S06-Management Deception PDF

Management Deception, Big-Bath Accounting, and Information Asymmetry:
Evidence from Linguistic Analysis
Ole-Kristian Hope
Rotman School of Management
University of Toronto
[email protected]
Jingjing Wang
Rotman School of Management
University of Toronto
[email protected]
February 26, 2018
Acknowledgements:
We thank the Michael Lee-Chin research grant for financial support. We thank Stefan Anchev,
Muhammad Azim, Stephanie Cheng, Mahfuz Chy, Bingxu Fang, Shushu Jiang, Ross Lu, Gordon
Richardson, Barbara Su, Aida Wahid, Eyub Yegen, Wuyang Zhao, two anonymous reviewers, and
seminar participants at the Rotman School of Management and the Canadian Academic
Accounting Association for their comments and inputs. Hope gratefully acknowledges funding
from the Deloitte Professorship.
Electronic copy available at: https://ssrn.com/abstract=3003259

ABSTRACT
Accounting big baths are pervasive in practice. While big baths can improve the information
environment and reduce information asymmetry, they can also degrade the information
environment and obscure operating performance. In this study, we examine the role of
management ethics. Specifically, we investigate whether managers’ truthfulness (or conversely,
deceptiveness) affects how investors perceive big baths. Using linguistic analysis on earnings-
conference calls to measure managerial deception and employing a difference-in-differences
research design with propensity-score matching, we find that information asymmetry is
significantly higher following big baths taken by deceptive CEOs, compared with big baths taken
by less deceptive CEOs.
Keywords: Management Deception; Big-Bath Accounting; Linguistic Analysis; Information
Asymmetry

1. Introduction
How does a firm’s information environment change after accounting big baths? Prior literature
provides evidence on both positive (Elliott and Shaw 1988; Francis, Hanna, and Vincent 1996;
Haggard, Howe, and Lynch 2015) and negative (Moore 1973; Kirschenheiter and Melumad 2002;
Kothari, Shu, and Wysocki 2009; Bens and Johnston 2009) consequences of big baths on the
information environment. As managers have discretion regarding whether to incur a large write-
off, and can decide the timing and amount of the write-off, management’s incentives are important
in studying the effects of big baths. However, such incentives are unobservable. Investors may use
managerial characteristics to infer management incentives. Among the most salient managerial
characteristics in this setting is truthfulness; thus this study examines how truthfulness (or
conversely, deceptiveness) affects investors’ perceptions of big baths.
According to upper-echelons theory (Hambrick and Mason 1984), the ethical attentiveness
throughout the organization is instilled by its leaders (Patelli and Pedrini 2015). A series of
accounting fraud scandals over the last decades put leadership ethics at the forefront of the heated
debate on financial-reporting truthfulness (Tourish and Vatcha 2005; Mihajlov and Miller 2012).
Ethics is an intrinsic part of managers’ behavior (Solomon 1992). As firms’ high-level decision
makers, top managers are likely to follow a cognitive and rational approach that revolves around
moral judgments about the issues when making ethical decisions, just as an individual making a
choice when facing an ethical dilemma (Kohlberg 1981; Rest 1986; Weber 1990; Vitell, Lumpkin,
and Rawwas 1991; Reynolds 2006; Albert, Scott, and Turan 2015). Big baths are managerial
decisions that can be the result of managers’ ethical considerations of the firms’ welfare, or can be

the result of managers’ incentives to maximize their personal utility. Being truthful or deceptive to
investors and other stakeholders also indicates management’s ethical choice of how they view their
responsibility to the firm’s stakeholders.
On one hand, big baths can manifest themselves as exceptionally large negative discretionary
accruals. On the other hand, big baths can consist of one-time, large write-offs, and may include
restructuring charges, asset impairments, and litigation losses. These write-offs are generally
reflected as “special items” in the financial statements. There are two ways to look at a big bath.
If a company reports a loss that is larger than expectations, it could be the case that there are certain
issues within the firm that warrant such actions and that managers are truthfully conveying such
information to the capital market and other stakeholders. In line with that view, some analysts
interpret big baths as managers’ positive response to existing problems (Elliot and Shaw 1988).
Big baths can also “clear the air” (Haggard et al. 2015). That is, by writing off assets when their
carrying values are less than the market values, the reported values of the assets are realigned with
their economic values. As a result, firm-level information asymmetry following a big bath should
decrease.
However, big baths are sometimes used as an earning-management technique to shift current
earnings to future periods. As Levitt (1998) points out, if big-bath charges are overly
conservatively estimated with “extra cushioning,” they can miraculously be reborn as income
when future earnings fall short. Big baths can also be used to secure bonus payments. Often,
managers’ rewards are tied to meeting certain performance targets. In an economic downturn,
managers may follow the big-bath approach by bundling as much bad news into the current period
as possible, aiming to make their targets easier to achieve in the next period.1 In cases of new
1
This big-bath approach is discussed in the article “why honesty is the best policy” in the special report section of
The Economist, March 7, 2002.

management, big-bath accounting can be used to mitigate top executives’ job-security concerns.2
The new manager can benefit from taking a big bath, blaming the low earnings on the previous
manager so as to display an improved financial performance in future (Moore 1973).
From investors’ perspectives, the action of taking big baths by a firm is observable, but the
motivations behind the action are not entirely clear. Hou (2015) finds that in a well-diversified
market, idiosyncratic information risk is priced when information is subject to managers’
discretion and thus ambiguous. Can investors infer the motivations of managers by observing their
types – truthful or deceptive – and associating their actions with the types? By taking advantage
of newly-developed technologies, investors are now analyzing the linguistic patterns displayed in
management speech. Investors have been using algorithmic textual analysis, CIA lie-detection
techniques, and more recently, audio analysis of management speech, to seek an edge with stock
calls, sector sentiment, and overall market direction.3
Earnings-conference calls, in which managers discuss their firms’ financial performance with
analysts and investors, are important information sources to search for signs of management
deception. If a manager is discovered to be deceptive when discussing her firm’s financial results,
and the firm takes a big bath at the time, will investors perceive this big bath to have low
credibility? Will investors associate this big bath with such motivations as meeting earnings targets
or securing bonus payments? If this is the case, we would expect to observe an increase in
information asymmetry following big baths taken by deceptive managers. Conversely, if a firm’s
manager takes a big bath and she is considered truthful, investors may perceive this big bath as
having high credibility. In this case, the information asymmetry may decrease.
A primary reason both practicing accountants and researchers care about information
2
“Some new bank CEOs take an earnings bath when they start,” The Wall Street Journal, March 3, 2014.
3
A discussion on the topic of “how to tell if a CEO is lying” can be found at http://www.CNBC.com, July 7, 2015.

