, and A.B. Summerfield). Cambridge, UK: Cambridge University Press.

Zuckerman, M., Amidon, M.D., Bishop, S.E., and S.D. Pomerantz. Face and tone of voice in the

communication of deception. Journal of Personality and Social Psychology 43(2) 347-357

43

Table 1

Descriptive Statistics and Sample Characteristics This table reports descriptive statistics and sample characteristics for 615 fiscal year 2006 4th quarter earnings conference calls occurring between January 1 and March 31, 2007. PAFF is positive affect measured as the percentage of CEO and CFO spoken audio during the entire conference call with Emotion Level scores above the critical value of 110 as measured by LVA; NAFF is negative affect measured as the percentage of CEO and CFO spoken audio during the entire conference call with Cognition Level scores above the critical value of 120 as measured by LVA; ROA is return on assets measured as income before extraordinary items at the beginning of the quarter; STDROA is the standard deviation of ROA over the prior four fiscal quarters; ASSETS is total assets in millions at fiscal quarter end, NEGWORDS is the percentage of negative words spoken by the CEO and CFO during the entire conference call less the percentage of negative words as measured from all the dialog on the firm’s prior quarter earnings conference call; FREV is analyst forecast revisions for one quarter ahead earnings; FDISP is the standard deviation of analyst per share earnings forecasts that comprise the consensus earnings estimate; FFCAR(i,j) is daily abnormal returns cumulated over day i through j, where expected returns are derived from the size-BE/ME portfolio to which the firm belongs; UEA is unexpected earnings measured as the difference between actual I/B/E/S earnings per share and I/B/E/S analyst consensus median earnings per share scaled by price per share to days before the conference call; UEE is unexpected earnings measured as the difference between reported income before extraordinary items and the same value four quarters ago, scaled by the market value of equity two days before the conference call; BEATA equals one if UE A is greater than one, zero otherwise; BEATE equals one if UE E is greater than one, zero otherwise; LNMVE is the natural logarithm of the market value of equity measured in millions at fiscal quarter end; MOM is momentum measured as FFCAR(-127,-2); BM is the ratio of the book value of equity to the market value of equity at fiscal quarter end; VOL is the stock return volatility measured as the standard deviation of daily stock returns over the period (-127,-2) relative to the conference call date. Panel A reports descriptive statistics for the sample observations. Panel B reports industry concentrations for the sample observations. Panel C reports correlations between positive and negative emotional states and sample firm characteristics.

Panel A: Descriptive statistics

Variable N Mean Std Dev Median Min Max

PAFF 615 0.1028 0.0174 0.1039 0.0347 0.1610 NAFF 615 0.1663 0.0644 0.1632 0.0256 0.3372 ROA 615 0.0037 0.0538 0.0115 -0.2750 0.1312 STDROA 615 0.0154 0.0261 0.0065 0.0001 0.1760 ASSETS 615 9,196 37,312 1,219 10 70,7121 NEGWORDS 615 0.0020 0.0042 0.0020 -0.0136 0.0196 FREV 615 -0.0012 0.0043 0.0000 -0.0313 0.0084 FDISP 615 0.0324 0.0471 0.0200 0.0000 0.4800

FFCAR(-1,1) 615 0.0043 0.0690 -0.0001 -0.3055 0.2738 FFCAR(2,180) 615 -0.0533 0.3579 -0.0403 -2.6180 0.9685

AtUE 615 -0.0004 0.0105 0.0005 -0.0712 0.0216

AtBEAT 615 0.7073 0.4554 1.0000 0.0000 1.0000

AtUE 1+ 615 -0.0002 0.0069 0.0003 -0.0367 0.0201

AtBEAT 1+ 615 0.6602 0.4740 1.0000 0.0000 1.0000

EtUE 615 0.0008 0.1477 0.0016 -1.2432 3.0520

EtBEAT 615 0.6098 0.4882 1.0000 0.0000 1.0000

EtUE 1+ 615 -0.0089 0.1672 0.0011 -4.0896 0.1541

EtBEAT 1+ 615 0.5951 0.4913 1.0000 0.0000 1.0000

LNMVE 615 7.2724 1.5348 7.2060 3.8778 11.2854 MOM 615 0.0235 0.2164 0.0200 -1.0523 1.1964 BM 615 0.4232 0.2551 0.3897 -0.1074 1.2354 VOL 615 0.0197 0.0087 0.0183 0.0072 0.0466

