almost-cheap talk in earnings press...

39
Almost-cheap talk in earnings press releases: Textual tone and managerial equity incentives Özgür Arslan-Ayaydin a,1 , Kris Boudt b,c,2 , James Thewissen b,d,3 July 29, 2014 ABSTRACT We are the first to examine whether managers’ equity-based incen- tives affect the tone of the qualitative information contained in earn- ings press releases and how these incentives affect the abnormal stock return at the earnings announcement. We analyze the earnings press releases of S&P1500 firms between 2004 and 2011 and find evidence that managers inflate the tone of their earnings press releases as the sensitivity of their stock-based compensation to the share price in- creases. We show that this effect cannot be explained by differences in past performance, industry and year fixed effects. Measures of tone in earnings press releases thus do not simply reflect the economic events of the firm but also managers’ incentives to maximize their stock-based compensation. In addition, as theoretically predicted by the Almost-cheap talk model of Kartik (2009), we find that managers’ tone inflation in earnings press releases is not completely free. We de- fine a two-regime-threshold model, where in the first regime, the price sensitivity of managers’ stock-based compensation is small and, in the second regime, managers’ compensation has a significant price sensi- tivity. We find that the price of stocks belonging to this second cat- egory reacts proportionally less to the tone of earnings press releases when managers have more equity incentives. KEYWORDS: Cheap talk, Earnings press release, Managerial portfo- lio, Market efficiency, Price sensitivity AUTHORS I NFO a Department of Finance, University of Illi- nois at Chicago, United States b Solvay Business School, Vrije Universiteit Brussel, Belgium c Faculty of Economics and Business, V.U.University of Amsterdam, The Nether- lands d Accounting, Finance and Insurance, Katholieke Universiteit Leuven, Belgium 1 [email protected] 2 [email protected] 3 [email protected] PAPER I NFO AMS CLASSIFICATION: G14, G30, G32 This research was carried out thanks to financial support in the form of a grant from the Intercollegiate Center for Management Science, the Dutch Science Foundation and the Hercules Foundation (Project No. AKUL/11/02). We thank Angela Davis, Lalitha Naveen and Dalia Marciukaityte for helpful comments and suggestions. Corresponding author: James Thewissen, I.C.M. Fellow, Katholieke Universiteit Leuven, Naamsestraat 69, 3000 Leuven, Belgium 1

Upload: others

Post on 19-Jun-2020

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

Almost-cheap talk in earnings press releases:Textual tone and managerial equity incentivesI

Özgür Arslan-Ayaydina,1, Kris Boudtb,c,2, James Thewissenb,d,3

July 29, 2014

ABSTRACT

We are the first to examine whether managers’ equity-based incen-tives affect the tone of the qualitative information contained in earn-ings press releases and how these incentives affect the abnormal stockreturn at the earnings announcement. We analyze the earnings pressreleases of S&P1500 firms between 2004 and 2011 and find evidencethat managers inflate the tone of their earnings press releases as thesensitivity of their stock-based compensation to the share price in-creases. We show that this effect cannot be explained by differencesin past performance, industry and year fixed effects. Measures of tonein earnings press releases thus do not simply reflect the economicevents of the firm but also managers’ incentives to maximize theirstock-based compensation. In addition, as theoretically predicted bythe Almost-cheap talk model of Kartik (2009), we find that managers’tone inflation in earnings press releases is not completely free. We de-fine a two-regime-threshold model, where in the first regime, the pricesensitivity of managers’ stock-based compensation is small and, in thesecond regime, managers’ compensation has a significant price sensi-tivity. We find that the price of stocks belonging to this second cat-egory reacts proportionally less to the tone of earnings press releaseswhen managers have more equity incentives.

KEYWORDS: Cheap talk, Earnings press release, Managerial portfo-lio, Market efficiency, Price sensitivity

AUTHORS INFOa Department of Finance, University of Illi-nois at Chicago, United Statesb Solvay Business School, Vrije UniversiteitBrussel, Belgiumc Faculty of Economics and Business,V.U.University of Amsterdam, The Nether-landsd Accounting, Finance and Insurance,Katholieke Universiteit Leuven, Belgium1 [email protected] [email protected] [email protected]

PAPER INFO

AMS CLASSIFICATION: G14, G30, G32

IThis research was carried out thanks to financial support in the form of a grant from the Intercollegiate Center forManagement Science, the Dutch Science Foundation and the Hercules Foundation (Project No. AKUL/11/02).We thank Angela Davis, Lalitha Naveen and Dalia Marciukaityte for helpful comments and suggestions.Corresponding author: James Thewissen, I.C.M. Fellow, Katholieke Universiteit Leuven, Naamsestraat 69, 3000Leuven, Belgium

1

Page 2: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

1 Introduction

The use of performance-based executive compensation schemes has increased significantly inthe last decade (see Murphy, 1999). Although these compensation schemes are clearly intendedto align managers’ and shareholders’ incentives, there is ample empirical evidence that stock-based compensation contracts increase managers’ incentives to manipulate earnings numbersin accounting statements (see e.g. Bergstresser and Philippon, 2006; Burns and Kedia, 2006;Gaver, Gaver, and Austin, 1995; Ke, 2003).1 Yet, the disclosures made by managers in earningspress releases are not limited to quantitative information. In fact, prior research shows thatthere is incremental information content in the narrative section of earnings press releases andthat investors positively react to the tone in earnings announcements (Davis, Piger, and Sedor,2012; Davis and Tama-Sweet, 2012; Henry, 2008). This evidence suggests that managers canopportunistically influence stock prices by inflating the tone of their narrative disclosures toincrease their compensation. However, it remains an open question whether managers’ equity-based incentives also influence their misreporting behaviour of qualitative information.

This paper is the first to examine whether management’s equity incentives affect the tone ofthe qualitative information contained in earnings press releases and how these incentives affectthe stock price reaction to the tone. This question is related to prior research on the presenceof tone manipulation in earnings press releases before important accounting events, such asmergers and acquisitions or equity issuance. Huang, Teoh, and Zhang (2013) find that the tonein earnings press releases is, on average, more positive when firms are issuing new equity orundertaking mergers and acquisitions, and more negative when granting stock options. Similarly,Davis and Tama-Sweet (2012) argue that managers act strategically in choosing the narrativeoutlets to describe firm performance, e.g., Management, Discussion & Analysis (MD&A) vs.earnings press releases. They argue that, because investors react more strongly to the informationdisclosed at the time of the earnings press release relative to the market reaction at the time ofthe 10-K or 10-Q filings (which includes MD&A), managers use less pessimistic language andmore optimistic language in earnings press releases relative to the MD&A. Both studies thusexplain their result by the fact that managers’ underlying objective is to maximize the expectedutility of their portfolio by positively influencing the stock price.

We extend these findings and investigate if managers inflate the tone with the objective tomaximize their private benefits, e.g., maximize the value of their stock-based compensation. Wefollow Coles, Daniel, and Naveen (2006) and Daniel, Li, and Naveen (2013) in measuring man-agers’ equity incentives as the sensitivity of the stocks and options in the managerial portfoliosto changes in the firm’s stock price. Since the tone of financial disclosures positively affectsstock prices (see e.g. Davis et al., 2012; Henry, 2008), it is natural to expect that managers witha more price sensitive portfolio have an incentive to inflate the tone in the qualitative statementsof their narrative disclosures in order to positively influence the price reaction to the earningspress release.

We refer to such inflation as “cheap talk”. This is analogous to the theory on games of cheap

1Throughout the paper, we refer to managers as the group of CEO, CFO and the three highest-compensated namedexecutive officers. One exception is that in Subsection 4.2.1 we do the the analysis for the CEO and the remainderof the team, separately.

2

Page 3: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

talk pioneered by Crawford and Sobel (1982) and Green and Stokey (2007). In the “cheap talk”model, there is strategic transmission of private information between an informed sender and anuninformed receiver, whereby the sender can lie without direct costs. There exist, however, var-ious indirect costs that temper managers’ over-optimism in narrative disclosures, such as repu-tation or potential litigation with shareholders. To account for these costs, Kartik (2009) extendsthe cheap talk theory to describe the case of small lying costs as “almost-cheap talk”. Cheap andalmost-cheap talk typically arise when the information is non-binding and non-verifiable, likethe tone in earnings press releases.

Our sample covers the earnings press releases written by managers of the S&P 1500 firmsbetween 2004 and 2011. To estimate the tone of the language used in earnings press releases,we adopt a content-analysis approach and measure sentiment as the (standardized) average ofthe difference in the frequency of positive and negative words identified using three differentwordlists: 1) the general DICTION wordlist used in Davis et al. (2012), 2) the wordlist based onearnings press releases and developed in Henry (2008) and 3) the wordlist based on 10-K filingsdeveloped by Loughran and McDonald (2011).

We find that the higher the sensitivity of the manager’s compensation to the share price ofthe firm, the more optimistic the manager is in his earnings press release. This effect cannot beexplained by differences in past performance, industry and year fixed effects. In addition, wefind that the CEO’s portfolio sensitivity to stock price is the primary driving force behind theobserved tone inflation. We thus conclude that measures of tone in earnings press releases donot simply reflect the economic events of the firm but also managers’ incentives to maximizetheir compensation and limit their risk exposure.

Based on this result, we expect the well-documented positive reaction to tone in earnings pressreleases decreases when firm managers have more equity incentives and model the price reactionto the tone as a two-regime-threshold model. The first regime corresponds to firms for whichthe price sensitivity of managers’ stock-based compensation is small such that investors do notdiscount the potential tone inflation. In the second regime, managers’ stock-based compensationhas a significant price sensitivity and investors therefore react proportionally less to the tone ofthe press release. The transition between the two regimes is governed by a threshold that weestimate by non-linear least squares. We interpret investors’ anticipation as one of the costs tothe cheap talk of the manager. These findings are consistent with the Almost-cheap talk modelof Kartik (2009) in which managers interpret the investors’ anticipation as a cost to their cheaptalk.

Our analysis also shows that there is a clear size effect regarding the relationship betweenthe price sensitivity of the managerial stock-based compensation, the tone in the earnings pressreleases and the market impact of the release. We find that, vis-à-vis large firms’ announce-ments, the tone of earnings press releases of small-cap firms signals more valuable informationconcerning managers’ expectations about future performance.2 This is as expected since thecompensation structure depends on the size of the firm (Cyert, Kang, and Kumar, 2002; Hub-bard and Palia, 1995; Jensen and Murphy, 1990; Murphy, 1985, 1986) and the size of the firmis also a proxy for the information asymmetry (Atiase, 1980, 1985). However, despite being

2Implicit in the information efficiency definition of Latham (1986) is the notion that an announcement is informativeif prices change upon publication of the information.

3

Page 4: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

more informative, we also find cheap talk to be more frequent in the earnings press releases ofsmall-cap firms. Accordingly, we show that investors start discounting the tone of smaller firmsat relatively lower levels of managerial compensation.

This study contributes to the literature on the information transmission between managers andinvestors in three ways. First, despite this widespread concern over earnings management, re-search has not addressed the question of how cheap talk might erode the quality and credibilityof the narrative sections of earnings press releases. While quantitative disclosures are subject toGAAP enforced by independent auditors or the SEC monitoring of periodic filings, words aremuch more elastic than numbers in conveying an impression. For the textual component of earn-ings press releases there exist only a few general guidelines that advise managers on how to bestreport information on future performance. Earnings press releases are required to be “accurateand complete so as not to mislead” (Trautmann and Hamilton, 2003). The National Investor Re-lations Institute and the Financial Executives Institute also recommend that “managers presentin earnings press releases a reasonably balanced perspective of operating performance”. Addi-tionally, New York Stock Exchange rules require that press releases place news in the “properperspective” and that managers avoid “overly optimistic forecasts, exaggerated claims and un-warranted promises” (NYSE). Such disclosures are thus harder to regulate or to litigate against(Henry, 2008) and offer to managers an opportunity to more subtly manipulate market partici-pants’ perceptions of future firm performance. Our results are thus of importance to investorsand regulators interested in the credibility of financial disclosures.

Second, while prior research primarily focuses on tone as a signal of managers’ private infor-mation about future corporate performance (Davis et al., 2012; Davis and Tama-Sweet, 2012;Henry, 2008), it generally ignores managers’ incentives to inflate the tone of corporate disclo-sures. We show that managers seem to inflate the tone as the sensitivity of their compensationincreases and that caution should thus be taken when interpreting the tone of earnings pressreleases as a pure signal for managers’ private information. Third, in contrast to Davis et al.(2012), Davis and Tama-Sweet (2012) and Henry (2008) who define investors’ reaction to thetone of earnings press releases through a linear model, we show the presence of a clearly non-linear relationship between the abnormal price reaction to the earnings press release and the toneof the press release. Our model not only incorporates managers’ opportunistic use of languagein earnings press releases but also shows that investors anticipate the inflated tone and (at leastpartially) discount managers’ cheap talk.

The rest of the paper is organized as follows. In Section 2, we discuss prior literature onmanagers’ compensation and quantitative misreporting. In Section 3, we describe the variablesand the data selection method used. Section 4 describes our results and a robustness analysis.Finally, Section 5 concludes and suggests directions for further research.

2 Literature review and hypothesis development

Information transmission from managers to shareholders and its interaction with the stock val-uation is a central topic of investigation in financial economics. We focus on the manager’sprivate benefits of influencing the stock’s valuation and refer to Baker and Wurgler (2002) foran alternative theory in which the manager aims at exploiting stock mispricing to issue shares at

4

Page 5: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

high prices and repurchase them at low prices, or to Salomon (2012) for a theory in which moreincentivized managers have already spun the good news before the earnings press release, whichinduces a smaller price reaction at earnings announcement.