asymmetry is that it reflects the information environment.4 Clearly an important issue related to
big baths is whether these accounting events improve or deteriorate the firm’s information
environment. However, to broaden the scope of our study, in additional analyses we also test for
trading volume (another commonly employed outcome variable in this line of literature).
This study builds on Larcker and Zakolyukina (2012) who find that certain words are
significantly associated with management deception. For instance, deceptive CEOs use more
“reference to general knowledge” and “extreme positive emotion” words, and use fewer “anxiety”
and “shareholder value” words. The linguistic approach proposed by Larcker and Zakolyukina
(2012) is based on psychological theories linking deception to linguistic behavior (Vrij 2008), and
is built up by applying a well-developed and frequently used psychosocial dictionary – LIWC.
There are an increasing number of applications of LIWC analyses in deception detection, and also
in personality, forensic, clinical, relationship, and cultural assessments (Chung and Pennebaker
2012). Providing further validity to Larcker and Zakolyukina (2012), Loughran and McDonald
(2013) demonstrate that the credibility of managers is diminished by having an overly positive S-
1 in the IPO process, consistent with Larcker and Zakolyukina (2012) who find that deceptive
CEOs use significantly more positive emotion words in conference calls. Another piece of
evidence substantiating the usefulness of CEOs’ linguistic patterns in signifying deception is
provided by Hobson, Mayew, Peecher, and Venkatachalam (2017), who demonstrate that once
given instructions on the "cognitive dissonance" in the CEOs remarks, auditors are able to more
precisely detect fraudulent companies as well as the unidentified "red flag" sentences in earnings-
conference calls.
4
For example, Haggard et al. (2015) use the terms information environment and information asymmetry
interchangeably.

We use Larcker and Zakolyukina’s (2012) approach to identify truthful and deceptive
managers in this paper, and examine whether the change in information asymmetry around a big-
bath event is a function of managerial deception. As CEOs play the largest role in corporate
decision making, we focus on the linguistic pattern of CEOs.
In our primary analyses we employ a difference-in-differences research design coupled with
propensity-score matching of treatment and control firms. This approach provides strong control
for potential confounding events as well as omitted-variable biases. We find evidence that investors
are able to discern managers’ deception levels from conference calls and that information
asymmetry is affected accordingly. Specifically, we find that information asymmetry (proxied
using Amihud’s 2002 illiquidity measure and bid-ask spreads) increases significantly after big
baths taken by deceptive CEOs as compared to those taken by less deceptive CEOs. In additional
analyses, we find that this effect is more pronounced when a CEO who has been truthful in the
past becomes deceptive in the bath year. Our inferences are robust to a variety of regression
specifications and other robustness tests.
The study adds to the big-bath line of research by examining a potentially important factor
that could affect the impact of big-bath taking on information asymmetry. Second, our study
contributes to research on how investors use ex-ante credibility of CEOs to interpret financial
reporting quality (e.g., Loughran and McDonald 2013). We do this by applying textual-analysis
techniques to accounting issues to infer management’s intentions using subtle linguistic cues.
Finally, the study adds to management-ethics research by examining management deception and
its associated capital-market consequences in the setting of big-bath taking, an economically

important event as documented in prior research.5 This paper thus also contributes to the literature
by studying the financial outcomes of CEO idiosyncratic characteristics and psychological patterns.
2. Prior Literature and Hypothesis Development
2.1 Big-Bath Literature
Prior research has examined the timing, motivation for, and consequences of taking big baths.
Moore (1973) finds that discretionary accounting decisions that reduce income are more likely to
be made in a period of management changes. The write-offs, many of which have substantial
economic consequences, reflect decisions by corporate management. Kirschenheiter and Melumad
(2002) construct a theoretical model to demonstrate that managers under-report earnings the most
when the news is sufficiently “bad.” Management, on average, delays the release of bad news to
investors (Kothari, Shu, and Wysocki 2009). Mendenhall and Nichols (1988) find that most bad
news, including large write-offs, takes place in the fourth quarter and that the market reaction for
fourth-quarter bad news is smaller than the reaction to similar news in other quarters.6
On one hand, big baths can be motivated by the desire to manipulate earnings. On the other
hand, big baths may be used to reflect declines in the values of assets due to poor performance,
increased market competition, and changes in the economic environment. Managers can make use
of big baths to turn adversity into financial advantage. Large accounting write-offs often follow
5
Maak and Pless (2006) call for research that focuses more on the identification and measurement of leadership styles
that lead to responsible leadership. The extent to which CEOs influence accounting choices is fundamental to the
understanding of how organizations work, but this linkage is poorly understood (Mackey 2008).
6
There are different types of write-offs. Write-offs in PP&E and inventory accounts are typically considered less
reflective of earnings manipulation, while restructuring charges and write-offs of goodwill are considered more
reflective of such a motif (Francis, Hanna, and Vincent 1996). Similarly, write-offs of long-lived assets after the
adoption of SFAS 121 are less reflective of firms’ fundamentals and such big baths are associated with managers’
opportunistic actions (Riedl 2004). Bens and Johnston (2009) find that before EITF No. 94-3, when fewer codified
rules existed regarding restructuring charges, managers were more likely to engage in earnings management by
recognizing overly large charges.

CEO turnover (Strong and Meyer 1987; Murphy and Zimmerman 1993), as new CEOs can blame
the losses taken on their predecessors and take credit for subsequent improvements in reported
profitability. Not only do CEOs have incentives to maximize their bonus payments, they also wield
significant influence over the firm’s reported financial results and may use discretionary accruals
to increase compensation or maintain their positions in firms.
However, the motivation for and consequences of big baths may not all be negative. Strong
and Meyer (1987) discuss how write-down decisions may be used by managers to provide signals
to investors that actions are taken to eliminate those assets generating little or no return. By
cleaning up the balance sheet and reducing reported equity, a company can boost future profits and
increase its per-share return. However, a series of write downs can erode investors’ confidence in
management and induce declines in a firm’s stock price. Elliot and Shaw (1988) discuss the
favorable-resolution hypothesis in which write-offs reflect managers’ acknowledgement of
existing problems and their constructive responses. Consistent with that view, the financial press
often interprets large write-offs positively.
Christensen, Paik, and Stice (2008) examine management’s motifs when faced with the
opportunity of making a big bath even larger. They find that after SFAS 109, which prescribes the
establishment of a deferred-tax valuation allowance when relevant, the majority of the larger-than-
expected valuation allowances reflect informed management pessimism about the future of the
firms, as such firms experience poorer operating performance in subsequent periods. Finally, and
in a similar spirit, Haggard, Howe, and Lynch (2015) examine the information environment of
firms following large, non-recurring charges and find that earnings become smoother, firm-level
information asymmetry decreases, and stock returns become more responsive to unexpected
earnings.