44

Panel B: Industry Composition

Industry Sample Firms

All Compustat Firms

N % N % Chemicals 11 1.79 411 1.82 Computers 83 13.5 2,908 12.85 Extractive 23 3.74 904 3.99 Financial 90 14.63 3,050 13.48 Food 9 1.46 401 1.77 Insurance/RealEstate 42 6.83 2,306 10.19 Manf:ElectricalEqpt 19 3.09 767 3.39 Manf:Instruments 42 6.83 1,062 4.69 Manf:Machinery 12 1.95 544 2.40 Manf:Metal 6 0.98 473 2.09 Manf:Misc. 3 0.49 214 0.95 Manf:Rubber/glass/etc 3 0.49 371 1.64 Manf:TransportEqpt 11 1.79 340 1.50 Mining/Construction 10 1.63 622 2.75 Pharmaceuticals 45 7.32 900 3.98 Retail:Misc. 32 5.2 933 4.12 Retail:Restaurant 7 1.14 286 1.26 Retail:Wholesale 11 1.79 781 3.45 Services 66 10.73 2,064 9.12 Textiles/Print/Publish 31 5.04 845 3.73 Transportation 38 6.18 1,388 6.13 Utilities 19 3.09 658 2.91 Not assigned 2 0.33 405 1.79 Total 615 100.00 22,633 100.00

45

Panel C: Pearson Correlations among emotion levels and firm characteristics (significance levels in parentheses)

PAFF NAFF

NAFF -0.03 (0.51)

ROA -0.04 (0.32)

-0.08 (0.06)

STDROA -0.05 (0.19)

0.09 (0.03)

ASSETS 0.01 (0.79)

-0.11 (0.01)

NEGWORDS -0.02 (0.66)

0.01 (0.72)

FREV -0.06 (0.16)

-0.02 (0.60)

FDISP 0.03 (0.51)

-0.02 (0.56)

FFCAR(-1,1) 0.01 (0.81)

0.07 (0.10)

FFCAR(2,180) 0.03 (0.44)

-0.09 (0.03)

AtUE -0.05

(0.25) -0.05 (0.23)

AtBEAT -0.02

(0.57) -0.01 (0.71)

AtUE 1+ 0.00

(0.97) -0.13 (0.00)

AtBEAT 1+

-0.02 (0.54)

-0.11 (0.01)

EtUE 1+ -0.04

(0.38) -0.01 (0.73)

EtBEAT -0.03

(0.40) -0.04 (0.32)

EtUE 1+ 0.01

(0.87) 0.03

(0.51) E

tBEAT 1+ -0.01

(0.74) -0.10 (0.01)

LNMVE -0.01 (0.79)

-0.16 (0.00)

MOM -0.07 (0.10)

-0.03 (0.44)

BM 0.05 (0.21)

-0.03 (0.47)

VOL 0.03 (0.48)

0.14 (0.00)

46

Table 2 Estimations of the association between affect and future earnings news c

This table reports logistic regression estimates of managerial affect’s (PAFF and NAFF) ability to predict future quarterly earnings news. PAFF is positive affect measured as the percentage of CEO and CFO spoken audio during the entire conference call with Emotion Level scores above the critical value of 110 as measured by LVA. NAFF is negative affect measured as the percentage of CEO and CFO spoken audio during the entire conference call with Cognition Level scores above the critical value of 120 as measured by LVA. Panel A reports the coefficients and summary statistics from four regressions: Indicator variables for whether the firm met or exceeded analyst based ( A

tBEAT 1+) or seasonal earnings based ( E

tBEAT 1+) earnings targets one-quarter ahead regressed on positive and

negative affect, both without and with control for linguistic content (NEGWORDS). NEGWORDS is the percentage of negative words spoken by the CEO and CFO during the entire conference call less the percentage of negative words as measured from the entire dialog on the firm’s prior quarter earnings conference call. All regressions include control variables for current quarter values of the dependent variable, firm size, book-to-market, momentum, and analyst forecast revision revisions and dispersion (see text and Table 1 for details). Panel B reports re-estimation of Panel A regressions while using two and three quarter ahead earnings news and controlling for linguistic content. White-corrected standard errors in the presence of heteroskedasticity are presented in parentheses below the coefficient estimates. ***, **, * Significant at .01, .05 and .10 level, respectively, in a two-tailed test (one-tailed when predicted). Panel A: Logistic regression of one period ahead future earnings news on affect measures and control variables

jtBEAT 1+

(j=A, Analyst Forecast Benchmark)

jtBEAT 1+

(j=E, Lagged Seasonal Earnings Benchmark)