This paper focuses on the impact of the price sensitivity of managerial equity incentives onthe tone of earnings press releases. Those incentives often result from the firm’s actions toreward and stimulate managerial good performance by granting managers option and sharesat a discount compared to the market price. However there is abundant evidence provided bythe literature on the distorting effects this type of performance-based compensation can haveon managerial decisions (see among others, Bergstresser and Philippon, 2006; Peng and Roell,2003). Most studies evaluate this in a flow setting and study the impact of the yearly compensa-tion on the decision of the manager that year. Coles et al. (2006) take a different angle and studythe price sensitivity of the managerial portfolio that can be seen as a stock variable aggregat-ing the annually awarded compensations in terms of stock and option compensation programs.They find that sensitivity of managerial compensation to stock price has a strong influence onmanagers’ value-critical decisions, specifically, those derived from both investment policy anddebt policy.

We examine the tone in narrative disclosures as a potentially new channel through which thecompensation of managers can influence their behaviour. We argue that managers inflate thetone for at least two reasons. First, they are risk averse and want to maximize the firm’s valueby minimizing the potential negative value effects that their tone can have on their portfoliovalue (Markowitz, 1952). Second, from a short term value maximization perspective, managershave an interest in diffusing information such that investors are overoptimistic about the futureprospects of the firm and therefore overvalue the stock price around the earnings announce-ments, leading to a higher stock-based compensation. At the same time, we hypothesize thatinvestors are aware of this inflation and will react proportionally less to the tone of the earningspress release when the price sensitivity of the manager’s stock-based compensation is higher.Such a feedback effect is consistent with the Almost-cheap talk theory of Kartik (2009) that wereview in Subsection 2.1. In Subsection 2.2 we discuss the main finding of prior literature of thepredictive power of textual sentiment for the future stock price performance.

The remaining sections develop our hypotheses regarding the following two questions: (i)whether a manager whose compensation is stock-dependent inflates the textual tone in voluntarydisclosures to maximize his wealth and (ii) how do investors react to the tone of earnings pressreleases, given that they know managers’ incentives.

2.1 The cheap talk and almost-cheap talk model for tone inflation by managers

Cheap talk arises in a situation of information transmission between an informed sender andan uninformed receiver, whereby the sender can lie without direct costs. Crawford and Sobel(1982) and Green and Stokey (2007) pioneered the theory on games of cheap talk, which hasrecently been generalized by Kartik (2009) to account for lying costs. When lying costs aresmall, Kartik (2009) describes the situation as an almost-cheap talk game between the informedsender and uninformed receiver.

In Kartik (2009)’s Almost-cheap talk model, inflated language naturally arises in a situation ofstrategic communication between an informed sender and an uninformed receiver. Kartik (2009)

5

Page 6: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

shows that the sender will always claim to be a higher type than he would under completeinformation. The receiver, however, is not systematically deceived by the sender’s languageinflation, because he recognizes that the sender is inflating his messages and adjusts his decisionaccordingly. While this equilibrium may appear paradoxical at first glance, the key point is thatif the sender were to tell the truth (i.e. send a message with no inflation), the receiver wouldinfer that his type is in fact lower than it actually is, since the receiver expects messages to beinflated. There is thus an inescapable inefficiency for the sender in equilibrium.

A practical example that Kartik (2009) gives for his model is the one of inflated stock recom-mendations by analysts at brokerage firms that have an underwriting relationship with the stock.Lin and McNichols (1998) and Michaely and Womack (1999) show that analysts at brokeragefirms that have an underwriting relationship with a stock tend to provide inflated recommenda-tions on that firm. The literature discusses two basic factors that determine an affiliated analyst’sincentives: first, the presence of an underwriting relationship creates a pressure to induce clientsto buy the stock. Second, there are compensation and future reputation rewards for more ac-curate recommendations. In terms of Kartik’s model, the more an analyst’s compensation isdetermined by accurate assessments, the higher is the intensity of lying costs. The position ofthe firm manager writing the earnings press release and being exposed to a price sensitivity ofhis portfolio is similar to the situation of the analyst.3

2.2 The information content of textual sentiment for firm valuation

Managers acquire private information about the firm’s future expected cash flows and risk throughtheir proximity to operating activities. There exists strong evidence that the market uses quali-tative information from earnings press releases to infer the manager’s private information aboutthe firm’s prospects and value. For instance, Davis et al. (2012) document the immediate effectof the tone of earnings press releases on the perception of investors about its future performance.They show that the three-day cumulative abnormal return around the earnings announcement in-creases with the tone of earnings press releases. Henry (2008), Demers and Vega (2010) andPrice, Doran, Peterson, and Bliss (2012) also conclude that the tone of earnings press releasesis significantly positively correlated with short window contemporaneous returns around thedate that the disclosures are made even after controlling for a firm’s financial information andearnings surprises. Engelberg (2008) finds that, over longer horizons, qualitative earnings infor-mation embedded in Dow Jones News Service stories about a firm’s earnings announcementshas additional predictability for asset prices beyond the predictability of hard financial infor-mation, such as the unexpected earnings surprise, return-on-assets or analyst dispersion. Heexplains that qualitative information has a higher processing cost than quantitative informationover longer horizons, which explains why it diffuses slowly into asset prices. This evidence,either over short or longer periods, supports the fact that qualitative information contained inearnings releases holds information that is related to both current and future firm profitability

3Alternative models exist on the idea of strategic information transmission between a sender and a receiver. Al-though the assumptions differ, the main results remain the same: while the message sent by the sender is biased,at the equilibrium, the receiver is able to reduce (partly or completely, depending on the assumptions) this bias.See for instance the Signal Jamming theory between investors and managers modeled by Stein (1989) or themisreporting bias model defined by Fischer and Verrecchia (2000).

6

Page 7: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

that affects stock prices, above and beyond hard, financial information.Early research on the qualitative information of earnings press releases mainly interprets the

tone as an unbiased signal of a manager’s private information about future corporate perfor-mance and generally ignores the managerial incentives to inflate the tone of corporate disclo-sures (see e.g. Davis et al., 2012; Davis and Tama-Sweet, 2012; Henry, 2008). It is only recentlythat increasing evidence shows that managers can intentionally affect the optimistic languagein earnings press releases. For instance, Huang et al. (2013) provide evidence that managersmanipulate investors’ perceptions to hype a stock before important events. They find that thetextual tone in earnings press releases is, on average, more positive when firms are issuing newequity or undertaking mergers and acquisitions, and more negative when granting stock op-tions. Similarly, Davis and Tama-Sweet (2012) argue that managers act strategically in choosingthe narrative outlets to describe firm performance, e.g., Management, Discussion & Analysis(MD&A) vs. earnings press releases. They argue that, because investors react more strongly tothe information disclosed at the time of the earnings press release relative to the market reactionat the time of the 10-K or 10-Q filings (which includes MD&A), managers use less pessimisticlanguage and more optimistic language in earnings press releases relative to the MD&A. How-ever, in both cases, a potential underlying objective is for managers to maximize their portfoliovalue by positively influencing the stock price. We thus extend this research and investigate ifmanagers inflate the tone with the objective to maximize their private benefits, e.g., maximizethe value of their portfolio.

2.3 Textual tone and sensitivity of managers’ equity-based compensation to thestock price

Stock-based compensation aims at aligning managers’ and shareholders’ interests. The basicpremise is that by making managers’ compensation positively dependent on the firm’s stockprice, they will maximize shareholders’ long run value. After a series of accounting scandalssuch as Enron, it has become clear that stock-based compensation can also fail in aligning theinterests and lead to self-opportunistic behavior of the manager.

Several authors have provided evidence on managers’ strategic misreporting behaviour ofearnings numbers with the attempt to maximize their performance-based compensation. Forinstance, Gaver et al. (1995), Ke (2003), Peng and Roell (2003), Burns and Kedia (2006) andBergstresser and Philippon (2006) all show that firms with more incentivized managers havehigher levels of earnings management and that management trade an unusually large amount ofstocks and options in periods of overstated earnings. Cheng and Warfield (2005) make a similarpoint and provide evidence that stock-based compensation induces managers to meet or justbeat analysts’ forecasts. Gao and Shrieves (2002) also find that earnings management intensityis significantly and positively associated with the sensitivity of the CEO’s compensation packageto stock price.

We examine a more subtle form of misreporting to investors which is the inflation of the tonein earnings press releases. Language in earnings press releases may be particularly susceptibleto the influence of managers’ stock-based incentives because the choice of language is relativelyunconstrained and difficult to verify ex post when compared to audited financial statements and

7

Page 8: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

explicit earnings numbers. Managers are obviously constrained from making statements that arefalse, but positioning factual statements in a more positive way is less likely to cause regulatoryintervention. In addition, given the positive linkage in the tone contained in earnings pressreleases and firm value over the short and long term, one can expect that managers who aredirectly rewarded using a variety of stock-based compensation plans, such as stock or stock-option grants, will engage in the inflation of the tone in voluntary disclosures to maximize theirwealth. Our first hypothesis is as follows:

Hypothesis 1. The more sensitive the managerial equity-based compensation is to the firm’sstock price, the higher the manager’s incentive for tone inflation in earnings press releases willbe.

The manager is expected to balance the benefits of tone inflation with the costs of doing so.There exist various costs that temper managers’ over-optimism in narrative disclosures. Prob-ably the most important cost a manager would bear is its loss of credibility. Credibility hasbeen shown to be an important factor for investors in reacting to financial disclosures. For in-stance, Gordon, Henry, Peytcheva, and Sun (2008) study whether management credibility isassociated with a less negative market reaction to restatements and find that greater managementcredibility is associated with a less negative market reaction at the time of the restatement. Liti-gation risk also constitutes an important cost to misreporting in qualitative disclosures. Rogers,Van Buskirk, and Zechman (2011) evidence that managers’ use of optimistic language increaseslitigation risk. They show that plaintiffs target optimistic statements in their lawsuits and that,controlling for a firm’s economic condition, sued firms have unusually linguistically optimisticearnings announcements. The fear of litigation based on voluntary disclosures that are ex postoverly optimistic can thus reduce managers’ tendency towards over-optimism in voluntary out-lets.4 Similarly, Baginski, Demers, Wang, and Yu (2011) argue that optimism in financial dis-closures is also costly because it signals high industry profitability, which in turn may encourageentry into the industry by potential competitors.

2.4 Feedback effects between tone inflation and investors’ reaction to textual tone

We expect that the tone in an earnings press release is predominantly informative about the futureprospects of the firms and that stock prices therefore react positively to positive tone. This is astandard prediction confirmed in the literature (Davis et al., 2012; Davis and Tama-Sweet, 2012;Engelberg, 2008; Henry, 2008). We first replicate this result and test to what extent investorsreact to the tone in earnings press releases.

Hypothesis 2. Ceteris paribus, the tone in earnings press releases is positively associated withabnormal returns on the firm’s stock price around the earnings announcement date.

An essential part in the almost-cheap talk theory of Kartik (2009) is the prediction that the in-vestor anticipates the manager’s incentive to send a too optimistic signal through the tone of his

4The firm may suffer what stock traders call a "liar’s discount", which is the discounting of a firm’s stock price belowthat of its competitors when analysts refuse to trust a firm’s management after its prior voluntary disclosure isexposed to misleading (King, 1988).

8

Page 9: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

report and therefore will react proportionally less to the tone in the earnings press release. Weexpect this to happen when managers’ stock-based compensation as a ratio of his total compen-sation is sufficiently high. The perceived importance of the stock market effect for the manager’stotal portfolio is likely to depend on how large the price sensitivity is compared to the manager’stotal compensation.

Hypothesis 3. For a sufficiently high level of managers’ compensation sensitivity, the firm’sstock price reacts proportionally less to the textual tone when the managers’ compensation sen-sitivity to the stock price increases.

This hypothesis is related to the notion that investors are aware of managers’ strategic biasesand discount what they view as opportunistically motivated. For instance, Balsam, Batov, andMarquard (2002) investigate the stock price reaction to quarterly earnings news for a sampleof firms for which there is ex post evidence of earnings management. They find evidence of anegative association between the degree of accruals management and the stock price reactionaround the 10-Q filing date.

However, contrarily to Kartik (2009), we cannot argue that investors fully adjust for the cheaptalk bias in earnings press releases. While the long-standing assumption in both economicsand accounting is that investors fully adjust for known biases in reported information, recentevidence shows that investors underweight the impact that managerial bias has on informationand are largely unable to correct their expectations. Smith (2012) experimentally documentsthat investors are behaving contrary to economic theory as they are not able to fully adjust forknown biases in managerial communications, although they know the quantitative amount ofthe manager’s bias. Drawing on insights from psychology, he argues that individuals have atendency to believe plausible information, even when other evidence suggests the information isunreliable or false (Evans, Barston, and Pollard, 1983; Markovits and Nantel, 1989). His resultoccurs because, as individuals comprehend the information, they overly focus on the content ofthe information and are not sufficiently attentive to indicators of the veracity of the information(Burgoon, Blair, and Strom, 2008; Gilbert, Tafarodi, and Malone, 1993). Based on Smith (2012),we expect that investors, at least partly, correct for managers’ cheap talk in earnings pressreleases.

We further distinguish between positive and negative tone. The prior literature on manage-ment earnings forecasts finds that bad news forecasts are consistently informative (i.e., credible),while good news forecasts tend to be discounted (see, e.g., Forst, 1997; Hutton, Miller, and Skin-ner, 2003; Jennings, 1987). In the case of the tone in earnings press releases, because cheap talkis more likely to be optimistic, we expect rational investors to anticipate the bias and to discountthe cheap talk in the positive tone of earnings press releases accordingly. We hypothesize:

Hypothesis 4. Earnings press releases that convey good news are less informative to the stockmarket than earnings press releases conveying negative news.