2.2 Management Ethics
CEOs are the primary decision makers within firms, and they go through a moral decision-
making process and trade off costs and benefits when they face ethical choices. Several factors
play a role in influencing executives’ ethical behavior, such as the behavior of superiors (for CEOs
this could potentially be interpreted as the board), the ethical climate of the industry, the behavior
of one’s equals, the lack of formal company policy, and one’s personal financial needs (Baumhart
1961; Ekin and Tezölmez 1999).7 Importantly, managers are given the power of discretion. In the
case of high-ranking business executives, discretion to do good is also discretion to do bad (Carson
2003). Managers often face an ethical dilemma in reporting decisions (Evans, Hannan, Krishnan,
and Moser 2001; Liu, Wright, and Wu 2015). There is a social standard of ‘‘doing the right thing’’
and not acting in a deceptive, unethical manner, but at the same time, managers’ incentive to
maximize self-interests may lead to actions that are detrimental to the firms’ stakeholders.8
Extant research in management emphasizes the role of the CEO as the key driver of corporate
strategy (Hambrick and Mason 1984; Zona, Minoja, and Coda 2013). “Tone at the top” not only
affects corporate strategy, but also influences financial reporting choices. Amernic, Craig, and
Tourish (2010) point out that a fundamental and direct manifestation of tone at the top is the
narrative language of the CEO. Craig, Mortensen, and Iyer (2013) examine the linguistic features
of Ramalinga Raju, the main individual involved in the biggest corporate scandal in India, and find
that his word choices changed noticeably in his five annual-report letters prior to the collapse of
his firm, as the scale of his financial misstatements increased. After analyzing 535 annual CEO
letters to shareholders with DICTION, a computer-based language analysis program, Patelli and
7
In a recent study, Cardinaels (2016) provides experimental evidence about how the interaction between a company’s
earnings and its information system influences the degree of honest reporting by managers.
8
Top management is often identified as key antecedents of corporate fraud (Baucus 1994; Efendi, Srivastava, and
Swanson 2007; Khanna, Kim, and Lu 2015; Chen, Cumming, Hou, and Lee 2016).
10

Pedrini (2015) find that aggressive financial reporting is associated with three thematic indicators
in DICTION, namely Certainty, Realism, and Commonality.
2.3 Linguistic Analysis and the Conference-Call Literature
With technological development, investors and analysts are now able to look at both verbal
and non-verbal cues that are indicative of management deception, not only from 10-Ks and 10-Qs,
but also from press releases, conference calls, and other information sources. Such information has
become increasingly important for investors to make decisions and for analysts to make earnings
forecasts and stock recommendations.
Importantly, linguistic analysis can be applied to assess the likelihood of management
deception. One stream of linguistic analysis is based on existing dictionaries or the modification
of these dictionaries (also known as the “bag-of-words” approach). The most commonly used
dictionaries are Harvard General Inquiry, LIWC, Diction, and Loughran and McDonald (2011).
This “bag-of-words” approach, even though it does not take into account the actual and contextual
meaning of words, has the advantages of being easy to understand, implement, and replicate.
Subjectivity is also removed by relying on well-established dictionaries. In a well-cited study,
Loughran and McDonald (2011) examine whether negative, uncertainty, and litigious words in 10-
Ks can predict 10b-5 fraud lawsuits, after weighting these words to account for rarity. Using
linguistic inquiry and word count (LIWC) software, Newman, Pennebaker, Berry, and Richards
(2003) find a set of “lying words” that identify deceptive language in a variety of experimental
settings. Most important for this study, Larcker and Zakolyukina (2012) employ LIWC as well as
11

several self-constructed word categories to predict misstatements from linguistic features of CEO
speech during quarterly earnings-conference calls.9
In terms of information sources to analyze management linguistic cues, conference calls have
been used extensively by researchers, as these conference calls are essential in resolving the
information asymmetry between firms and outside investors. Earnings-conference calls are more
spontaneous than 10-Ks or 10-Qs, and the information disclosed during conference calls is mostly
voluntary (Frankel, Mayew, and Sun 1999; Bowen, Davis, and Matsumoto 2002; Kimbrough
2005; Davis, Ge, and Matsumoto 2015). Management linguistic features such as vocal emotion
cues (Mayew and Venkatachalam 2012), excessive use of negative words (Druz, Wagner, and
Zeckhauser 2015), and tone dispersion (Allee and Deangelis 2015) have been examined in prior
studies and these linguistic features are demonstrated to have information content. Among
practitioners, earnings-conference calls are used by equity-research firms to search for cues of
deception (Javers 2010). In addition, Auditing Standard No. 12 issued by the Public Company
Accountability Oversight Board in 2010 mandates that auditors consider “observing or reading
transcripts of earnings calls” as part of the process for identifying and assessing risks of material
misstatements. However, to our knowledge, no prior research has examined conference calls (or
employed other textual-analysis techniques) in relation to big-bath accounting.
2.4 Hypothesis Development
The research question we examine is whether big baths taken at firms with truthful CEOs are
more credible, and whether investors can perceive different motivations behind big baths by
9
Another stream of linguistic analysis research involves supervised machine learning, such as Naïve Bayes method
and Support Vector Machines. Purda and Skillicorn (2015) use a decision-tree approach to establish a rank-ordered
list of words from the MD&A sections that are best able to distinguish between fraudulent and truthful reports, and
use support vector machines to predict the status of each report and assign it a probability of truth.
12

looking at CEOs’ linguistic cues. As CEOs possess private inside information of their firms,
financial-statement users are likely to assess the credibility of management disclosure regarding
the extent to which such disclosure represents management’s unbiased beliefs about the true nature
of the transactions and events within the firm (Hodge, Hopkins, and Pratt 2006).
Research in psychology indicates that the credibility or reputation of a source influences
reactions to a message (Petty and Cacioppo 1986; Pornpitakpan 2004; Cianci and Kaplan 2010).
Attribution theory illustrates that individuals have motivation to interpret and analyze events for a
better understanding of the environment (Ross and Nisbett 1991), and that information is gathered
and combined by an individual to form a causal judgment (Fiske and Taylor 1991).
In line with the above psychological theories, Williams (1996) and Mercer (2005) find that
the quality of managers’ disclosures influences how the managers are judged by analysts and
investors. Hodge, Hopkins, and Pratt (2006) show that the level of discretion in the reporting
environment and management’s reporting reputation affect the importance of incentive
consistency in explaining the credibility of management disclosures. Cianci and Kaplan (2010)
argue that even for nonprofessional investors who possess limited investment expertise, these
investors might still scrutinize the available information and assess the plausibility of
management’s disclosures.
We predict that, if CEOs are deceptive, big baths are more likely to reflect an accounting
choice driven by such incentives as securing bonus payments or meeting earnings targets. Such
big baths would have low credibility and should increase the firm-level information asymmetry.
In contrast, a big bath taken by a less deceptive (or more truthful)CEO is more likely to be
reflective of the CEO’s constructive response to an existing problem or the intention to “clear the
air”– realign a firm’s accounting numbers with their economic values. If this is the case, big baths
13