(1) (2) (3) (4)

Intercept -1.1079 -0.9730 -0.8422 -0.7372

(0.7821) (0.7844) (0.7517) (0.7554)

PAFF -1.8946 -2.1552 5.1336 4.9819

(5.2478) (5.2312) (5.1658) (5.2171)

NAFF -2.8316** -2.8273** -2.7286** -2.7321**

(1.4137) (1.4250) (1.3793) (1.3789) j

tBEAT 0.9259*** 0.9051*** 0.7331*** 0.7277***

(0.1940) (0.1953) (0.1822) (0.1826)

FREV 38.4415* 39.3456* 92.6505* 92.6893*

(21.6688) (22.0235) (50.6846) (50.6602)

FDISP -2.0410 -1.7780 -4.5088** -4.2537**

(1.8790) (1.9199) (1.9409) (1.9382)

LNMVE 0.2802*** 0.2755*** 0.1978*** 0.1931***

(0.0625) (0.0628) (0.0615) (0.0618)

MOM -0.1089 -0.1236 0.9891** 0.9739**

(0.4148) (0.4123) (0.4096) (0.4128)

BM -0.1701 -0.1609 -1.1257*** -1.1314***

(0.3650) (0.3715) (0.3646) (0.3672)

NEGWORDS -33.0770* -26.9758 (21.2604) (21.2830) N 615 615 615 615

Pseudo R2 8.51% 8.80% 11.45% 11.64%

47

Panel B: Logistic regression of two and three period ahead future earnings news on affect measures and control variables

AtBEAT 2+

AtBEAT 3+

EtBEAT 2+

EtBEAT 3+

(j=A, Analyst Forecast

Benchmark)

(j=E, Lagged Seasonal Earnings Benchmark)

(1) (2) (3) (4)

Intercept -0.6223 2.1027*** -2.1242*** 0.7770

(0.8037) (0.8124) (0.7659) (0.7592)

PAFF 3.5587 -8.4757 9.6378** 0.1492

(5.1188) (5.1671) (4.9754) (4.9742)

NAFF -2.6403** -3.8887*** 0.4542 -1.9191*

(1.4324) (1.4234) (1.3565) (1.3209) j

tBEAT 0.7051*** 0.6539*** 0.5063*** 0.5123***

(0.1948) (0.1972) (0.1768) (0.1783)

FREV -23.3354 -15.2356 48.5071 40.7767*

(19.5556) (18.6596) (24.3278) (24.4119)

FDISP -0.8605 -2.8720 0.6624 2.2757

(1.9439) (1.8956) (1.8843) (2.0348)

LNMVE 0.1929*** 0.0544 0.1923*** 0.0066

(0.0622) (0.0603) (0.0596) (0.0588)

MOM 0.7600* 0.4245 0.8307** 0.6532

(0.3978) (0.4733) (0.4186) (0.4385)

BM -0.8588** -1.0476*** -0.4794 -1.0794***

(0.3742) (0.3751) (0.3499) (0.3603)

NEGWORDS -39.7327** -27.4634* -21.7478 -47.6264** (21.8869) (20.8446) (20.5294) (20.9158) N 613 614 610 610

Pseudo R2 6.63% 5.72% 5.89% 5.16%

48

Table 3 Estimation of the association between affect and contemporaneous stock returns