9

Page 10: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

2.5 Information asymmetry, managers’ tone inflation and investors’ reaction

The presence of information asymmetry between the manager and the investors is a sine qua noncondition for the existence of tone inflation. It is natural to expect that the higher the informationasymmetry, the higher the inflation of tone in earnings releases. This hypothesis cannot be di-rectly tested, as the magnitude of information asymmetries are not directly measurable. Instead,we will focus on the firm’s market capitalization. Following the "firm-size related differentialinformation hypothesis" of Atiase (1980), the level of information asymmetry decreases withfirm size since large firms tend to be more mature, produce, disclose and disseminate more pri-vate information, as well as to receive more attention from the market and regulators. Basedon this hypothesis, Atiase (1985) shows that the amount of unexpected information conveyedto the market by actual earnings reports is inversely related to firm capitalization and explainshis result by the differential in information asymmetry of firms with different market capitaliza-tion. While Atiase (1980, 1985) focuses on actual earnings numbers, we extend Hypothesis 2 bytesting for a size-effect in investors’ reaction to the tone in earnings press releases. We expect,ceteris paribus, the tone in earnings press releases to be positively associated with abnormalreturns on the firm’s stock price around the earnings announcement date and this effect to bestronger for small-cap firms.

Smaller firms have a higher level of information asymmetry, which gives their manager moreopportunity to inflate the tone of their qualitative disclosures. Following Atiase (1985), the earn-ings press releases of larger firms have less informative value as these firms are more scrutinizedby market participants. This increased scrutiny of market participants leads to an increasedlikelihood in shareholder litigation in the case of over-optimism in earnings press releases or agreater credibility cost. We thus expect managers of small-cap firms to have more incentives toinflate the tone in earnings press releases to increase their compensation. This is consistent withRichardson (2000) who shows that the greater the information asymmetry between managementand its shareholders, the more likely the firm is to manage accruals and earnings. Accordingly,we expect investors to see through this tone inflation and to start discounting the tone of earningspress releases of small firms at lower levels of stock-based compensation (as compared to totalcompensation).

3 Methods and Data

To test our first hypothesis, we define a linear least-squares model in which we regress thetone on managers’ stock-based compensation. In the model we control for variables explainingthe intrinsic informative value of the tone for firm performance and variables proxying for theinformation asymmetry and earnings uncertainty. These variables are described in Subsection3.5.

For Hypothesis 2, we define a threshold least squares model in which we regress the cu-mulative abnormal return around the earnings announcement date on the tone and test whetherinvestors discount for cheap talk as managers’ compensation sensitivity to stock price increases.

In the estimation, we exploit the longitudinal structure of the data by using panel least squaresestimation in which we take industry and year fixed effects into account. The panel approach

10

Page 11: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

requires the slope coefficients to be the same for all entities. This might seem as a restrictiveassumption, as some managers might not engage in tone inflation. This might be true, but itbiases us against finding an effect. If we do find confirmation of our hypothesis, then we canconclude that the effect is probably even greater for those firms for which the price sensitivity ofmanagers’ stock-based compensation induces tone inflation by the management.

3.1 Sample selection

Our initial sample consists of the S&P 1500 constituents over the period 2004-2011 for whichcompensation data is available on ExecuComp. While the earnings press releases are availableon a quarterly basis, the information on the manager’s compensation is only available at a yearlyfrequency on ExecuComp. We will work with average sentiment per year and do our analysis atthe yearly frequency.

The S&P 1500 universe covers the S&P small-cap, mid-cap and large-cap indices. We do nottrack exactly the composition of those three subuniverses but proxy it by doing the analysis onthe time varying universe of small-cap, mid-cap and large-cap stocks obtained through an equalsplit of the universe using the end of year market capitalization of the previous year.

Our sample consists of publicly-listed firms, which, since 2003, are required to disclose earn-ings releases to the SEC on Form 8-K within four days of the earnings announcement (SEC,2003). The initial sample of 57,194 firm-years shrinks to 44,917 after requiring that the 8-Kreport is available on the EDGAR website. We further require the 8-K to be filed within thefour trading days after the earnings announcement date. We then select 8-K filings that includean earnings release identified by Item Code 2.02.5 We aggregate the data at a yearly frequencyand obtain 7,409 firm-year observations. After merging with ExecuComp, the sample contains6,143 yearly observations. The specific COMPUSTAT, CRSP and IBES data requirements forthe control variables described below shrink the sample further to 2,795 firm-year observationsbetween 2004 and 2011. Each variable is defined in Table 1.

Table 2 presents summary statistics of the sample obtained and the different variables used.For the moment, let us focus on the distribution of the firm size. The size of firms included inour sample is quite large with an average of $10.127 bil. If we focus on the size tertiles, we seethat the average market capitalization starts at $0.621 bil. to increase to $2.487 bil. in tertile 2and $27.267 bil. in tertile 3. This heterogeneity reflects the variety of firms included in the S&P1500.

3.2 Managers’ equity incentives

We follow the approach in Core and Guay (2002) and Coles et al. (2006) in measuring themanagers’ equity-based incentives as the dollar change in the value of executives’ stock and

5A further technical detail is that we only keep the text within the tag "<TYPE> EX 99.1". To ensure also thatthe sample includes only earnings press releases, we read all electronic files with a size of at least 10 kilobytesand eliminate the files containing conference call announcements or other non-earnings-related announcements.We also consider firms for which the total number of earnings press releases in a fiscal year equals four. Theseconditions allow us to have an accurate estimation of the average tone in a year. Following Davis et al. (2012),we also remove files where the total number of words in an earnings press release is inferior to 100 words. Moredetails on the parsing method of earnings press releases used are given in the Appendix.

11

Page 12: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

Table 1: Variable definitions

Panel A – Compustat/CRSP/IBES itemsItem Expected sign Item name Compustat/CRSP item #

Aj,t Total assets COMPUSTAT #6Yj,t Income before extraordinary items COMPUSTAT #18CSHOj,t Common shares outstanding COMPUSTAT #25Bj,t Book value of equity COMPUSTAT At − LtPj,dt Price on day d, year t CRSPRetj,dt Return of stock j on day d of year t CRSPRetM,dt S&P500 return on day d of year t CRSPConsensusj,t Mean forecast made by analysts’ (last) forecast for firm j in year t IBESForecasti,j,t Forecast made by analyst i for firm j in year t IBES

Panel B – Dependent variablesVariable Variable Description Definition

PWj,qt Total number of positive words in the earnings press release of quarter q offiscal year t

NWj,qt Total number of negative words in the earnings press release of quarter q offiscal year t

Tonej,qt (in %) Tone in the earnings press release of quarter q of fiscal year t of firm j Tonej,qt = 100 · PWj,qt−NWj,qt

TWj,qt

Tonej,t (in %) Average net tone in the earnings press release of fiscal year t of firm j Tonej,t =∑4

q=1NetSentj,qtCAR[−1,+1]j,qt Cumulative abnormal return for firm j for the [-1,1] trading day window, where

t = 0 is the earnings announcement day of fiscal quarter q. αj,qt and βj,qt areestimated following the market model that spans the [-275,-31] time window,where t = 0 is the day of the quarterly earnings announcement.

∑1d=−1 [E[Rj,dt]− αj,qt − βj,qt · E[RM,dt]]

CAR[−1,+1]j,t Average of CARj,qt over fiscal year t 1Q

∑Qq=1CARj,qt

Panel C – Managerial portfolio sensitivity to stock priceVariable Expected sign Variable Description Definition

∆j,t + Price sensitivity of the stock-based compensation of the team of executives offirm j stock price (in $ mil.)

Follows the Coles et al. (2006) method

∆Stck,j,t + Delta of the stock portfolio of firm’s j team of executives’ in year t (in $ mil.) Follows the Coles et al. (2006) method∆Opt,j,t + Delta of the option portfolio of firm’s j team of executives’ in year t (in $ mil.) Follows the Coles et al. (2006) method∆j,t + Relative price sensitivity of the stock-based compensation of the team of exec-

utives of firm j stock price (in %).)

∆j,t

TotCompj,t

∆Stck,j,t + Relative price sensitivity of the managerial stock portfolio of the team of ex-ecutives of firm j stock price (in %).)

∆Stck,j,t

TotCompj,t

∆Opt,j,t + Relative price sensitivity of the managerial option portfolio of the team of ex-ecutives of firm j stock price (in %).)

∆Opt,j,t

TotCompj,t

Panel D – Control variablesVariable Expected Sign Variable Description Definition

MCj,t + Market capitalization on the last day d of the fiscal year t (in $ mil) Pj,dt · CSHOj,tBTMj,t - Book-to-market ratio of firm j for fiscal year t Bj,t/MCj,tROAj,t + Income before extraordinary items for firm j of fiscal year t, divided by Aj,t at

the beginning of the yearYj,t/Aj,t

αj,t + Estimated intercept of the market model that spans the [-275,-31] time window,where t = 0 is the last day of fiscal year t

E[Rj,t] = αj,t + βj,t · E[RM,t]

CAR[−30, 0]j,t + Cumulative abnormal return for firm j for the [-30,0] trading day window,where t = 0 is the last day of fiscal year t

∑0d=−31 [E[Rj,dt]− αj,t − βj,t · E[RM,dt]]

FEj,t + Analysts’ forecast error for stock j in year t, standardized by the stock priceon the last day of fiscal year t

|Actualj,t−Consensusj,t|Pj,t

NOAj,t - Number of analysts’ forecasts for stock j in year t Sum of individual forecasts

V olumej,t - Share turnover over fiscal year t, standardized by the number of shares out-standing on the last day of fiscal year t

∑250d=0 V OLj,dtCSHOj,t

DISPj,t - Dispersion of analysts’ (last) forecasts for stock j in fiscal year t, standardizedby the stock price on the last day of fiscal year t

σForecast,i,j,tPj,t

σROA,j,t - Standard deviation of ROA over the 4 years preceding the end of fiscal year t√

14 ·∑4

i=1

(ROAj,t−i − ̂ROAj,t−i

)βj,t - Systematic risk of stock j in year t, based on the market model that spans the

[-252,-31] time windowE[Rj,t] = αj,t + βj,t · E[RM,t]

Y eary Year y is defined as the fiscal year during which the earnings press released ismade public

Inds Industry s is defined as the first two digits of the Global Industry ClassificationStandard

12

Page 13: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

Table 2: Summary statistics for all firms and by size tertile

All firms Small-Cap Mid-Cap Large-Cap

Mean Median Mean Median Mean Median Mean Median

Tone in earnings press releasesTonej,t 0.045 -0.032 -0.222 -0.194 0.101 -0.044 0.257 0.226

Price sensitivity of managers’ stock-based compensation in $mil∆j,t−1 1.954 0.504 0.329 0.180 1.001 0.513 4.529 1.406∆Stck,j,t−1 1.330 0.155 0.199 0.064 0.597 0.156 3.194 0.355∆Opt,j,t−1 0.623 0.265 0.130 0.089 0.404 0.301 1.335 0.882

Relative price sensitivity of managers’ stock-based compensation∆j,t 0.078 0.048 0.058 0.033 0.080 0.050 0.097 0.063∆Stck,j,t 0.028 0.014 0.024 0.011 0.030 0.015 0.031 0.016∆Opt,j,t 0.040 0.027 0.026 0.017 0.040 0.029 0.053 0.041

Cumulative abnormal return at earnings announcement (in %)CAR[−1,+1]j,t 0.302 0.279 0.488 0.496 0.446 0.565 -0.027 -0.060

Control variablesBTMj,t−1 0.549 0.455 0.725 0.616 0.511 0.460 0.410 0.331MCj,t−1 10.127 2.290 0.621 0.601 2.487 2.290 27.267 13.505ROAj,t−1 0.059 0.057 0.029 0.035 0.062 0.059 0.086 0.077αj,t−1 0.014 0.011 -0.017 -0.019 0.026 0.021 0.034 0.025CAR[−30, 0]j,t−1 0.006 0.003 0.016 0.007 0.004 0.004 -0.001 -0.002FEj,t−1 0.026 0.003 0.060 0.005 0.010 0.003 0.007 0.002NOAj,t−1 18.774 17.000 11.591 10.000 17.276 16.000 27.454 26.000V olumej,t−1 126.276 97.070 123.625 96.051 137.599 106.600 117.615 90.749Dispj,t−1 0.028 0.005 0.060 0.009 0.015 0.005 0.009 0.004σROA,j,t−1 0.051 0.025 0.069 0.039 0.045 0.023 0.038 0.020βj,t−1 1.197 1.130 1.307 1.236 1.189 1.139 1.094 1.051

Note: This table reports the summary location statistics for the variables defined in Table 1. The first two columnsreport the mean and median values for the whole sample and columns 3-8 for each of the firm size tertiles.

option holdings that would come from a one percentage point increase in the firm stock price. Itis the delta of the stock times the number of shares executive k holds in stocks, plus the delta ofthe option portfolio times the number of shares of options in option portfolio for executive k’sstock-based compensation in firm j for fiscal year t

∆k,j,t = ∆Stk,k,j,t + ∆Opt,k,j,t, (3.1)

with ∆Stk,k,j,t and ∆Opt,k,j,t the price sensitivity of the stock and option portfolio, respectively:

∆Stk,k,j,t = Nj,k,t

(Sj,t100

)(3.2)

∆Opt,k,j,t =

[∂Vopt,j,k,t∂Sj,t

](Sj,t100

), (3.3)

13

Page 14: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

where Nj,k,t is the number of stocks that manager k of firm j has in his portfolio at the end ofyear t, Sj,t is firm’s j stock price at the end of year t and VOpt,j,k,t is the value of the optionportfolio for manager k of firm j at the end of year t. We then compute the option value usingthe Black and Scholes (1973) option valuation model as modified by Merton (1973) to accountfor dividends. Under this approach, there exists an explicit function for the option portfolio pricesensitivity, which is given in the Appendix of Core and Guay (2002).

To compute the overall delta for firm j in fiscal year t, we then sum over the entire manage-ment team:

∆j,t =K∑k=1

∆j,k,t, (3.4)

which gives the total compensation sensitivity to a 1% change in stock price for the executivesof firm j in year t. The compensation of the CEO is substantially different in structure andmagnitude from that of other executives. In robustness checks, we thus further distinguish theeffect of the CEO from the rest of the team of executives on the tone of earnings press releases.