taken by less deceptive CEOs would have higher credibility than big baths taken by deceptive
managers and should decrease information asymmetry. Conversely, the more deceptive the CEO,
the less credible the big bath should be and subsequently, the higher the information asymmetry
should be in the post big-bath period. Stated formally (in alternative form):
H1: Compared with a big bath taken by a less deceptive CEO, a big bath taken by a deceptive
CEO will result in an increase in the information asymmetry in the post-bath period.
We focus on information asymmetry in our paper for two main reasons. First, information
asymmetry is a fundamental issue to the health of capital markets. Addressing the effects of
asymmetric information has always been on the top of the agenda for the SEC, FASB, and other
regulators, as well as oversight boards. As discussed in a major conference held by SEC, the
existence of asymmetric information can enable market participants to engage in deleterious
strategic behaviors that would be impossible in a world of complete information. Hidden
information or hidden actions may permit executives to collect unwarranted rents for themselves,
at the expense of shareholders.10 Thus, we study information asymmetry against the backdrop of
big baths and management deception, in the hope of shedding light on how to prevent and mitigate
the negative consequences resulting from asymmetric information in the capital markets. Second,
information asymmetry has also aroused substantial interest among academic researchers.
Reduction in information asymmetry, or increase in liquidity, can benefit capital markets from
various perspectives. For example, Amihud and Mendelson (1986) provide theoretical and
empirical evidence that higher liquidity can lower a firm’s cost of capital. Supporting that
10
Excerpts of the keynote address by Mark J. Flannery, the chief economist and director of Division of Economic
and Risk Analysis at SEC, in a 2015 conference on auditing and capital markets.
14

prediction, Lang, Lins, and Maffett (2012) find that increased liquidity is associated with lower
implied cost of capital and with higher valuation. They conclude that liquidity is a major channel
through which financial reporting transparency can affect firm valuation and cost of capital.
3. Data and Research Design
3.1 Big Baths
Identifying big-bath events is not straightforward. Generally speaking, the trade-off is between
having a large enough sample that utilizes an objective (or “less subjective”) approach versus the
researcher using her intuition to attempt to capture the spirit of big baths by coming up with a self-
constructed measure.11 For our primary analyses we consequently employ two different empirical
proxies for big baths.
For our first proxy, intuition and practical insights suggest that big baths should capture the
idea that firms “over-state certain charges.” We consider that firms that have especially large
negative discretionary accruals would meet the definition of “over-stating charges.” Specifically,
we employ the performance-adjusted version of Kothari, Leone, and Wasley’s discretionary
accruals model (2005). We require industry-year adjusted income before extraordinary items
minus special items to be in the bottom tercile and we require performance-matched discretionary
accruals to be in the bottom quintile of the distribution. Please see Appendix A for further details.
Consequently, this measure – “the Accruals Approach” - should capture especially egregious
charges that are likely to fall under the caption of big baths. However, this comes at the cost of a
11
Not surprisingly, researchers have employed a variety of empirical approaches. Definitions of big baths include
announced asset write-downs (Strong and Meyer 1987; Zucca and Campbell 1992; Francis et al. 1996) and non-
discretionary asset write-downs (Elliott and Shaw 1988). However, these definitions of big baths suffer from certain
limitations. For example, only focusing on announced asset write-downs ignores multiple small write-downs that do
not warrant disclosure individually, but can be substantial when combined together; manual deletion of non-
discretionary items can introduce subjectivity (and thus measurement error) into the sample selection (Haggard et al.
2015).
15

relatively small sample size (resulting from our matched-sample design described below).
As our second approach, we follow a more objective approach that allows us to benchmark
our results on recent research (Haggard et al. 2015). Specifically, we identify big baths as fiscal
year-end observations in Compustat for which Special Items are negative and exceed one percent
of lagged total assets (i.e., they are especially large and non-recurring charges). This definition is
consistent with prior research including Elliott and Shaw (1988) and Haggard et al. (2015). We
name this approach as the “Special Items Approach” for future discussion. Appendix A provides a
detailed description of what goes into the “Special Items Approach.”
Considering the research design and the availability of earnings-conference calls, we choose
2005 to 2015 as the sample period to define big baths, generate treatment baths, and perform
propensity-score matching.12 From this sample, we remove financial firms (SIC codes between
6000 and 6199), firms with total assets less than five million dollars, and firms with the absolute
value of Special Items exceeding 100 percent of total assets. A big-bath indicator is generated that
equals one for each big bath identified satisfying our definition, zero otherwise. There are 4,557
such cases under the Accruals Approach, and 14,209 under the Special Items Approach. To cleanly
analyze how the information asymmetry changes from the pre to post period, baths are removed
that occur within one year of another bath, both before and after, to avoid complications arising
from multiple consecutive baths. A total of 3,180 unique treatment baths are identified under the
Accruals Approach and 6,174 are identified under the Special Items Approach.
To test how the market reacts, we use two event windows in the empirical analyses: three and
six months pre and post baths.13 To create a benchmark sample, event windows for the control
12
Quarterly conference-call transcripts are obtained from SeekingAlpha.com. We start the sample period in 2005 as
this is when the website provides sufficient conference-call transcripts.
13
In choosing the event window, the trade-off is between isolating the effect (i.e., using a short window) versus
investors having sufficient time to respond (i.e., using a longer window).
16

group are also generated in a similar way. The control group consists of firms that did not take any
bath during the period of 2005 to 2015. Through propensity-score matching (see below), each
treatment bath firm is matched to a control firm that does not take a big bath in the event window.
We comment on alternative matching approaches as well as non-matching based tests in Section
5.1.1.
3.2 Management Deception
Larcker and Zakolyukina (2012) propose a linguistic approach that can be used to detect
management deceptive discussions during earnings-conference calls. They find that deceptive
managers’ speech displays certain linguistic patterns. In this paper, we apply their methodology to
identify whether CEOs are classified as high or low in terms of deceptiveness. A summary of the
linguistic approach in Larcker and Zakolyukina (2012) is provided in Appendix B.
To apply Larcker and Zakolyukina’s (2012) approach to our setting, we first need to obtain
earnings conference-call transcripts for linguistic analysis. Specifically, we use Python to crawl
the website http://seekingalpha.com/earnings/earnings-call-transcripts. SeekingAlpha.com is a
crowd-sourced content service for financial markets. SeekingAlpha.com is also used by Allee and
Deangelis (2015) as their source for collecting such transcripts.
A typical conference-call transcript can be divided into three parts. Basic information is listed
in the first part: company name, ticker symbol, fiscal year, quarter, transcript date, and a list of
inside and outside participants with their names and occupations. We use Python to extract the
basic information.
The second part is management prepared remarks. Following the operator’s introduction, the
CEO will discuss the financial situation, financial performance, and other major events or changes
17

during the period covered by the conference call. The third part is the question and answer session
(Q&A), in which analysts ask questions for managers to answer. To some extent, the prepared
remarks section is more scripted than the Q&A section. However, Larcker and Zakolyukina (2012)
find similar results in their linguistic models regardless of pooling these two sections together or
separating them. As combining these two sections provides more instances of words to analyze,
we follow Larcker and Zakolyukina (2012) and do not make a distinction between the prepared
remarks and Q&A. As an additional test, we also run tests using linguistic patterns displayed in
prepared remarks and Q&A, separately, to identify management deception. The test results are
discussed in section 5.2.
All phrases that belong to a CEO in the conference calls are gathered for linguistic analysis.
The transcripts are structured in the chronological order of the speech: each speaker’s name is
followed by the content of her speech, and then the next speaker’s name and her speech. For each
transcript, we use Python to collect all the parts of the speech that belongs to the same speaker, so
that the content of each transcript is classified and allocated to different speakers. Next, we pick
the speech of CEOs from the classified transcripts and combine CEO speech with company name,
ticker symbol, fiscal year, quarter, and transcript date, constructing a dataset that can be linked
with Compustat through ticker symbols.
Larcker and Zakolyukina (2012) use the Linguistic Inquiry and Word Count psychosocial
dictionary (LIWC) for their textual analysis (see also Pennebaker, Chung, Ireland, Gonzales, and
Booth 2007). They construct several word categories based on their readings of the conference-
call transcripts. To be consistent with Larcker and Zakolyukina (2012), we use LIWC 2007
software to count the word frequency for each word category.14
14
We remove CEOs with fewer than 150 total word counts from the sample.
18