This table reports ordinary least squares regression estimation of the association between managerial affect (PAFF and NAFF) and the contemporaneous stock market reaction (FFCAR(-1,1)). The dependent variable, FFCAR(-1,1), is the daily abnormal stock return cumulated over the three days centered on the conference call date, where expected returns are derived from the size-BE/ME portfolio to which the firm belongs. PAFF is positive affect measured as the percentage of CEO and CFO spoken audio during the entire conference call with Emotion Level scores above the critical value of 110 as measured by LVA. NAFF is negative affect measured as the percentage of CEO and CFO spoken audio during the entire conference call with Cognition Level scores above the critical value of 120 as measured by LVA. Controls for earnings news, linguistic content, size, momentum, book-to-market and volatility are included in the regression (see text and Table 1 for details). The first column estimates the regression using raw independent variables and the second column estimates using decile ranks of independent variables. White-corrected standard errors in the presence of heteroskedasticity are presented in parentheses below the coefficient estimates. ***, **, * Significant at .01, .05 and .10 level, respectively, in a two-tailed test (one-tailed when predicted).

Raw Independent Variables

Decile Rank Independent

Variables

(1) (2)

Intercept -0.0019 -0.0186

(0.0298) (0.0148)

PAFF 0.0957 0.0052

(0.1548) (0.0083)

NAFF 0.0799* 0.0178**

(0.0414) (0.0083)

UEA 0.7775* 0.0730***

(0.5667) (0.0132)

BEATA 0.0310*** -0.0016

(0.0066) (0.0084)

LNMVE -0.0025 -0.0012

(0.0020) (0.0099)

BM -0.0177 -0.0145*

(0.0111) (0.0086)

VOL -0.5030 -0.0085

(0.4727) (0.0102)

MOM -0.0050 -0.0043

(0.0145) (0.0094)

NEGWORDS -1.3859*** -0.0195** (0.5645) (0.0080) N 615 615

Adjusted R2 10.57% 13.69%

49

Table 4 Estimation of the association between affect and analyst one quarter ahead forecast revisions

This table reports ordinary least squares regression estimation of the association between managerial affect (PAFF and NAFF) and analyst one quarter ahead earnings forecast revisions (FREV). The dependent variable, FREV, is the measured as difference between the median forecast for quarter t+1 earnings issued after and before the quarter t earnings announcement date, scaled by price two days before the earnings announcement. The median forecast before (after) the quarter t earnings announcement is measured as the last (first) forecast of all individual analysts issuing forecasts during the 90 day period prior to (after) the quarter t announcement date. PAFF is positive affect measured as the percentage of CEO and CFO spoken audio during the entire conference call with Emotion Level scores above the critical value of 110 as measured by LVA. NAFF is negative affect measured as the percentage of CEO and CFO spoken audio during the entire conference call with Cognition Level scores above the critical value of 120 as measured by LVA. Controls for earnings news, linguistic content, contemporaneous stock market reaction, size, momentum, book-to-market and volatility are included in the regression (see text and Table 1 for details). White-corrected standard errors in the presence of heteroskedasticity are presented in parentheses below the coefficient estimates. ***, **, * Significant at .01, .05 and .10 level, respectively, in a two-tailed test (one-tailed when predicted).

FREV

Intercept 0.0040

(0.0020)

PAFF -0.0087

(0.0087)

NAFF -0.0009

(0.0023)

UE 0.0446

(0.0381)

BEAT 0.0005

(0.0004)

LNMVE -0.0002*

(0.0001)

BM -0.0023***

(0.0009)

VOL -0.1066***

(0.0328)

MOM 0.0016

(0.0010)

FFCAR(-1,1) 0.0129***

(0.0033)

NEGWORDS 0.0251

(0.0380) N 615

Adjusted R2 14.95%

50

Table 5 OLS estimations of the association between affect variables and future stock returns

This table reports ordinary least squares regression estimation of the association between managerial affect (PAFF and NAFF) and future stock returns (FFCAR(2,180)). The dependent variable, FFCAR(2,180), is the daily abnormal stock return cumulated over the period beginning 2 days through 180 days after the conference call, where expected returns are derived from the size-BE/ME portfolio to which the firm belongs. PAFF is positive affect measured as the percentage of CEO and CFO spoken audio during the entire conference call with Emotion Level scores above the critical value of 110 as measured by LVA. NAFF is negative affect measured as the percentage of CEO and CFO spoken audio during the entire conference call with Cognition Level scores above the critical value of 120 as measured by LVA. Controls for earnings news, linguistic content, size, momentum, book-to-market and volatility are included in the regression (see text and Table 1 for details). The first column estimates the regression using raw independent variables and the second column estimates using decile ranks of independent variables. White-corrected standard errors in the presence of heteroskedasticity are presented in parentheses below the coefficient estimates. ***, **, * Significant at .01, .05 and .10 level, respectively, in a two-tailed test (one-tailed when predicted).