Executive compensation data are from ExecuComp database. ExecuComp compiles annualsalary and bonus information as well as annual stock and option grant data for the top fiveexecutives. We have ExecuComp data for the years 2004-2011 for firms in the S&P 1500.These years and firms define our sample. ExecuComp recently changed its compensation datareporting, after the accounting changes imposed by the Financial Accounting Stardards Boardas well as expanded compensation disclosure requirements imposed by the SEC. The new rulerequires that equity-based compensation awards must be expensed based on the fair value at thegrant date. Thus, for fiscal years 1992-2005, all firms on ExecuComp report by using the oldformat. For fiscal years after 2007, all firms on ExecuComp report their compensations underthe new requirements. To measure executives’ portfolio delta in the pre-2006 format, we followthe Core and Guay (2002) methodology. For the new format, we follow the approach adopted byColes et al. (2006) and Daniel et al. (2013) and distinguish between managers’ stock and optionportfolios.6

Summary statistics on our delta measure are given in Table 2. The average (median) deltaof managers’ portfolio equals 0.328 (0.180), which means that if stock prices increase by 1%,managers’ portfolio value increases by $0.328 (0.180) mil. The average delta equals 0.199 and0.129 for the stock and option portfolios, respectively. The distribution of managers’ delta ishighly skewed, we thus use logarithms of the delta measures in the subsequent analyses. InTable 2 we also show that the average sensitivity of a manager’s portfolio (either the whole orthe individual portfolios) increases with firm size. In the first tertile, the average delta equals0.329 and progressively increases to 4.529 in the last tertile.

6The following discussion focuses on the new format only. The pre-2006 format is extensively described in Coreand Guay (2002).

14

Page 15: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

3.3 Standardized equity incentives of managers

The delta measure in the previous section is an absolute measure of the price sensitivity ofmanagers’ stock-based compensation. Because of its skewed distribution, we use the log-transformation of this variable in our regressions.

In the analysis of the impact of tone on the price reaction to the earnings press release weallow for the possibility that the investor considers that discounting tone is only needed if theincentive for tone inflation is sufficiently large for the manager. To measure this incentive, weuse the price sensitivity of managers’ stock-based compensation, relatively to the total yearlycompensation of the management.

In fact, for a given delta of $1,000, a manager’s incentive to inflate the tone and refer to cheaptalk in earnings press releases will be much lower if his total compensation equals $1,000,000 ascompared to a manager whose overall compensation equals $100,000. To correct for differencesin total compensation, we calculate the variable ∆j,t, which measures the dollar change in wealthfor a one percentage point change in firm value, divided by annual compensation. This measureof incentives is drawn from Edmans, Gabaix, and Landier (2009) and captures the share of theteam of executives’ total compensation that would come from a one percentage point increasein the value of equity of its firm, as shown below:

∆j,t =∆j,t

TotCompj,t, (3.5)

where TotCompj,t is the total annual compensation of executives of firm j in fiscal year t. Totalcompensation for the individual year, comprises the following: salary, bonus, other annual, totalvalue of restricted stock granted, total value of stock options granted (estimated following theprocedure described in Subsection 3.2), long-term incentive payouts, and all other total. Asshown in Table 2, the sample cross-sectional average (median) for ∆j,t equals 7.8% (4.8%) andincreases with firm size.

3.4 Tone of earnings press releases

The tone of the earnings press releases is measured through a content analysis in which the toneis defined as the spread between the percentage of positive and negative words:

Tonej,qt = 100 · PWj,qt −NWj,qt

TWj,qt, (3.6)

with TWj,qt the total number of words in the earnings press release of firm j in year quarter qof year t, and PWj,qt and NWj,qt are the number of positive and negative words, respectively.This approach thus treats a document as a “bag of words”, and counts the number of times aword appears in a list of positive words and negative words.

There exist different lists of words, called dictionaries, and there is no consensus in the liter-ature regarding which wordlist is more appropriate for the analysis of language in contexts suchas financial disclosures. As argued by Rogers et al. (2011), among others, the tone obtainedusing a single wordlist should be seen as a (noisy) proxy for the true, but unknown, tone of the

15

Page 16: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

text. To avoid the model risk of choosing an inappropriate library and hence reduce the noise, weaverage over the (standardized) tone obtained by using three established lists of words, namelythe positive and negative wordlists defined by Henry (2008), those defined by Loughran andMcDonald (2011) and the word list in the DICTION 7.0 software. All three of them are alreadypopular choices in practice. DICTION was used by Davis et al. (2012); Davis and Tama-Sweet(2012); Demers and Vega (2010) to analyze earnings press releases. A limitation of generalword lists such as DICTION is that they do not analyze language in the context of financial dis-closures. Prior studies (see e.g. Demers and Vega, 2010; Henry and Leone, 2009; Loughran andMcDonald, 2011) suggest that generic linguistic algorithms such as DICTION may yield noisymeasures of "positive" and "negative" linguistic tone in the context of financially-oriented textpassages. To overcome this, we also use the Henry (2008) and Loughran and McDonald (2011)word lists specifically designed for financial disclosures.

As shown in Figure 3 in Appendix, the tone measure based on the three lists are significantlypositively correlated but not perfectly. This correlation is consistent with each measure capturingthe underlying construct of tone plus some idiosyncratic error.7 While correlations between thetone from individual libraries are between 54 and 60%, we find that the correlation of eachlibrary with our standardized average measure of tone (Tonej,t) is between 81% and 84%. Inthe following sections, we focus our discussion on the standardized average tone measure andshow the robustness of our results to the different libraries in Section 4.

3.5 Control variables

We argue that there are two main components to the tone of earnings press releases: (i) the in-formative part and (ii) the biased part. To capture the tone inflation in earnings releases, we needto control for the informative part contained in the release. Just as a firm’s voluntary disclosurebehavior is influenced by its economic performance (Miller, 2002), we expect disclosure tone tobe correlated with a variety of economic attributes, such as firm characteristics, past firm perfor-mance, earnings uncertainty and the firm’s informational environment. We thus include each ofthese categories as a control for the informative part of the tone in our regressions.

We include two variables to control for firm characteristics, firm size and book-to-marketratio. We measure firm size as the log of market equity (logMCj,t) and the log book-to-market(logBTMj,t) at the end of the preceding calendar year, following Fama and French (1992).

We define three variables to control for a firm’s past performance. Davis et al. (2012) andDavis and Tama-Sweet (2012) show that the tone in earnings releases is highly correlated witha firm’s past performance. We control for past firm performance by lagged return on assets(ROAj,t−1). The other two of our control variables are based on a simple earnings announce-ment event study methodology: the cumulative abnormal return from the [-30,0] trading daywindow (CAR[−30, 0]j,t) and the estimated intercept from the event study regression that spansthe [-252,-31] time window, αj,t, where t = 0 is the last trading day of year t. We estimatebenchmark returns using the market model with an estimation window of [-275,-31] trading

7An alternative is to use the scores of the first principal component factor analysis (PCA) as in Rogers et al. (2011).Since the PCA loads almost equally on the three sentiment measures, this leads to almost identical results as withthe average (standardized) sentiment approach, but the average sentiment approach has the advantage that theresulting tone is easier to interpret.

16

Page 17: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

Figure 1: Tone by year, using the DICTION 7.0, Henry (2008) and Loughran and McDon-ald (2011) libraries and their simple average

Year

Net

Sen

timen

t

2004 2005 2006 2007 2008 2009 2010 2011

−0.

50.

00.

51.

01.

5Equally−weighted scoreHenry (2008)DICTION 7.0Loughran and McDonald (2011)

Note: This figure reports the average tone of the earnings press releases of the S&P 1500 between 2004 and 2011.The Equally-weighted measure of tone is measured as the standardized average of the tone measures obtained usingthree established lists of words, namely: the positive and negative wordlists defined by Henry (2008), those definedby Loughran and McDonald (2011) and the word list in the DICTION 7.0 software. We also report the average tonemeasures using the DICTION 7.0, Henry (2008) and Loughran and McDonald (2011) libraries.

days prior to the end of the current fiscal year t. We interpret αj,t as the firm’s in-sample cu-mulative abnormal return over the previous calendar year, skipping the most recent month. Weexpect a positive relationship between the tone in earnings press releases and the firm’s pastperformance.

To account for differences in a firm’s informational environment, we include three proxies forinformation asymmetry, namely analysts’ forecast error (FEj,t), the logarithm of the numberof financial analysts following the firm in fiscal year t (logNOAj,t) and the logarithm of shareturnover (log V olumej,t) measured over fiscal year t and standardized by the number of sharesoutstanding. Firms with high information asymmetry have a higher forecast error (Thomas,2002), a lower analyst coverage (Roulstone, 2003) and a lower share turnover because unin-formed traders are less likely to trade in these shares, knowing that they will lose to informedtraders (Mohd, 2005). We expect firms with less information asymmetry to be less overly opti-mistic in their earnings pres releases.

We also add three variables to control for information uncertainty: (i) analyst forecast dis-

17

Page 18: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

persion (DISPj,t) as defined by Das, Levine, and Sivaramakrishnan (1998), (ii) the standarddeviation of ROAj,t over the four years preceding the end of fiscal year t, as defined by Coreand Guay (2002) and (iii) the firm’s systematic risk (βj,t), estimated in the event study regres-sion that spans the [-252,-31] time window. We expect managers to be less optimistic in theirearnings press releases when the information environment is more uncertain.

Summary statistics are reported in Table 2. When comparing small- and large-cap firms,important differences appear in terms of firm characteristics, past performance and informationenvironment. First, the book-to-market ratio (BTMj,t−1) of large firms is substantially lowerthan that of smaller firms. Furthermore, past performance of large firms is much higher in termsofROA and the number of analysts (NOAj,t−1) following large firms is substantially more thanfor smaller firms. Moreover, analysts’ error (FEj,t−1) as well as analysts’ forecast dispersion(Dispj,t−1) is much lower for larger firms. Finally, smaller firms also have a smaller variationin past performance (σROA,j,t−1). These results support the findings of Atiase (1980, 1985).

4 Results

The results in this section are divided into two subsections. In Section 4.1, we evaluate execu-tives’ tone inflation and test the relation between managers’ compensation sensitivity to changesin stock prices and tone in earnings press releases across firms between 2004 and 2011. We findthat earnings press releases are more optimistic as managers’ stock-based compensation sensi-tivity to stock price increases. In Section 4.1.1, we distinguish between the sensitivity of theCEO’s and the team of executives’ portfolios. We find that it is CEO’s portfolio sensitivity tostock price that is the primary driving force behind the observed tone inflation within earningspress releases and that the longer the tenure of the CEO, the less he/she will engage in toneinflation within the narrative sections of the firm’s financial disclosure.

Section 4.2 tests for market efficiency with respect to investors’ reaction to the linguistictone in earnings press releases. We find that the market reacts more positively to optimisticearnings press releases, but less so when the executives’ compensation package is more sensitiveto changes in stock prices. We define a two-regime-threshold least-squares model that estimatesthe cutoff point at which investors start discounting the inflated tone in earnings press releases.This cutoff point reflects the change in managers’ total wealth to a 1% change in stock price. Theresults indicate that investors start discounting for cheap talk in earnings press releases when thedollar change in managers’ total wealth to a 1% change in stock price equals 1.740% of theirtotal compensation.

Taken together, these results suggest that the use of equity-based incentives reduces the infor-mativeness of the tone of earnings press releases. Highly incentivized managers are more likelyto inflate the tone of their earnings press releases, but investors appear to correct for this bias bypartly discounting the cheap talk in the earnings press release.

18

Page 19: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

4.1 The effect of managers’ equity incentives on the tone in earnings pressreleases

In our first hypothesis, we test whether an increase in an executive’s stock-based compensationsensitivity incentivizes the manager to inflate the tone in earnings press releases, provided it iseffective and cheap to do so. We define the following linear least-squares model in which weregress the tone on managers’ stock-based compensation (∆j,t−1):

Tonej,t = β0 + β1 · log ∆j,t−1 + β3 · Γj,t−1 + (4.1)

6∑y=1

βy · Y eary,t +

9∑s=1

βs · Inds,j + εj,t,

where Γj,t−1 corresponds to a set of variables that have been shown in the prior literature toexplain the tone in earnings press releases, each of which are defined in Panel C of Table 1.Y eary,t is an indicator variable taking the value of one if the fiscal year of the earnings pressrelease is for year t and zero otherwise. Inds,j is an indicator variable taking the value of one ifthe earnings press release is for a firm in industry s and zero otherwise.

Because all equity incentive variables are measured during or at the end of fiscal year t andare made available one to four months after the fiscal-year end (Cheng and Warfield, 2005), weconsider a one-year lag in the sensitivity of managers’ stock-based compensation variables. Toavoid any look-ahead bias and endogeneity issues, we also consider a one-year lag in the setof control variables. Table 5 presents the estimation results for the model in Equation (4.1),where we have suppressed the estimated coefficients on the industry and year dummy variablesfor presentation purposes. We test for the significance of the coefficients using Newey-Weststandard errors.

Model (1) of Table 5 presents the results based on Equation 4.1 estimated without controlvariables. Consistent with Hypothesis 1, we find a positive coefficient for the log ∆j,t−1 vari-able, which suggests that an increase in managers’ compensation sensitivity is associated with asignificant increase in the tone of earnings press releases. In Model (2), we distinguish betweenthe delta of the stock and option portfolios. Both variables are highly significant.8

In Models (3) and (4), we add the control variables. In Model (3), we control for the firm’spast accounting and market performance and book-to-market ratio (logBTMj,t−1), as well asfor year and industry fixed effects. We control for market-to-book ratios in an effort to exclude apotential alternative explanation for our findings. Smith and Watts (1992) look at data aggregatedof the level of industries. They show that there is a positive relationship between firm’s growthopportunities and their pay-performance sensitivity. The estimated coefficient of log ∆j,t−1 issubstantially smaller but remains highly significant at a 99% confidence level. We add in Model(4) the control variables for information environment and earnings uncertainty. The coefficientfor the log ∆j,t−1 slightly decreases but remains highly significant. In addition, out of the 10

8For presentation purposes, we did not report the estimation output of Model (2) that includes control variables.The result remains qualitatively similar if we add the control variables, with the exception that the coefficient forthe variable ∆Opt,j,t−1 becomes insignificant at traditional confidence levels.