Larcker and Zakolyukina (2012) build linguistic models by regressing accounting
irregularities or restatements on word frequencies as well as total word count. They find that
deceptive CEOs use more “reference to general knowledge” and “extreme positive emotion”
words, and use fewer “anxiety” and “shareholder-value” words. We provide details of deceptive
words in Appendix C.
We calculate aggregated deception scores that indicate the tendency of managers to be
deceptive. First, for each category of deceptive words, we calculate the sample-median word
frequency and code a score of 2 (1) if a CEO is associated with an above (equal to or below)
sample-median word frequency. Word frequencies are multiplied by minus one for words that are
negatively associated with deception. Then we add up the scores of each word category to generate
a total deception score. The score is further standardized so that the range of the variable remains
between zero and one. If a firm has a deception score that is above (equal to or below) the sample
median of the deception score, we classify the firm into the high (low) deception group. The
indicator variable DECEPTION equals one for the CEOs who are classified to the high deception
group, zero otherwise.
3.3 Propensity-Score Matching
When analyzing changes in information asymmetry between the pre- and post-bath periods, it
is important to control for non-bath events. For example, big baths are generally associated with
negative income in the pre-bath period, and it is possible that changes in information asymmetry
are common among firms with poor operating performance. Failing to control for such factors
would introduce an omitted-variable bias into the regression model. It is also essential to control
for the potential firm characteristics that are associated with a firm’s big-bath taking choice. In
19

order to mitigate such endogeneity concerns and isolate the effect of big baths, we use a difference-
in-differences approach and in particular employ propensity-score matching to match treatment
firms and control firms. Propensity-score matching enhances the comparability between treatment
firms and control firms, as the propensity score summarizes across all relevant matching variables
and offers a diagnostic on the comparability of the treatment and comparison groups (Dehejia and
Wahba 1999).
Each treatment firm is matched to a control firm using propensity-score matching. To generate
propensity scores, we run logit regressions, for both the Accruals Approach and Special Items
Approach, with the big-bath indicator as the dependent variable and with firm size, book-to-market
ratio, net income scaled by total assets, revenue scaled by total assets, annual cumulative stock
returns, annual Amihud illiquidity, annual share turnover, leverage, sales growth, institutional
ownership, and analyst coverage as explanatory variables.15 Treatment baths with missing data
points on the matching variables are dropped after this stage. Treatment baths with no conference-
call data available are also dropped, resulting in 327 treatment baths under the Accruals Approach
and 1,137 treatment baths under the Special Items Approach for the next step in which each
treatment firm is matched to a control firm that has the same industry classification, the same
fiscal-year end, and the closest propensity score. The control group consists of firms that have not
taken any bath during the sample period. A total of 254 treatment firms are matched to 254 control
firms in the Accruals Approach; 442 treatment firms are matched to 442 control firms in the
Special Items Approach.
15
Among these matching variables, following Francis et al. (1996) and Haggard et al. (2015), size, book-to-market
ratio, net income scaled by total assets, revenue scaled by total assets, annual cumulative stock returns, annual Amihud
illiquidity, and annual share turnover are lag values. Leverage, sales growth, institutional ownership, and analyst
coverage are current-year observations.
20

We employ nearest-neighbor matching without replacement. A caliper is imposed in the
matching process, as involving caliper is generally a best practice to decrease the likelihood of
‘‘poor’’ matches and to improve covariate balance (Shipman, Swanquist, and Whited 2017). For
each successful match, the maximum allowable distance between propensity scores is restricted
within the range of a caliper distance. A better covariate balance can be achieved following this
approach, but it may come at the cost of a reduced sample size. We set a strict caliper distance of
0.00001 for the propensity-score matching procedure for both the Accruals Approach and Special
Items Approach, as in this way a matched treatment and control group that are comparable across
many important matching variables can be generated, without unduly reducing the sample size.
3.4 Difference-in-Differences Approach and Research Model
We use the following model to test our hypothesis:
IA = α + β1 TREAT + β2 POST + β3 DECEPTION + β4 TREAT × POST + β5 TREAT ×
DECEPTION + β6 POST × DECEPTION + β7 TREAT × POST × DECEPTION + δ Controls +
Year Fixed Effects + Industry Fixed Effects + ε
We use a difference-in-differences research design to examine whether CEOs being truthful
can increase the credibility of big baths and whether investors can perceive the different
motivations behind big baths taken by the two types of CEOs. In the model, TREAT equals one for
firms in the treatment bath group and zero for firms in the control group. POST is an indicator
variable that equals one (zero) for the months after (prior to) a bath. DECEPTION equals one for
CEOs that belong to the high-deception group. The main variable of interest is β7, which measures
21

the difference between the high- and low-deception groups in terms of the changes in information
asymmetry post bath. We use this triple-difference framework to study whether the effects of big
baths on information asymmetry differ for baths taken by more deceptive CEOs versus those taken
by less deceptive CEOs.
The dependent variable is information asymmetry (IA), as captured by Amihud’s (2002)
illiquidity measure and bid-ask spreads, both of which are widely used in the literature (e.g.,
Balakrishnan, Billings, Kelly, and Ljungqvist 2014; Haggard et al. 2015). Goyenko, Holden, and
Trzcinka (2009) conclude that the Amihud measure is better at capturing price impact than other
liquidity measures. Amihud is defined as the monthly mean of the daily absolute returns divided
by dollar volume. In the regressions, we take the log of one plus Amihud. Spread is defined as the
monthly average of the daily bid-ask spreads. Data used to calculate the liquidity measures are
obtained from the CRSP daily profile.
Our differences-in-differences approach controls for time-invariant determinants of
information asymmetry. However, to reduce the possibility that our findings are confounded by
omitted variables, we include a number of control variables that are motivated by prior research.
The control variables include SIZE (log of total assets), BTM (book-to-market ratio), INCOME
(income scaled by total assets), LEVERAGE (debt to equity ratio), RETURN_A_LAG (lagged
annual return), INSTOWN (institutional ownership), ANALYST (analyst coverage),
EARNING_SHOCK (absolute value of earnings shock), ROA_CHANGE (change in ROA),
BEAT (indicator variable of beating analyst forecast), DELTA (CEO equity compensation delta),
VEGA (CEO equity compensation vega), DELTA_M and VEGA_M (indicator variables of
whether CEO delta or vega is missing), GENKNLREF_PRIOR, POSEMOEXTR_PRIOR,
SHVALUE_PRIOR, and ANX_PRIOR (CEO prior linguistic patterns). SIZE, MTB, INCOME,
22