Raw Independent Variables

Decile Rank Independent

Variables (1) (2)

Intercept 0.0587 0.0252

(0.1886) (0.0867)

PAFF 0.8719 0.0314

(0.8448) (0.0459)

NAFF -0.4334** -0.0902**

(0.2282) (0.0469)

UEA 2.4048 0.0962*

(3.1880) (0.0696)

BEATA 0.0264 0.0027

(0.0386) (0.0441)

LNMVE -0.0061 -0.0148

(0.0124) (0.0574)

MOM 0.1639** 0.0679

(0.0758) (0.0519)

BM -0.1320* -0.1100**

(0.0788) (0.0521)

VOL -1.2691 0.0007

(2.4632) (0.0616)

NEGWORDS -12.6546*** -0.1418*** (3.3147) (0.0454) N 615 615

Adjusted R2 6.77% 5.22%

51

Table 6 Estimations of the moderating effects of information environment

on the association between affect and future earnings news and returns This table reports how proxies for strong information environments of firm size (Hsize), institutional holding (Hinst), and analyst following (Hanal) moderate the association between negative affect (NAFF) and both one-period ahead analyst based earnings news (BEATA) and future stock returns (FFCAR(2,180)). PAFF is positive affect measured as the percentage of CEO and CFO spoken audio during the entire conference call with Emotion Level scores above the critical value of 110 as measured by LVA. NAFF is negative affect measured as the percentage of CEO and CFO spoken audio during the entire conference call with Cognition Level scores above the critical value of 120 as measured by LVA. BEATA is an indicator variable for whether the firm met or exceeded analyst based earnings targets one-quarter ahead. FFCAR(2,180), is the daily abnormal stock return cumulated over the period beginning 2 days through 180 days after the conference call, where expected returns are derived from the size-BE/ME portfolio to which the firm belongs. Hsize is an indicator variable that equals 1 when firm size is above the sample median, zero otherwise. Hinst is an indicator variable that equals 1 when the level of institutional holding is above the sample median, zero otherwise. Hanal is an indicator variable that equals 1 when the level of analyst following is above the sample median, zero otherwise. The logistic regression on the determinants of future earnings news includes control variables for current quarter values of the dependent variable, firm size, book-to-market, momentum, and analyst forecast revision revisions and dispersion. The ordinary least squares regression of future stock returns controls for earnings news, linguistic content, size, momentum, book-to-market and volatility (see text and Table 1 for details). White-corrected standard errors in the presence of heteroskedasticity are presented in parentheses below the coefficient estimates. ***, **, * Significant at .01, .05 and .10 level, respectively, in a two-tailed test (one-tailed when predicted). Raw Independent variables A

tBEAT 1+

(1)

Decile Rank Independent Variables FFCAR(2,180) (2)

Intercept -0.6663 Intercept -0.0196

(0.9523) (0.1022) PAFF -0.9518 PAFF 0.0269 (5.2788) (0.0459) NAFF -4.8144*** NAFF -0.0195 (1.6760) (0.0814) NAFF*Hsize -1.1529 NAFF*Hsize -0.0694 (1.5737) (0.0734) NAFF*Hinst 2.9522*** NAFF*Hinst -0.0389 (1.0529) (0.0514) NAFF*Hanal 2.7166** NAFF*Hanal -0.0296 (1.1723) (0.0592)

AtBEAT 0.8646*** UEA

0.0979 (0.1982) (0.0700) FREV 46.2938 BEATA

0.0054 (25.7682) (0.0442) FDISP -2.1044 LNMVE 0.0603 (1.9917) (0.0878) LNMVE 0.2121 MOM 0.0658 (0.0964) (0.0522) MOM -0.1248 BM -0.1070** (0.4167) (0.0530) BM -0.0606 VOL 0.0102 (0.3863) (0.0647) NEGWORDS -27.4888 NEGWORDS -0.1447***

(21.6533) (0.0451)

N 615 N 615

Pseudo R2 10.68% Adjusted R2 5.58%