19

Page 20: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

Table 3: Tone in earnings press releases and managers’ equity incentives

Panel A – All firms Panel B – Size tertiles

Tonej,t Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7Small-Cap Mid-Cap Large-Cap

(Intercept) 0.130∗∗∗ 0.270∗∗∗ 0.010 0.619∗∗ 0.799 1.031∗∗∗ 0.787(0.029) (0.039) (0.224) (0.272) (0.665) (0.272) (0.638)

Sensitivity of the managerial portfolios to the stock pricelog ∆j,t−1 0.137∗∗∗ 0.058∗∗∗ 0.050∗∗ 0.073∗∗ 0.041∗∗ 0.067

(0.016) (0.018) (0.020) (0.032) (0.020) (0.046)log ∆Stck,j,t−1 0.068∗∗∗

(0.015)log ∆Opt,j,t−1 0.070∗∗∗

(0.013)

Control variablesROAj,t−1 0.590∗∗∗ 0.677∗∗∗ 0.995∗∗∗ 0.353∗∗ 0.380

(0.170) (0.163) (0.257) (0.163) (0.365)logBTMj,t−1 -0.129∗∗∗ -0.132∗∗∗ -0.147∗∗∗ -0.175∗∗∗ −0.064

(0.033) (0.034) (0.049) (0.034) (0.054)αj,t−1 0.378∗∗∗ 0.375∗∗∗ 0.216∗ 0.398∗∗∗ 0.601∗∗∗

(0.080) (0.081) (0.120) (0.081) (0.176)CAR[−30, 0]j,t−1 0.058 0.061 −0.050 0.135 0.034

(0.133) (0.127) (0.188) (0.127) (0.206)FEj,t−1 0.056∗∗∗ 0.048∗∗∗ -0.261∗∗∗ 0.864∗∗∗

(0.017) (0.017) (0.017) (0.289)logNOAj,t−1 0.023 0.131∗∗ 0.005 −0.075

(0.042) (0.058) (0.042) (0.124)log V olumej,t−1 -0.140∗∗∗ -0.110∗∗ -0.232∗∗∗ −0.118

(0.037) (0.052) (0.037) (0.075)Dispj,t−1 -0.078∗∗ -0.055∗ -1.073∗∗∗ -3.952∗

(0.035) (0.032) (0.035) (2.098)σROA,j,t−1 -0.428∗ −0.052 -0.818∗∗∗ −0.343

(0.245) (0.348) (0.245) (0.374)βj,t−1 0.028 −0.048 0.172∗∗∗ 0.146

(0.046) (0.066) (0.046) (0.097)

Year fixed effects No No Yes Yes Yes Yes YesIndustry fixed effects No No Yes Yes Yes Yes Yes

R2 0.055 0.063 0.180 0.193 0.196 0.198 0.262Adj. R2 0.055 0.062 0.174 0.186 0.174 0.175 0.241Num. obs. 2795 2795 2795 2795 932 931 932

Note: This table presents the estimation results for Equation 4.1. In Panel A, we report the estimation results forthe whole sample and Panel B reports the results for Equation 4.1 for each size tertile. Tonej,t is measured as thestandardized average of the tone measures obtained using three established lists of words, namely: the positive andnegative wordlists defined by Henry (2008), those defined by Loughran and McDonald (2011) and the word list inthe DICTION 7.0 software. Each variable is defined in Table 1. The significance of the coefficients is tested usingNewey-West standard errors. ∗, ∗∗, and ∗∗∗ denote statistical significance at the 10 percent, 5 percent, and 1 percentlevels, respectively, based on a two-sided t-test.

20

Page 21: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

variables, 8 (7) have the predicted sign (and are significant). As shown in Model (4), controllingfor the book-to-market ratio, past performance, information environment and earnings uncer-tainty suggests that our results are not driven by the more volatile operating environment offirms that use a high amount of stock-based compensation.

In Models (5)-(7), we replicate Model (4) but perform the analysis based on size-tertiles.Although the summary statistics showed that managers of small-cap firms have a substantiallylower portfolio sensitivity than large-cap firms, we find that the managers inflate the tone sub-stantially more for small-cap firms as compared to large-cap firms. The coefficient for thelog ∆j,t−1 decreases from a significant 0.073 for the first size tertile to an insignificant 0.067in the last tertile. Consistent with our expectations, this result can be explained by the factthat large-cap firms are more scrutinized by market participants such as financial analysts andinvestors. This increased scrutiny of market participants thus increases the likelihood of share-holder litigation in the case of over-optimism in earnings press releases (Rogers et al., 2011).On the other hand, small-cap firms have a higher level of information asymmetry, which givesmanagers more room to inflate the tone in their earnings press releases, while limiting the costto their credibility or litigation with shareholders.

Managerial compensation sensitivity to stock price thus does not just increase managers’ like-lihood to manipulate earnings numbers (see, e.g., Bergstresser and Philippon, 2006; Burns andKedia, 2006), but also increases managers’ incentives to inflate the tone of the narrative sectionof earnings press releases. It implies that the tone measure should not merely be considered asa signal of future performance, but also as a signal that reflects managers’ underlying incen-tives to maximize their wealth. Because of the higher and predictable level of cheap talk inthe earnings press releases, we expect and show in the next section that investors discount thetone of the release for a sufficiently high level of stock-based compensation compared to totalcompensation.

4.1.1 CEO and the team of executives

Thus far in the paper, we referred to managers as the group of CEO, CFO and the three highest-compensated named executive officers. As we show in Table 4, the sensitivity of the CEO’sportfolio is substantially larger for CEOs than for other executives in the management team. Infact, the average delta of a CEO equals 0.478, while the average sensitivity of the top four exec-utives equals $0.099 mil. for a 1% increase in stock price. To further examine the consistencyof our results to our specifications, we thus estimate Equation 4.1 using the CEOs’ portfoliosensitivity to stock price only, rather than the full management team.

In each model in Table 4, we also add two new CEO-specific control variables to the regres-sion. As a first variable, we add the CEO’s tenure Tenurej,t, defined as the number of yearsan executive has served as CEO in firm j. There is ample evidence that longer-tenured exec-utives may be less willing to engage in unethical or fraudulent behaviors that might ruin theirestablished reputations (Gray and Cannella, 1997; Zhang, Bartol, Smith, Pfarrer, and Khanin,2008). In addition, there is evidence that new CEOs may have less to lose and may be moreaggressive and take chances in order to build their personal wealth (Brouthers, Brouthers, andWerner, 2000). We thus expect tenure to interact negatively with linguistic optimism in earn-ings releases. Dechow, Sloan, and Sweeney (1996) find that having a Chairman who is also

21

Page 22: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

Table 4: CEO and managers’ equity incentives

All firms Small-Cap Mid-Cap Large-Cap

Mean Median Mean Median Mean Median Mean Median

∆CEO,j,t−1 0.478 0.239 0.175 0.089 0.403 0.244 0.856 0.634∆Team,j,t−1 0.099 0.054 0.033 0.020 0.086 0.056 0.178 0.143

∆CEO,Stck,j,t−1 0.161 0.072 0.090 0.032 0.149 0.071 0.242 0.148∆Team,Stck,j,t−1 0.029 0.014 0.013 0.006 0.029 0.016 0.046 0.031

∆CEO,Opt,j,t−1 0.252 0.122 0.064 0.038 0.198 0.134 0.492 0.409∆Team,Opt,j,t−1 0.060 0.031 0.017 0.012 0.047 0.033 0.115 0.097

Note: This table reports the location summary statistics for the stock-based compensation sensitivity to stock pricefor the CEO and the top-four executives for the whole sample and each of the size tertiles.

simultaneously the CEO increases the likelihood of manipulating earnings. We thus also in-clude the dummy isCHMNj,t, which takes the value of one if the CEO is also the Chairman ofthe Board of Directors and zero otherwise. We expect a positive relationship between Tonej,tand isCHMNj,t.

Models (1)-(3) in Table 4 show that the CEO’s portfolio sensitivity has a significant positiveimpact on the tone included in earnings press releases.9 The results are thus basically unchangedand show that CEOs resort to cheap talk to inflate the tone of press releases as their portfolio’ssensitivity to stock price increases. However, tone inflation is mitigated by the CEO’s tenure.Tenurej,t is significantly negatively correlated with the tone in earnings press releases in allmodels. On the other hand, isCHMNj,t is positive but insignificant.

To compare the influence of CEO’s and the top four executives, we report the result of Equa-tion 4.1 with both the delta of the CEO’s and the top four executives’ (other than the CEO)portfolios. We define the management team delta as the sum of the deltas of the top four ex-ecutives other than the CEO, divided by the number of executives in the team. Model (4) ofTable 4 shows that log ∆CEO,j,t−1 remains positive and significant. In fact, only the sensitivityof the CEO’s stock portfolio explains the tone inflation in earnings releases, as the coefficienton the team of executives’ delta (log ∆Team,j,t−1) is substantially smaller and insignificant. TheCEO’s compensation is thus the primary driving force behind the observed tone inflation withinearnings press releases and, the longer his/her tenure, the less he/she will engage in tone inflationwithin the narrative sections of earnings press releases.

4.1.2 Robustness checks

Our main findings that the price sensitivity of managers’ stock-based compensation induces toneinflation completely align with our initial expectations. However we bear in mind that the exactquantification of this effect depends on the measurement of tone and price sensitivity, as well as

9In some cases, ExecuComp does not designate which of the executives is the CEO. In this case, we remove theobservation from the sample, which reduces the number of observations from 2,795 to 2,316.

22

Page 23: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

Table 5: CEO’s vs. team of executives’ equity incentives and the tone in earnings pressreleases

Tonej,t Model 1 Model 2 Model 3 Model 4

(Intercept) 0.392∗∗∗ 0.542∗∗∗ 0.928∗∗∗ 0.925∗∗∗

(0.074) (0.087) (0.272) (0.274)

Sensitivity of the CEO’s and team of executives’ managerial portfolios to the stock pricelog ∆CEO,j,t 0.152∗∗∗ 0.061∗∗∗ 0.062∗∗

(0.019) (0.024) (0.031)log ∆CEO,Stck,j,t 0.077∗∗∗

(0.017)log ∆CEO,Opt,j,t 0.071∗∗∗

(0.013)log ∆Team,j,t −0.002

(0.028)

CEO-specific variableslog Tenurej, t -0.100∗∗∗ -0.095∗∗∗ -0.056∗∗ -0.056∗∗

(0.028) (0.028) (0.026) (0.028)isCHMNj,t 0.037 0.031 0.034 0.035

(0.053) (0.052) (0.050) (0.050)

Control variables No No Yes YesYear fixed-effects No No Yes YesIndustry fixed-effects No No Yes Yes

R2 0.063 0.069 0.204 0.204Adj. R2 0.062 0.068 0.194 0.194Num. obs. 2331 2331 2331 2331

Note: This table presents the estimation results for Equation 4.1 and distinguishes between the portfolio sensitivityof CEOs and the other top four executives. Tonej,t is measured as the standardized average of the tone measuresobtained using three established lists of words, namely: the positive and negative wordlists defined by Henry (2008),those defined by Loughran and McDonald (2011) and the word list in the DICTION 7.0 software. Each variable isdefined in Table 1. The significance of the coefficients is tested using Newey-West standard errors. ∗, ∗∗, and ∗∗∗

denote statistical significance at the 10 percent, 5 percent, and 1 percent levels, respectively, based on a two-sidedt-test.

the control variables and model specification used. Therefore we now illustrate the robustnessof our findings.

As a first robustness check, we report the results for each list of words included in the Tonej,tmeasure, namely the DICTION 7.0, Henry (2008) and Loughran and McDonald (2011) libraries,as well as for the principal component score obtained on those three libraries.10 In Panel A,we find that, for each library, the tone in earnings press releases is significantly and positivelyrelated to managers’ compensation sensitivity to stock price. The Adj.R2 ranges between 14

10Each library contains words that overlap with other lists, but the overlap is far from complete, as only 25% of thewords in the Henry lists are contained in the DICTION list; 55% of the words in the Henry lists are contained inthe Loughran and McDonald (2011) lists and 13% of the words in the Loughran and McDonald (2011) lists arecontained in the DICTION lists.

23

Page 24: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

and 19%, among which the model based on the Henry (2008) lists is performing the best. Weaggregated the quarterly tone in earnings press releases to match with the frequency of managers’compensation. As a second robustness check, we verify in Panel B whether working with aquarterly frequency does not alter our results. We show that our main conclusion holds on aquarterly frequency. The magnitude of the coefficients for managers’ compensation sensitivityis the largest for quarter 1 and quarter 4. The third robustness check considers the relative pricesensitivity of managers’ stock-based compensation and replaces log ∆j,t−1 in Equation 4.1 bythe corresponding log ∆j,t−1 variables. The results in Panel C are qualitatively similar: thepercentage change in managers’ total compensation is significantly correlated with the tone inearnings press releases.

4.2 Nonlinearity in the market response to investor tone and managers’equity-based incentives

Hypothesis 2 defines our base-line model which replicates the results obtained by Davis et al.(2012), Davis and Tama-Sweet (2012) and Henry (2008). We predict that the textual tone inearnings press releases is positively associated with market returns around the earnings an-nouncement date. We estimate a multivariate regression model in which investors’ responseis regressed on Tonej,t. We measure investors’ response CAR[−1,+1]j,qt to quarterly earn-ings press releases as the cumulative abnormal return over the three-day window centered on theearnings press release date and average it over fiscal year t. To measure the incremental marketresponse to managers’ language, we include the set of control variables (Γ) defined in Table 1.Our model tests for the significance of the coefficients using Newey-West standard errors:

CAR[-1,+1]j,t = β0 + β1 · Tonej,t + β2 · Γj,t−1 + (4.2)

6∑y=1

βy · Y eary,t +9∑s=1

βs · Inds,j + εj,t.