LEVERAGE, INSTOWN, and ANALYST are firm characteristics shown to be associated with
information asymmetry in prior literature. Studies find that firm size is negatively associated with
percentage spread (Golsten and Harris 1988; Leuz 2003) and PIN (Brown, Hillegeist, and Lo 2004).
We therefore expect a negative coefficient on SIZE. INSTOWN is used to control for the presence
of potentially informed market participants and sophisticated investors (Brown et al. 2004). We
expect a negative coefficient on INSTOWN, given prior literature illustrates a negative association
between information asymmetry and the percentage of institutional ownership (Bhattacharya,
Desal, Venkataraman 2013). Analyst following can reflect the information collection process of
market participants and its interaction with firms’ financial reporting practice (Bhushan 1989).
Firms with higher analyst following are found to be associated with lower information asymmetry
(Bhattacharya et al. 2013), and with reduced profitability of insider trades (Frankel and Li 2004).
We expect the coefficient on ANALYST to be negative. BEAT is a measure indicating whether a
firm’s actual earning exceeds analysts’ consensus forecast by a small margin. Brown, Hillegeist,
and Lo (2009) find that “beat” firms experience a decrease in information asymmetry. Thus, we
expect a negative coefficient on BEAT.
We add several other measures in our regression model to control for their potential influential
effects, but do not provide a directional prediction on these coefficients due to the lack of consensus
findings in prior research. BTM, book to market ratio, is a proxy for growth opportunities or risk.
Information asymmetry and disclosure issues are pertinent for growth firms, generally
characterizing by a high level of intangible assets (Smith and Watts 1992; Leuz 2003). However,
analyst coverage is also higher for such firms and analysts expend more effort in analyzing these
firms. In terms of LEVARAGE, pecking order theory of capital structure implies that leverage is
negatively associated with the amount of firm-investor information asymmetry. However, the
23

incentive for private information acquisition is demonstrated to be increasing with leverage (Boot
and Thakor 1993). EARNING_SHOCK, the absolute value of earnings shock, controls for the
information content of the earnings announcement (Bushee, Core, Guay and Hamm 2010).
INCOME, RETURN_A_LAG, and ROA_CHANGE are added to control for firms’ financial
performance, as prior studies show that the level of firm investor information asymmetry is likely
to increase with performance variability (Brown and Hillegeist 2007). We also control for CEO
risk taking incentives, measured by DELTA, the sensitivity of a manager's wealth to the firm's
stock price, and VEGA, the sensitivity of a manager's wealth to the firm's stock return volatility.
CEO DELTA and VEGA are found to be significantly associated with firms’ financial policies such
as corporate leverage and cash holding policies. GENKNLREF_PRIOR, POSEMOEXTR_PRIOR,
SHVALUE_PRIOR, and ANX_PRIOR represent prior conference calls’ average word frequencies
in the use of “reference to general knowledge,” “extreme positive emotions,” shareholder value,”
and “anxiety.” These variables are included in the regressions to control for the effect of the CEO’s
linguistic habit.16 All continuous variables are winsorized at the 1st and 99th percentiles. Finally,
we include both year and industry fixed effects (based on two-digit SIC codes) and cluster the
standard errors by firm.17
We use the Compustat variable Datadate, which indicates the fiscal-year end for the financial
statements, to identify the cutoff for the pre and post periods. The advantage of using Datadate is
that it is consistently available across firm years and is closely associated with how we define big
baths. Each Datadate is matched with the most recent earnings-conference call held prior to this
Datadate. We further require that the time lag between a conference call and Datadate to be within
16
Results are robust if we remove these four linguistic characters as control variables. More generally, our inferences
are not affected by the inclusion or exclusion of particular control variables.
17
Alternatively, we cluster standard errors by industry instead of by firm. No conclusions are affected (untabulated).
24

one year. By doing so, we intend to capture investors’ perceptions of CEO deception most related
to the fiscal-year end big baths. The choice is also consistent with the recency effect in psychology
– people tend to recall things that arrive more recently (or appear at the end of a list).18
4. Empirical Results
4.1 Sample Selection and Effectiveness of Propensity-Score Matching
Table 1 presents the sample-selection process for the final treatment firms. Table 2 exhibits
the logistic regression of the big-bath indicator on important predictors of such big baths.19 Of all
the matching variables used in the propensity-score matching under the Accruals Approach in
defining big baths, BTM_LAG (lag of book to market ratio), TURNOVER_A_LAG (lag of annual
share turnover), LEVERAGE, and ANALYST are positively associated with big baths and the
coefficients are statistically significant (0.136, 0.0672, 0.177, 0.138, respectively). SIZE_LAG,
REVENUE_LAG, INCOME_LAG, RETURN_A_LAG, AMIHUD_A_LAG, SALEGROWTH,
and INSTOWN are negatively associated with big baths and the coefficients are statistically
significant (-0.361, -0.0775, -2.473, -0.0851, -0.0926, -0.148, and -0.424, respectively). For the
Special Items Approach in defining big baths, the logit regression shows that SIZE_LAG,
BTM_LAG, REVENUE_LAG, TURNOVER_A_LAG, LEVERAGE, and INSTOWN
(INCOME_LAG, RETURN_A_LAG, AMIHUD_A_LAG, SALEGROWTH, and ANALYST)
are positively (negatively) related to big baths.
18
Each big-bath event is associated with CEO linguistic characteristics identified in the most recent earnings
conference call prior to the fiscal year-end. To illustrate the timeline of the overall research design, if a firm has a year-
end of December 31, 2017 (Datadate), the associated conference call date could be October, 2017, which is the
conference call held closest to the year-end date. We then calculate the information asymmetry and trading volume
measures for each month from December 2016 to November 2018.
19
The area under the curve is 82.36% (70.76%) for the Accruals Method (Special Items Method).
25

Table 3 reports the summary statistics of the main variables of the treatment and control firms.
Table 4 presents the results of the differences in these variables across the treatment and control
groups. After propensity-score matching, there is considerable similarity between bath and non-
bath firms. Panel A illustrates the results of key variable difference after propensity-score matching
under the Accruals Approach. After propensity-score matching, only INCOME_LAG remains
significantly different between treatment and control firms, according to t-test results.
REVENUE_LAG, INCOME_LAG, and SALEGROWTH remain significant between treatment
and control firms, based on Kolmogorov-Smirnov (K-S) test. Panel B demonstrates the results of
variable difference after propensity-score matching under the Special Items Approach.
REVENUE_LAG, INCOME_LAG, RETURN_A_LAG, and Amihud_A_LAG remain different
in the t-test; BTM_LAG, REVENUE_LAG, and LEVERAGE are different in the K-S test. Table
5 illustrates the correlations among key variables for the main regressions under the Accruals
Approach.
4.2 Main Tests
Table 6 presents the results of different specifications of regressions testing how information
asymmetry changes after big baths taken by deceptive CEOs, compared with big baths taken by
less deceptive CEOs. AMIHUD and SPREAD are the dependent variables. Panel A represents
results from the Accruals Approach and Panel B represents results from the Special Items
Approach. The main variable of interest is the coefficient on the interaction term
TREAT×POST×DECEPTION. We report results with 3 month and 6 month as event-window
lengths.
26