Panel A of Table 7 presents the estimation results for Equation 4.2. As expected, the coef-ficient on Tonej,t is positive and highly significant, suggesting that higher values of tone areassociated with positive abnormal returns around the issuance of the earnings press release. Thisresult is consistent with Davis et al. (2012), Davis and Tama-Sweet (2012) and Henry (2008) andshows that managers use language in earnings press releases to signal their future expectations.Investors thus recognize that earnings press release contain information value and respond tomanagers’ use of language as a voluntary disclosure mechanism.

As Graham and Harvey (2001) note, differences in the corporate finance practices of smalland large firms are pervasive. We thus split the sample into tertiles based on firm size, whichpermits us to capture nonlinear size effects to investors’ reaction to tone. Although we found inthe previous subsection that cheap talk is prevalent for small-cap firms, the tone of earnings pressreleases is more informative for small-cap than for large-cap firms. As shown in Models (2)-(4),the average stock price reaction to the tone decreases from 0.905 for small-caps to 0.578 for

24

Page 25: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

Table 6: Robustness checks – Tone in earnings press releases and managers’ equity-basedincentives

Panel A – By library

Tonej,t Score Henry (2008) DICTIONLoughran and

McDonald (2011)(Intercept) 1.058∗∗ 2.357∗∗∗ 0.332 0.132

(0.473) (0.288) (0.208) (0.311)log ∆j,t−1 0.087∗∗ 0.040∗ 0.037∗ 0.046∗∗∗

(0.035) (0.022) (0.020) (0.018)

Control variables Yes Yes Yes YesYear fixed-effects Yes Yes Yes YesIndustry fixed-effects Yes Yes Yes Yes

R2 0.193 0.200 0.159 0.148Adj. R2 0.185 0.193 0.151 0.140Num. obs. 2795 2795 2795 2795

Panel B – By quarterTonej,t Quarter 1 Quarter 2 Quarter 3 Quarter 4(Intercept) 0.513∗ 0.573∗∗ 0.351 0.838∗∗∗

(0.278) (0.226) (0.274) (0.234)log ∆j,t−1 0.072∗∗∗ 0.054∗∗∗ 0.047∗∗∗ 0.073∗∗∗

(0.021) (0.016) (0.015) (0.017)

Control variables Yes Yes Yes YesYear fixed-effects Yes Yes Yes YesIndustry fixed-effects Yes Yes Yes Yes

R2 0.175 0.166 0.153 0.163Adj. R2 0.167 0.161 0.149 0.159Num. obs. 2779 5443 5475 5506

Panel C – Price sensitivity of managers’ stock-based compensation standardized by yearly total compensationTonej,t Model 1 Model 2 Model 3 Model 4(Intercept) 0.499∗∗∗ 0.669∗∗∗ 0.190 0.682∗∗

(0.066) (0.078) (0.241) (0.278)

log ∆j,t−1 0.154∗∗∗ 0.068∗∗∗ 0.055∗∗∗

(0.019) (0.019) (0.020)

log ∆Stck,j,t−1 0.073∗∗∗

(0.016)

log ∆Opt,j,t−1 0.086∗∗∗

(0.013)

Control variables No No No YesYear fixed-effects No No Yes YesIndustry fixed-effects No No Yes Yes

R2 0.047 0.059 0.179 0.193Adj. R2 0.047 0.058 0.173 0.186Num. obs. 2795 2795 2795 2795

Note: This table presents the robustness checks of Hypothesis 1. The estimation results are for Equation 4.1. In PanelA, we verify the robustness of our results with respect to the library used to measure tone. We measure Tonej,t usingthree established lists of words, namely: the positive and negative wordlists defined by Henry (2008), those definedby Loughran and McDonald (2011) and the word list in the DICTION 7.0 software. We also report the results for theprincipal component score obtained on those three libraries. Panel B reports the regression results by quarter. PanelC reports the results of 4.1 where log ∆j,t−1 in Equation 4.1 is replaced by the corresponding log ∆j,t−1 variables.Each variable is defined in Table 1. The significance of the coefficients is tested using Newey-West standard errors.∗, ∗∗, and ∗∗∗ denote statistical significance at the 10 percent, 5 percent, and 1 percent levels, respectively, based ona two-sided t-test. 25

Page 26: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

Table 7: Hypothesis 2 – The market response to tone in earnings press releases and man-agers’ equity incentives - Univariate models

Panel A – Hypothesis 2 Panel B – Hypothesis 3

CAR[−1,+1]j,t Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8All firms Small-Cap Mid-Cap Large-Cap All firms Small-Cap Mid-Cap Large-Cap

κ (%) - - - - 1.740 4.340 9.762 20.846∆κ - - - - 0.804 0.144 2.428 3.128% I

[∆j,t−1 ≤ κ

]- - - - 16.279 60.730 74.758 85.193

(Intercept) 0.289∗∗∗ 0.604∗∗∗ 0.412∗∗∗ −0.060 0.317∗∗∗ 0.686∗∗∗ 0.389∗∗∗ −0.051(0.081) (0.177) (0.136) (0.102) (0.081) (0.180) (0.138) (0.102)

Tonej,t 0.498∗∗∗ 0.905∗∗∗ 0.578∗∗∗ 0.218∗ 0.463∗∗∗ 1.203∗∗∗ 0.474∗∗∗ 0.184(0.109) (0.275) (0.163) (0.126) (0.104) (0.334) (0.168) (0.126)

Φj,t -0.313∗∗∗ -0.628∗∗ -0.973∗∗ -1.816∗∗

(0.110) (0.280) (0.471) (0.725)

R2 0.009 0.016 0.013 0.003 0.012 0.021 0.018 0.007Adj. R2 0.008 0.015 0.012 0.002 0.011 0.018 0.016 0.005Num. obs. 2795 932 931 932 2795 932 931 932

Note: This table presents the results of investors’ response to the tone in earnings press releases and the price sen-sitivity of managers’ equity incentives. Panel A reports the estimation results of Equation 4.2, without the controlvariables. Panel B reports the estimation results of Equation 4.3, without the control variables. Φj,t is the interactionvariable between Tonej,t and ∆j,t−1 and is defined in Equation 4.4. Tonej,t is measured as the standardized aver-age of the tone measures obtained using three established lists of words, namely: the positive and negative wordlistsdefined by Henry (2008), those defined by Loughran and McDonald (2011) and the word list in the DICTION 7.0software. Each variable is defined in Table 1. The significance of the coefficients is tested using Newey-West standarderrors. ∗, ∗∗, and ∗∗∗ denote statistical significance at the 10 percent, 5 percent, and 1 percent levels, respectively,based on a two-sided t-test.

mid-caps to finally reach an insignificant 0.218 for large-caps. This negative relation betweenthe firm size and investors’ reaction to tone is consistent with Hypothesis 2a and verifies theresults of Atiase (1985) on unexpected price changes: the level of information asymmetry ishigher for smaller firms, which explains the stronger price reaction to the tone. In Panel A ofTable 8, we add the control variables and obtain qualitatively similar results.

Yet, this result does not rule out the potential for managers’ opportunistic use of language inearnings press releases. In Hypothesis 3, we test whether, for a sufficiently high level of stock-based compensation compared to total compensation (∆), investors anticipate tone inflation andwill react proportionally less to the textual tone when the managers’ compensation sensitivity tothe stock price increases. To do this, we define a threshold model that separates two regimes.The first regime, in which the relative price sensitivity ∆ of the stock-based incentives is smallerthan a threshold value κ, is characterized by a linear response of the CAR to an increase in tone.In the second regime, investors expect that the relative price sensitivity ∆ is sufficiently high forthe manager to inflate the tone of the press release and therefore the response of the CAR to aincrease in tone decreases when the price sensitivity increases. We require that at the transitionbetween the two regimes, the left and right limit of the response of the CAR to the tone is thesame. This leads us to formulate the following threshold model that includes an interaction

26

Page 27: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

Table 8: Hypothesis 2 – The market response to tone in earnings press releases and man-agers’ equity incentives - Multivariate models

Panel A – Hypothesis 2 Panel B – Hypothesis 3

CAR[−1,+1]j,t Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8All firms Small-Cap Mid-Cap Large-Cap All firms Small-Cap Mid-Cap Large-Cap

κ (%) - - - - 0.040 4.340 9.762 19.546∆κ - - - - 0.003 0.144 2.428 3.128% I

[∆j,t−1 ≤ κ

]- - - - 0 60.730 74.758 85.193

(Intercept) 1.125 4.058 −1.404 1.961∗∗∗ 1.217 3.823 −1.358 2.120∗∗∗

(1.210) (3.287) (2.459) (0.534) (1.239) (3.300) (2.477) (0.622)

Textual tone in earnings press releasesTonej,t 0.618∗∗∗ 1.022∗∗∗ 0.704∗∗∗ 0.231∗ 1.779∗∗∗ 1.351∗∗∗ 0.588∗∗∗ 0.236∗

(0.111) (0.278) (0.171) (0.140) (0.544) (0.344) (0.185) (0.138)

Interaction variable between tone and price sensitivity of managers’ stock-based compensationΦj,t -0.228∗∗ -0.678∗∗ -0.989∗∗ -1.778∗∗

(0.098) (0.283) (0.461) (0.782)

Control variablesROAj,t−1 −1.283 −2.300 −1.143 −0.139 -1.501∗ −2.344 −1.121 −0.259

(0.870) (1.631) (1.143) (1.517) (0.869) (1.635) (1.125) (1.536)logBTMj,t−1 0.112 0.047 0.210 −0.074 0.110 0.018 0.195 −0.082

(0.130) (0.289) (0.220) (0.188) (0.129) (0.287) (0.219) (0.189)αj,t−1 0.432 0.357 1.388 −0.916 0.327 0.315 1.373 −0.904

(0.673) (1.178) (1.041) (1.034) (0.668) (1.173) (1.043) (1.034)CAR[−30, 0]j,t−1 −0.592 −1.735 −0.363 1.500 −0.611 −1.795 −0.417 1.470

(1.029) (1.654) (1.845) (1.318) (1.034) (1.663) (1.845) (1.324)FEj,t−1 −0.584 −0.511 −4.635 -7.179∗∗∗ −0.610 −0.522 −4.923 -7.225∗∗∗

(0.395) (0.450) (3.437) (0.877) (0.387) (0.446) (3.591) (0.860)Dispj,t−1 0.580 0.473 5.373 0.272 0.704 0.520 5.360 0.118

(0.512) (0.587) (3.458) (4.527) (0.469) (0.572) (3.591) (4.488)

Year fixed effects Yes Yes Yes Yes Yes Yes Yes YesIndustry fixed effects Yes Yes Yes Yes Yes Yes Yes Yes

R2 0.021 0.037 0.046 0.032 0.024 0.042 0.050 0.035Adj. R2 0.013 0.015 0.023 0.008 0.016 0.019 0.026 0.010Num. obs. 2795 932 931 932 2795 932 931 932

Note: This table presents the results of investors’ response to the tone in earnings press releases and the price sensitiv-ity of managers’ equity incentives. Panel A reports the estimation results of Equation 4.2, with the control variables.Panel B reports the estimation results of Equation 4.3, with the control variables. Φj,t is the interaction variablebetween Tonej,t and ∆j,t−1 and is defined in Equation 4.4. Tonej,t is measured as the standardized average of thetone measures obtained using three established lists of words, namely: the positive and negative wordlists definedby Henry (2008), those defined by Loughran and McDonald (2011) and the word list in the DICTION 7.0 software.Each variable is defined in Table 1. The significance of the coefficients is tested using Newey-West standard errors.∗, ∗∗, and ∗∗∗ denote statistical significance at the 10 percent, 5 percent, and 1 percent levels, respectively, based ona two-sided t-test.

27

Page 28: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

variable between Tonej,t and ∆j,t−1:

CARj,t = β0 + β1 · Tonej,t + β2 · Φj,t + (4.3)

β3 · Γj,t−1 +

6∑y=1

βy · Y eary,t +

9∑s=1

βs · Inds,j + εj,t,

where Φj,t is the interaction variable between Tonej,t and managers’ compensation ∆j,t−1 andis defined as:

Φj,t = [log ∆j,t−1 − log ∆κ] · Tonej,t · I[∆j,t−1 > κ

]. (4.4)

∆κ ensures the continuity of the sensitivity of the expected CAR to an increase in the average∆j,t around κ. More precisely, ∆κ is such that:

lim∆j,t−1→κ

∂E[CARj,t|∆j,t−1; Γt−1]

∂∆j,t−1= lim

κ←∆j,t−1

∂E[CARj,t|∆j,t−1; Γt−1]

∂∆j,t−1,

with Γt−1 the set of all explanatory variables but ∆j,t−1. In order words, ∆κ is such that β(measuring the sensitivity for ∆j,t−1 → κ) equals β+ω [∆j,t−1 −∆κ] (measuring the sensitivityfor κ← ∆j,t−1).11 Note that the model (4.2) is nested in the nonlinear model (4.3).

Model (5) in Panel B of Table 7 presents the results of our threshold model estimated withoutcontrol variables. As before, the coefficient on Tonej,t is positive and highly significant. In-vestors’ average reaction to the tone in earnings press releases equals 0.317 + 0.463 ·Tonej,t aslong as ∆ is below the threshold level (κ) of 1.740% (or a ∆j,t−1 of $ 804,000), which represents16% of the sample observations. As ∆ goes beyond κ, investors start discounting the cheap talkcontained within earnings press releases. The coefficient for the interaction variable Φj,t equals−0.313 and is highly significant at a 99% confidence level. The average stock reaction for alevel of compensation higher than κ thus becomes 0.317 + 0.463 ·Tonej,t− 0.313 ·Φj,t. Whencompared with Model (1) of Panel A, the univariate threshold model increases the adjusted R2

from 0.8 to 1.1%.The marginal investor reaction to the tone of earnings press releases is depicted in Figure

2a. It clearly illustrates the non-linearity of investors’ reaction in function to the sensitivity ofmanagers’ stock-based incentives. As the ∆j,t−1 increases, the marginal stock price reactiondecreases and reaches zero at a level of portfolio price sensitivity of about $2 million. In Figure2b, we distinguish between small-, mid- and large-caps. The figure clearly shows that infor-mativeness of the tone in earnings releases decreases as size increases and that the threshold atwhich investors start discounting the cheap talk in earnings releases substantially increases withfirm size.

Investors seem to interpret a high price sensitivity of managers’ stock-based compensationas a signal that the tone of the earnings press release has been inflated by the management and

11In practice, we set ∆κ as the best linear prediction of ∆j,t−1 given ∆j,t−1 for ∆j,t−1 = κ.

28

Page 29: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

Figure 2: Investors’ marginal reaction at earnings announcement to tone in function of theprice sensitivity of managers’ stock-based compensation

0 1 2 3 4

−1.

5−

1.0

−0.

50.

00.

51.

01.

5

Delta (in $mil)

Mar

gina

l inv

esto

r re

actio

n (%

)

First regime Second regime

(a) All firms

0 1 2 3 4

−1.

5−

1.0

−0.

50.

00.

51.

01.

5

Delta (in $mil)

Mar

gina

l inv

esto

r re

actio

n

Small−CapMid−CapLarge−Cap

(b) By Size Tertiles

Note: This figure depicts the marginal investor reaction to the tone of earnings press releases. In Figure 2a we reportthe marginal investor reaction to all firms and in Figure 2b, we illustrate investors’ reaction by size tertiles. Investors’reaction is estimated following Equation 4.3.

29

Page 30: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

therefore react proportionally less to the tone in earnings press releases as the sensitivity oftheir stock-based compensation increases. Although investors are aware that the tone containsinformation about a firm’s future performance and react positively to it (first regime), they alsoknow that the tone is biased and will adjust their reaction downward. We interpret investors’anticipation as the cost to the cheap talk of the manager and, on the basis of the Almost-cheap talkmodel of Kartik (2009), we empirically confirm that investors interpret a high price sensitivityof managers’ stock-based compensation as a signal that the tone of the earnings press releasehas been inflated by the management and therefore investors react less to the tone in earningspress releases as the sensitivity of managers’ compensation increases.

In Panel B of Table 8, we further add the different control variables to the threshold modeland obtain qualitatively similar results. We observe, however, a large decrease in the level ofκ. While κ equals 1.740% in our univariate model, it equals 0.04% in our multivariate setting.Such a low threshold level means that investors systematically discount the tone in earnings pressreleases. We explain this difference by the fact that we cover heterogenous firms in our sample.As we described in Section 3, our sample contains very large and very small firms. Whendividing the firms in three different categories based on size, we observed large differences intheir characteristics. We thus replicate the results of our threshold model based on the sizetertiles. As shown in Models (6)-(8), the average stock price reaction in the first regime is mostimportant for small firms than for large firms. As the firm size increases, the average stock pricereaction and its significance decrease. The coefficient for the Tonej,t variable equals 1.351 forsmall firms and is significant at a 99% confidence level, whereas for large firms it equals 0.236and is significant at a 90% confidence level.

When controlling for firm size, we find that it is only for a minority of the firms that investorsreact proportionally less to the tone when the price sensitivity increases. The tone adjustment byinvestors is present for 39% of the small-cap firms, 25% of the mid-cap and 15% of the large-cap firms. This decrease confirms that tone inflation is more prevalent for smaller firms than forlarger firms.

4.2.1 Positive vs. negative tone

Cheap talk is more likely to appear in an earnings press release with an optimistic tone. Wethus expect rational investors to anticipate this bias in an optimistic rather than in a pessimisticearnings press release. We verify this in Table 9 where we regress the cumulative abnormalreturn around the earnings announcement on positive and negative tone. The linear model showsthat the stock price reaction to negative tone is highly significant and equals 1.032, whereas thestock price reaction to positive tone is substantially lower with a coefficient of 0.289, significantat a 90% confidence level. This difference in coefficients indicates that earnings press releaseswith a negative tone are substantially more informative of future firm performance than thosewith positive tone. In Model (2) and (3), we report the results of the regression based on positiveand negative tone individually and the difference in the Adj.R2 confirms this result: the Adj.R2

for the Model (2) is about half of its negative counterpart in Model (3).Our two-regime-threshold model also shows a clear difference in investors’ reaction to the

tone in function of the sign of the tone in earnings releases. In Model (4), both the negativeand positive Tonej,t variables are significant, but here again the coefficient of the negative tone

30

Page 31: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

variable Tonej,t · I[Tonej,t ≤ 0] is substantially larger than for Tonej,t · I[Tonej,t > 0]. Thedifference in the interaction variables is interesting as only the coefficient for Φj,t·I[Tonej,t > 0]is negative and significant, whereas Φj,t · I[Tonej,t ≤ 0] is positive and insignificant.

In Model (5) and (6), we report the results of the regression based on positive and negativetone individually. All the main variables have the expected sign and are significant, but the maindifference appears in the level of κ: investors start discounting the tone of an earnings pressrelease at a much lower level of relative compensation (∆j,t−1) than for pessimistic earningspress releases. While investors start discounting the tone at a level of relative compensationof 2.140% for an earnings press release with a positive tone, they start discounting the tone ata level of relative compensation of 29.640% for firms with a negative tone. For the former,21.682% of the observations are not discounted by investors at earnings announcement and thelatter includes 93.452% of the sample. This difference confirms the intuition that cheap talkappears in optimistic earnings press releases more than in pessimistic releases and that investorsreact proportionally less to the tone in earnings press releases as the sensitivity of managers’compensation increases.

4.2.2 Robustness checks

In Subsection 4.1.2 we showed that the positive impact of equity-based compensation on tone in-flation is robust to the choice of the library used and the frequency of the data analyzed. In Table10 we now examine whether the results on the interaction effect of equity-based compensationand tone inflation on stock return on the announcement day is robust to the same methodologicalchoices.

Panel A of Table 10 shows that the choice of the library does change our results. The tone ofearnings press releases provides investors with information that is incremental to hard financialdata. The tone measure based on the Henry (2008) library shows the highest Adj.R2 with acoefficient of 0.599. The results for the two-regime-threshold model indicate that investors reactproportionally less to the tone in earnings press releases as the sensitivity of managers’ compen-sation increases. The coefficient for the Φj,t variable is negative and highly significant acrossall libraries. In addition, we observe that the Adj.R2 of the threshold model is substantiallylarger than for the linear model used in prior literature. This is particularly the case for the tonemeasure based on the DICTION library, which nearly doubles from 0.7 to 1.3%.

We show in Panel B of Table 10 that segregating the sample in quarters does not alter our con-clusion that a high level of equity-based compensation tends to reduce the positive impact of theoptimism signal in earnings press releases on the stock’s return on the day of the announcement.

31

Page 32: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

Table 9: The market response to positive vs. negative tone in earnings press releases andmanagers’ equity incentives – Multivariate linear and threshold models

Panel A – Linear model Panel B – Threshold modelCARj,t Model 1 Model 2 Model 3 Model 4 Model 5 Model 6

κ (%) - - - 2.140 2.140 29.640∆κ - - - 0.456 0.456 2.774% I

[∆j,t−1 ≤ κ

]- - - 21.682 21.682 93.452

(Intercept) 1.386 0.906 1.528 1.386 0.935 1.507(1.243) (1.187) (1.250) (1.276) (1.230) (1.249)

Tonej,t · I[Tonej,t > 0] 0.289∗ 0.607∗∗∗ 0.598∗∗∗ 0.941∗∗∗

(0.156) (0.142) (0.192) (0.176)Tonej,t · I[Tonej,t ≤ 0] 1.032∗∗∗ 1.154∗∗∗ 0.969∗∗∗ 1.137∗∗∗

(0.264) (0.244) (0.265) (0.243)Φj,t · I[Tonej,t > 0] -0.465∗∗∗

(0.134)Φj,t · I[Tonej,t ≤ 0] 0.062 -0.529∗∗∗ -3.035∗∗∗

(0.198) (0.132) (0.664)

Control variables Yes Yes Yes Yes Yes YesYear fixed-effects Yes Yes Yes Yes Yes YesIndustry fixed-effects Yes Yes Yes Yes Yes Yes

R2 0.023 0.015 0.022 0.026 0.019 0.023Adj. R2 0.015 0.007 0.014 0.017 0.011 0.015Num. obs. 2795 2795 2795 2795 2795 2795

Note: This table presents the results of investors’ response to the tone in earnings press releases and the price sensi-tivity of managers’ equity incentives and distinguishes between the positive and negative tone. Panel A reports theestimation results of Equation 4.2, with the control variables. Panel B reports the estimation results of Equation 4.3,with the control variables. Φj,t is the interaction variable between Tonej,t and ∆j,t−1 and is defined in Equation4.4. Tonej,t is measured as the standardized average of the tone measures obtained using three established lists ofwords, namely: the positive and negative wordlists defined by Henry (2008), those defined by Loughran and Mc-Donald (2011) and the word list in the DICTION 7.0 software. Each variable is defined in Table 1. The significanceof the coefficients is tested using Newey-West standard errors. ∗, ∗∗, and ∗∗∗ denote statistical significance at the 10percent, 5 percent, and 1 percent levels, respectively, based on a two-sided t-test.

32

Page 33: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

Table 10: Robustness checks of Hypothesis 2 – The market response to net tone in earningspress releases and managers’ equity incentives – Multivariate linear and thresh-old models

Panel A – By libraryLinear model Threshold model

Score Henry (2008) DICTION 7.0 L&M Score Henry (2008) DICTION 7.0 L&M

κ (%) - - - - 0.040 2.140 0.040 6.140∆κ - - - - 0.003 0.456 0.003 1.605% I

[∆j,t−1 ≤ κ

]- - - - 0 21.682 0 60.143

(Intercept) 1.128 0.171 1.184 1.217 1.217 0.314 1.246 1.273(1.208) (1.215) (1.193) (1.226) (1.239) (1.273) (1.220) (1.250)

Tonej,t 0.355∗∗∗ 0.599∗∗∗ 0.442∗∗∗ 0.519∗∗∗ 1.021∗∗∗ 0.732∗∗∗ 2.284∗∗∗ 0.473∗∗∗

(0.065) (0.088) (0.131) (0.127) (0.314) (0.095) (0.617) (0.130)Φj,t -0.131∗∗ -0.245∗∗∗ -0.363∗∗∗ -0.341∗

(0.057) (0.053) (0.110) (0.204)

Control variables Yes Yes Yes Yes Yes Yes Yes YesYear fixed-effects Yes Yes Yes Yes Yes Yes Yes YesIndustry fixed-effects Yes Yes Yes Yes Yes Yes Yes Yes

R2 0.021 0.023 0.015 0.017 0.024 0.029 0.021 0.018Adj. R2 0.013 0.016 0.007 0.009 0.016 0.021 0.013 0.010Num. obs. 2795 2795 2795 2795 2795 2795 2795 2795

Panel B – By quarterLinear model Threshold model

Quarter 1 Quarter 2 Quarter 3 Quarter 4 Quarter 1 Quarter 2 Quarter 3 Quarter 4

κ (%) - - - - 2.924 27.807 0.224 2.207∆κ - - - - 0.289 2.152 0.022 0.492% I

[∆j,t−1 ≤ κ

]- - - - 31.690 92.471 1.224 23.140

(Intercept) −1.557 −2.147 1.681 −3.660 −1.440 −1.976 1.943 −3.534(3.372) (2.207) (1.591) (2.612) (3.395) (2.222) (1.568) (2.595)

Tonej,t 0.797∗∗∗ 0.923∗∗∗ 0.862∗∗∗ 1.073∗∗∗ 1.091∗∗∗ 0.954∗∗∗ 2.028∗∗∗ 1.177∗∗∗

(0.179) (0.146) (0.166) (0.151) (0.253) (0.147) (0.471) (0.161)Φj,t -0.397∗∗ -3.832∗∗∗ -0.390∗∗∗ -0.503∗∗∗

(0.182) (1.074) (0.131) (0.154)

Control variables Yes Yes Yes Yes Yes Yes Yes YesYear fixed-effects Yes Yes Yes Yes Yes Yes Yes YesIndustry fixed-effects Yes Yes Yes Yes Yes Yes Yes Yes

R2 0.023 0.015 0.009 0.022 0.025 0.018 0.011 0.024Adj. R2 0.015 0.011 0.005 0.018 0.016 0.013 0.006 0.020Num. obs. 2779 5443 5473 5500 2779 5443 5473 5500

Note: This table presents the robustness checks of Hypothesis 2 and 3. The estimation results are for Equation 4.2 and4.3. In Panel A, we verify the robustness of our results with respect to the library used to measure tone. We measureTonej,t using three established lists of words, namely: the positive and negative wordlists defined by Henry (2008),those defined by Loughran and McDonald (2011) and the word list in the DICTION 7.0 software. We also report theresults for the principal component score obtained on those three libraries. Panel B reports the regression results byquarter. Each variable is defined in Table 1. The significance of the coefficients is tested using Newey-West standarderrors. ∗, ∗∗, and ∗∗∗ denote statistical significance at the 10 percent, 5 percent, and 1 percent levels, respectively,based on a two-sided t-test. 33

Page 34: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

5 Conclusion

Managers whose portfolio value depends on the firm’s stock price wear two hats while writingthe earnings press release. On the one hand, as the shareholders’ agent, their goal is the reductionof information asymmetries. In this case, one could expect managers to disclose a crediblesignal regarding their future-earnings expectations. On the other hand, as an investor, they mayopportunistically use these voluntary disclosures to influence investors’ expectations of futureearnings and maximize their stock and option portfolios. As earnings press releases are oftenconsidered as a major news event for many firms and are generally accompanied by a largemarket response, we investigate whether managers inflate the tone of the narrative section ofearnings press releases to maximize the value of their stock and option compensation.