In Panel A, the dependent variable is AMIHUD in columns 1 and 2, and SPREAD in columns
3 and 4. The estimated coefficients on TREAT×POST×DECEPTION are all positive and
statistically significant (at the 5% level or better using two-sided tests). Specifically, the
coefficients are 0.0957, 0.105, 0.196, and 0.191. Consistent with our hypothesis, Panel A shows
that big baths taken by deceptive managers lead to significant increase in information asymmetry
after big baths, compared with big baths taken by less deceptive managers.20
For control variables, the signs on the variables are mostly consistent with prior literature.
SIZE, RETURN_A_LAG, and INSTOWN are negatively associated with information asymmetry
and statistically significant. BTM, LEVERAGE, and ROA_CHANGE are positively associated
with information asymmetry.
Using the larger sample but likely coarser measure of big baths in Panel B, again the estimated
coefficients on TREAT×POST×DECEPTION are all positive and statistically significant. Thus,
our inferences are consistent using an alternative approach to measuring big baths that has been
employed in prior research.
20
The primary coefficients of interest, in the triple-interaction specification, are the coefficients on TREAT × POST
and especially TREAT × POST × DECEPTION. The coefficient on TREAT × POST captures the net effect of big
baths for less deceptive CEOs; the coefficient on TREAT × POST × DECEPTION measures the differential effect of
big baths for deceptive CEOs, compared with less deceptive CEOs. From Panel A, we observe that coefficients on
TREAT × POST are negative through column 1 to column 4 (statistically significant for column 2, 3, and 4), indicating
that information asymmetry decreases, or information environment improves, post big baths for less deceptive CEO
group. The coefficients on TREAT × POST × DECEPTION are positive and statistically significant in all columns,
indicating that information asymmetry increases, or information environment worsen off, if the baths are taken by
deceptive CEOs. In terms of numerical magnitude, using regression result in column 1 to illustrate, the net effect of
big baths on information asymmetry for less deceptive CEO is -0.0311, while the net effect of big baths on information
asymmetry for deceptive CEO is 0.0646 (-0.0311+0.0957). We find consistent results in Panel B, in which big baths
are identified using special items. As an alternative approach to avoid the three-way interaction, we estimate our model
including TREAT, POST, and TREAT×POST in separate deceptive and less-deceptive subsample regressions. In
untabulated results, the coefficients on TREAT×POST are all significantly positive in regressions for deceptive CEOs,
whether using three month or six month as event window, or with Amihud or Spread as information asymmetry
measure. In contrast, the coefficients on TREAT×POST are negative in regressions for less deceptive CEOs. The
difference in coefficients on TREAT×POST between deceptive and less deceptive subgroups is positive and
statistically significant, according to a z-test (Paternoster, Brame, Mazerolle, and Piquero 1998). Overall, this approach
provides consistent evidence that information asymmetry increases after big baths taken by deceptive CEOs.
27

Taken together, Table 6 provides support for our hypothesis that big baths taken by more
deceptive CEOs are associated with larger increase in information asymmetry following the baths,
compared with big baths taken by less deceptive CEOs.
5. Additional Analyses and Robustness Tests
In this section, we first provide robustness tests related our primary analyses. In particular, we
examine the effects of our matching procedures as well as other research-design choices. Next, we
attempt to investigate the “mechanisms” through which investors learn about the big baths. Finally,
we explore the effects of investors’ prior experience with CEOs’ deceptive (or truthful) speech.
5.1 Robustness
We conduct several tests to assess the sensitivity of our inferences to research-design choices.
For brevity, we do not tabulate the results of all these analyses.
5.1.1 Sensitivity Analyses Related to Matching
As described in Section 4.1, only one variable is statistically different (based on a t-test)
between the treatment and control samples after our propensity-score matching approach when
using the Accruals Approach. For the Special Items - based big-bath definition, four variables
remain significantly different. This reflects a trade-off between reducing differences between the
two samples and having a sufficiently large sample (“generalizability”). Please note that we
include the matching dimensions as control variables in all tests. To assess the robustness of our
findings to our propensity-score matching specifications, we conduct the following tests. First, for
the variables with significant differences after matching, we include non-linear terms in the
28

regression tests. Specifically, we include either the square root or the squared of these variables as
additional regressors and find that conclusions are unchanged.
Next, we implement a different matching approach: entropy matching. This approach allows
the researcher to match not only on means but also on higher-order dimensions of the covariates.
We find that our results continue to hold. Finally, although we believe the propensity-score
matching approach provides strong control for potential omitted variables, to generalize our
findings we run the regression without a control sample and thus test for a pure pre-post effect for
the treated firms. This sensitivity test also ensures that our findings are not induced by matching
(i.e., are not driven solely by the control-sample firms). Again, inferences remain. In sum, our
findings are not sensitive to our PSM choices.
5.1.2 Alternative Outcome Variables
To broaden the scope of our analyses, we also consider trading volume, another commonly
employed empirical outcome variables in this line of research. As the literature has used different
measures of trading volume, we consider both Turnover and Dollar Volume.21 Table 7 provides
the empirical results and shows, consistent with our predictions and with the primary tests,
21
Turnover, calculated as the total monthly trading volume divided by shares outstanding, is a widely used information
asymmetry measure in research (e.g. Leuz 2003; Chae 2005; Mohd 2005; Haggard et al., 2015). Dollar volume,
calculated as monthly mean of log daily dollar trading volume, is used to proxy for the benefit of information collection
(Bhushan 1989), and also to measure (inversely) the liquidity-related trading costs (Utama and Cready 1994).
29

negative and statistically significant estimated coefficients on our test variables. Thus, our findings
are not restricted to information asymmetry as measured by Amihud illiquidity and bid-ask spreads.
5.1.3 Other Robustness Tests
First, we repeat the above analyses by generating quartiles for each deception-word category,
instead of using the median cutoff. We additionally use the mean rather than the median of
deception scores to divide into high- and low-deception groups. Results reveal that inferences are
not sensitive to a particular cutoff standard being used to generate the deception score.
Next, although we employ two different approaches to identifying big baths in the paper, given
the importance of this empirical proxy and the fact this is not an obvious choice (i.e., prior research
uses a plethora of approaches), as a sensitivity analysis we follow the approach in Bens and
Johnston (2009).22 We find that our conclusions are unaltered using this alternative proxy.
Further, recall that the regressions include year fixed effects. For example, these fixed effects
control for any particular effect associated with the financial crisis. As an alternative approach, we
replace the year fixed effects with a specific control for the financial-crisis period, and inferences
are unaffected.
Big baths are likely to happen when there is a CEO turnover. We control for this possibility
by including an indicator variable that equals one if there is a CEO turnover during the event
window, zero otherwise. CEO turnovers are identified from the Compustat ExecuComp database,
which provides time-series data for top executives in the S&P 1500 firms since 1992. Our results
remain consistent after controlling for CEO turnover.
22
Using the Bens and Johnston (2009) methodology, we define big baths as restructuring charges (item #376) and
asset impairments (items #368 and #380) that exceed one percent of beginning total assets.
30