We analyze the earnings press releases of S&P 1500 firms between 2004 and 2011 and findevidence that managers inflate the tone of their earnings press releases as the sensitivity of theirstock-based compensation to the share price increases. We show that this effect cannot be ex-plained by differences in past performance, industry and year fixed effects. In addition, we findthat it is the CEO’s portfolio sensitivity to stock price that is the primary driving force behindthe observed tone inflation. We thus conclude that measures of tone in earnings press releasesdo not simply reflect the economic events of the firm but also managers’ equity incentives.

As predicted by the Almost-cheap talk model of Kartik (2009), we find that managers’ toneinflation in earnings press releases is not completely costless to them. We define a two-regime-threshold model, where in the first regime, the price sensitivity of managers’ stock-based com-pensation is small. In the second regime, the value of the managerial portfolio has a relativelyhigh sensitivity to the stocks price and investors seem to anticipate the managerial incentive toinflate the tone in the earnings press release, as manifested by the fact that stock returns on theearnings press release reacts proportionally less to the tone when the price sensitivity of man-agers’ compensation increases. We also show that cheap talk is more prevalent among smallerfirms.

Our results complement prior research on the impact of managers’ incentives on earningsmanagement practices within the firm and take the literature on the textual analysis of corporatedisclosures in a new direction. While the general perception from prior literature is to interpretthe tone of earnings press releases as a pure signal of future performance, we show that thetone of earnings press releases (and more generally of corporate disclosures) should also beinterpreted as a tool to drive up stock performance at earnings announcement. Future researchmight be focused on the interplay between tone manipulation and earnings management.

34

Page 35: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

6 Appendix

6.1 Parsing earnings press releases

As in most studies using textual analysis in finance (see e.g. Davis et al., 2012; Davis and Tama-Sweet, 2012; Henry, 2008; Loughran and McDonald, 2011), we use the bag of words methodthat requires to parse the 8-K documents into vectors of words.12 Each complete text filing isthus read into a single string variable and parsed using the following sequence:

• Remove graphics (ASCII encoded graphics) – embedded graphics increase by orders ofmagnitude the character count and file size of the 8-K documents. All encoded graphicsare purged.

• Re-encode characters – translates "encoded" characters such as &NBSP (blank space) or&AMP (&) back to their original ACSII form.

• Remove tables – we remove all text appearing within <TABLE> HTML tags, where morethan 10% of the nonblank characters are numbers. (Note that this filter is important sincesome filers embed all of their text within TABLE tags.)

• Remove HTML – the quantity of HTML increased substantially over the 2004 and 2011period. Some earnings press releases include more HTML than actual content.

• Parse into tokens – we use a regular expression (regex) to parse the remaining string vari-able into all collections of two or more alphabetic characters. (Hyphens are also allowedin the character collections.) We first replace all hyphens followed by a line-feed with ahyphen so that the word boundary regex works.

12We process the 8-K filings using Python scripts.

35

Page 36: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

6.2 Scatter plots of the tone measures

Figure 3: Scatter plots of the tone measures using the DICTION 7.0, Henry (2008) andLoughran and McDonald (2011) libraries to classify positive and negative words

−3 −2 −1 0 1 2 3

−2

−1

01

23

45

Tone − Henry (2008)

Tone

− L

ougr

an &

McD

onal

d (2

011)

Corr = 0.59

(a) Tone – Loughran and McDonald (2011) vs.Henry (2008)

−3 −2 −1 0 1 2 3

−2

−1

01

23

Tone − DICTION 7.0

Tone

− L

ougr

an &

McD

onal

d (2

011)

Corr = 0.54

(b) Tone – Loughran and McDonald (2011) vs.DICTION 7.0

−2 −1 0 1 2 3 4 5

−2

−1

01

23

Tone − DICTION 7.0

Tone

− H

enry

(20

08)

Corr = 0.56

(c) Tone – Henry (2008) vs. DICTION 7.0

36

Page 37: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

ReferencesAtiase, R. (1980). Predisclosure informational asymmetries, firm capitalization, financial reports and

security price behavior.Atiase, R. (1985). Predisclosure information, firm capitalization and security price behaviour around

earnings announcements. Journal of Accounting Research 23, 21–36.Baginski, S., E. Demers, C. Wang, and J. Yu (2011). Understanding the role of language in management

forecast press releases. INSEAD Working Paper.Baker, M. and J. Wurgler (2002). Market timing and capital structure. Journal of Finance 57, 1–32.Balsam, S., E. Batov, and C. Marquard (2002). Accrual management, investor sophistication and equity

valuation: Evidence from 10-Q filings. Journal of Accounting Research 40, 987–1012.Bergstresser, D. and T. Philippon (2006). CEO incentives and earnings management. Journal of Financial

Economics 80, 511–529.Black, F. and M. Scholes (1973). The pricing of options and corporate liabilities. Journal of Political

Economy 81, 631–654.Brouthers, K., L. Brouthers, and S. Werner (2000). Influences on strategic decision-making in the Dutch

financial services industry. Journal of Management 26, 863–883.Burgoon, J., J. Blair, and R. Strom (2008). Cognitive biases and nonverbal cue availability in detecting

deception. Human Communication Research 34, 572–599.Burns, N. and S. Kedia (2006). The impact of performance-based compensation on misreporting. Journal

of Financial Economics 79, 35–67.Cheng, Q. and T. Warfield (2005). Stock-based compensation, earnings management and the informa-

tiveness of earnings. The Accounting Review 80, 441–476.Coles, J., N. Daniel, and L. Naveen (2006). Managerial incentives and risk-taking. Journal of Financial

Economics 79, 431–468.Core, J. and W. Guay (2002). Estimating the value of employee stock option portfolios and their sensi-

tivites to price and volatility. Journal of Accounting Research 40, 613–630.Crawford, V. and J. Sobel (1982). Strategic information transmission. Econometrica 50, 1431–1451.Cyert, R., S. Kang, and P. Kumar (2002). Corporate governance, takeovers and top-management com-

pensation: Theory and evidence. Management Science 48, 453–469.Daniel, N., Y. Li, and L. Naveen (2013). No asymmetry in pay for luck. Working paper, Drexel University.Das, S., C. B. Levine, and K. Sivaramakrishnan (1998). Earnings predictability and bias in analysts’

earnings forecasts. The Accounting Review 73, 277–294.Davis, A., J. Piger, and L. Sedor (2012). Beyond the numbers: Measuring the information content of

earnings press release language. Contemporary Accounting Research 29, 845–868.Davis, A. K. and I. Tama-Sweet (2012). Managers’ use of language across alternative disclosure outlets:

Earnings press releases versus MD&A. Contemporary Accounting Research 29, 804–837.Dechow, P., R. Sloan, and A. Sweeney (1996). Causes and consequences of earnings manipulation: An

analysis of firms subject to enforcement actions by the SEC. Contemporary Accounting Research 13,1–36.

Demers, E. and C. Vega (2010). Soft information in earnings announcements: news or noise? Workingpaper, INSEAD.

Edmans, A., X. Gabaix, and A. Landier (2009). A multiplicative model of optimal ceo incentives inmarket equilibrium. Review of Financial Studies 22, 4881–4917.

Engelberg, J. (2008). Costly Information Processing: Evidence from Earnings Announcements. Workingpaper, Northwestern University.

37

Page 38: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

Evans, J., J. Barston, and P. Pollard (1983). On the conflict between logic and belief in syllogisticreasoning. Memory and Cognition 11, 295–306.

Fama, E. and K. French (1992). The cross-section of expected stock returns. Journal of Finance 47,427–465.

Fischer, P. and R. Verrecchia (2000). Reporting bias. The Accounting Review 2, 229–245.Forst, C. (1997). Disclosure policy choices of UK firms receiving modified audit reports. Journal of

Accounting and Economics 23, 163–188.Gao, P. and R. Shrieves (2002). Earnings management and executive compensation: A case of overdose of

option and underdose of salary? Working paper, Northwestern University and University of Tennesseeat Knoxville.

Gaver, J., K. Gaver, and J. Austin (1995). Additional evidence on bonus plans and income management.Journal of Accounting and Economics 19, 3–28.

Gilbert, D., R. Tafarodi, and P. Malone (1993). You can’t not believe everything you rread. Journal ofPersonality and Social Psychology 65, 221–233.

Gordon, E., E. Henry, M. Peytcheva, and L. Sun (2008). Disclosure credibility and market reaction torestatements. Working paper, University of Miami.

Graham, J. and C. Harvey (2001). The theory and practice of corporate finance: evidence from the field.Journal of Financial Economics 60, 187–243.

Gray, S. and A. Cannella (1997). The role of risk in executive compensation. Journal of Management 23,517–540.

Green, J. R. and N. L. Stokey (2007). A two-person game of information transmission. Journal ofEconomic Theory 127, 90–104.

Henry, E. (2008). Are investors influenced by how earninigs press releases are written? Journal ofBusiness Communications 45, 363–407.

Henry, E. and A. Leone (2009). Measuring qualitative information in capital markets research. Workingpaper, University of Miami.

Huang, X., S. H. Teoh, and Y. Zhang (2013). Tone management. The Accounting Review Forthcoming.Hubbard, R. and D. Palia (1995). Executive pay and performance: Evidence from the u.s. banking

industry. Journal of Financial Economics 39, 305–360.Hutton, A., G. Miller, and D. Skinner (2003). The role of supplementary statements with management

earnings forecasts. Journal of Accounting Research 41, 867–890.Jennings, R. (1987). Unsystematic security price movements, management earnings forecasts, and revi-

sions in consensus analyst earnings forecasts. Journal of Accounting Research 25, 90–110.Jensen, M. and K. Murphy (1990). Performance pay and top-management incentives. Journal of Political

Economy 98, 225–264.Kartik, N. (2009). Strategic communiction with lying costs. Review of Economic Studies 76, 1359–1395.Ke, B. (2003). The influence of equity-based compensation on CEO’s incentives to report strings of

consecutive earnings increases. Working paper, Penn State University.King, R. (1988). The liar’s discount. Forbes May 30.Latham, M. (1986). Informational efficiency and information subsets . Journal of Finance 41, 39–52.Lin, H. and M. F. McNichols (1998). Underwriting relationships, analysts’ earnings forecasts and invest-

ment recommendations. Journal of Accounting and Economics 25, 101–127.Loughran, T. and B. McDonald (2011). When is a liability not a liability? Textual analysis, dictionaries,

and 10-Ks. The Journal of Finance 66, 35–65.

38

Page 39: Almost-cheap talk in earnings press releasesswfa2015.uno.edu/B_Asset_Pricing_II/paper_78.pdfAlmost-cheap talk in earnings press releases: Textual tone and managerial equity incentivesI

Markovits, H. and G. Nantel (1989). The belief-bias effect in the production and evaluation of logicalconclusions. Memory and Cognition 17, 11–17.

Markowitz, H. (1952). Portfolio selection. The Journal of Finance 7, 77–91.Merton, R. (1973). Theory of rational option pricing. Bell Journal of Economics and Management

Science 4, 141–183.Michaely, R. and K. Womack (1999). Conflict of interest and the credibility of underwriter analyst

recommendations. Review of Financial Studies 12, 653–686.Miller, G. (2002). Earnings performance and discretionary disclosure. Journal of Accounting Re-

search 40, 173–204.Mohd, E. (2005). Accounting for software development costs and information asymmetry. The Account-

ing Review 80, 1211–1231.Murphy, K. (1985). Corporate performance and managerial remuneration: An empirical analysis. Journal

of Accounting and Finance 7, 11–42.Murphy, K. (1986). Incentives, learning, and compensation: A theoretical and empirical investigation of

managerial labor contracts. Rand Journal of Economics 17, 59–76.Murphy, K. (1999). Executive compensation. In Orley, A., Cards, D. (Eds.), Handbook of Labor Eco-

nomics. North-Holland, Amsterdam,.Peng, L. and A. Roell (2003). Executive pay, earnings manipulation and shareholder litigation. Working

paper, Baruch College/CUNY and Princeton University.Price, S., J. Doran, D. Peterson, and B. Bliss (2012). Earnings conference calls and stock returns: The

incremental informativeness of textual tone. Journal of Banking & Finance 36, 992 – 1011.Richardson, V. (2000). Information asymmetry and earnings management: Some evidence. Review of

Quantitative Finance and Accounting 15, 325–347.Rogers, J., A. Van Buskirk, and S. Zechman (2011). Disclosure tone and shareholder litigation. The

Accounting Review 86, 2155–2183.Roulstone, D. (2003). Analyst following and market liquidity. Contemporary Accounting Research 20,

552–578.Salomon, D. (2012). Selective publicity and stock prices. Journal of Finance 67, 599–637.Smith, A. and R. Watts (1992). The investment opportunity set and corporate financing, dividend and

compensation policies. Journal of Financial Economics 32, 1935–1974.Smith, J. (2012). Can investors fully adjust for known biases in manager communications? Working

paper, McCombs School of Business.Stein, J. (1989). Efficient capital markets, inefficient firms: A model of myopic corporate behavior. The

Quarterly Journal of Economics 104, 655–669.Thomas, S. (2002). Firm diversification and asymmetric information: Evidence from analysts’ forecasts

and earnings announcements. Journal of Financial Economics 64, 373 – 396.Trautmann, B. and G. Hamilton (2003). Informal corporate disclosure under federal securities law: Press

releases, analyst calls and other communications. Chicago: CCH.Zhang, X., K. Bartol, K. Smith, M. Pfarrer, and D. Khanin (2008). CEOs on the edge: Earnings manipu-

lation and stock-based incentive misalignment. The Academy of Management Journal 51, 241–258.

39