Finally, although Larcker and Zakolyukina (2012) provide validity tests for the linguistic
approach we follow in this study, we also conduct additional validity tests. That is, we test whether
our measure of CEO deception is positively associated with benchmark beating. Using both
analysts’ forecasts and prior earnings as benchmarks, consistent with our prediction we find that
DECEPTION loads positively and is highly statistically significant (after controlling for known
determinants of benchmark beating).
We conclude that our inferences are robust to these research-design choices and that the
empirical evidence suggests that investors are able to process the information in conference calls
and the information environment is differentially affected depending on whether CEOs are deemed
to be deceptive or not.
5.2 How Do Investors Learn about Big Baths and Detect Truthfulness?
It is interesting and important to consider the “mechanisms” through which our primary results
ensue. We clearly acknowledge that by using U.S. data we are not able to completely answer this
question as data on individual investors and their characteristics are generally not available.
However, to potentially shed some light on these important issues we do the following.
First, we have carefully scrutinized the conference-call transcripts in our sample. Specifically,
we have read through them and paid attention to whether and how analysts and investors ask
questions related to big baths, and also to company executives’ responses to these queries. We
provide several examples of such exchanges in Appendix D. As the appendix makes clear, both
analysts and investors ask questions related to big-bath accounting, and company executives
respond to such questions. Thus, we believe that investors are cognizant of the possible effects of
31

big baths and consequently “pay attention” to company actions, including potentially by
scrutinizing CEO speech.
Next, we consider that the results should be stronger when the baths are more visible and
salient to investors. Accordingly, we partition the sample based on the medians of (1) institutional
ownership, (2) analyst coverage, (3) press coverage, and (4) the magnitude of the baths.23 The
results are reported in Table 8. We observe that the effects are much stronger (and in some cases
only present) in the subsamples with higher visibility or salience. It is in these cases that
management deception is likely to be visible to investors and to matter (especially the magnitude
of the baths) more to investors.24,25 As these tests are indirect we do not want to draw strong
conclusions from them; however the findings are consistent with the idea that salience matters in
investors paying attention to and detecting managerial deception.
Third, we explore which segment of the conference call, prepared remarks versus the question
and answer section (Q&A), provides linguistic patterns that are more indicative of management
deception and more relevant in driving the upsurge of information asymmetry. Prior literature
shows that both segments convey incremental information over the accompanying press release,
but that the Q&A section is relatively more informative (Matsumoto, Pronk, and Roelofsen 2011).
We calculate deception scores based on the linguistic characteristics from prepared remarks and
Q&A individually, and run regressions with AMIHUD as the dependent variable for prepared
marks deception and Q&A deception separately. As shown in Table 9, management deception
23
Data on press coverage are obtained from RavenPack and aggregated within a fiscal year for each firm.
24
It is possible that technological developments have made it easier to detect deception over time. That is, the
technology becomes more familiar and improved over time. In an untabulated test, we find that the effect is more
pronounced in the most recent time period, providing some indirect support for this idea.
25
In another (untabulated) test, we follow the approach in Haggard et al. (2015) and partition the baths based on how
“forced” they are. Specifically, we use the Barton and Simko (2002) methodology that relies on the opening Net
Operating Assets to identify “forces” and “voluntary” baths. We find that the information asymmetry is higher when
the baths are taken by deceptive managers and when the baths are most discretionary (which is consistent with the
findings of Haggard et al. 2015).
32

indicated by linguistic features in Q&A leads to a more significant decrease in information
asymmetry.
5.3 The Effect of Previous versus Event-Window Specific Management Deception
It is possible that the capital market can form an expectation on the truthfulness or
deceptiveness of a CEO based on prior earnings-conference calls. For each individual CEO, it
could be that the CEO is truthful to investors in general, but deceptive to investors in the event
(bath or non-bath) year; or it could be the contrary - the CEO is a deceptive person most of the
time, but truthful to investors in the event year. In order to identify whether and how the capital
market reacts differently under these two circumstances, we construct a Deception_prior score for
26
each CEO by using the available earnings conference-call transcripts. CEOs with
Deception_prior score that is above (equal to or below) the sample median of Deception_prior are
classified in the high (low) prior deception group, which is indicated by PRIOR DECEPTIVE
(PRIOR TRUTHFUL). We run the regressions for the high and low groups, respectively.
The regression results are presented in Table 10. The main variable of interest is the coefficient
on the interaction term TREAT×POST×DECEPTION. Column 1 represents the regression results
for the PRIOR TRUTHFUL group, with 3 months as the event window. The coefficient on the
interaction term is positive and significant at the 5% level, implying that a prior truthful CEO being
deceptive when taking a big bath leads to significant increase in information asymmetry post bath,
compared with a prior truthful CEO being truthful when taking a big bath. Column 2 contains the
26
Specifically, for each CEO, we calculate the average word frequency for the four deception-related word categories
– GenKnlRef, PosEmoExtr, Anxiety, and ShareValue – across all available prior quarterly earnings-conference calls.
By taking the average of word frequencies, we intend to capture the CEO deception on a general level prior to the
bath event, to be compared with the level of CEO deception at the event time. Deception_prior score is generated the
same way as DECEPTION, using the average prior word frequencies.
33

results for the PRIOR DECEPTIVE group, with 3 months as the event window. The coefficient on
the interaction term is not statistically significant. We observe similar results for regressions using
six months as the event window as shown in column 3 and 4.
Table 10 suggests that, if a CEO is deceptive in general, the capital market does not react
significantly different whether the CEO is being truthful or deceptive when taking a big bath.
However, information asymmetry increases significantly for a generally truthful CEO being
deceptive when taking a big bath. It is thus conceivable that the capital market can form an
expectation of whether a CEO is truthful or deceptive according to prior earnings-conference calls.
If a CEO is perceived to be deceptive in general, it is likely that investors will interpret his or her
disclosures with caution. In contrast, investors may less likely to rigorously scrutinize the behavior
of a generally truthful CEO, as Mercer (2004) points out that management’s reputation is a
relatively enduring trait.
6. Conclusion
While some studies find that big baths can improve the information environment, others find
that they degrade the information environment. To expand upon prior literature, this paper looks
at big baths in conjunction with management deception. In particular, we examine whether
management deception can decrease the credibility of big baths and alter investors’ perceptions.
We find that information asymmetry (proxied for using the Amihud illiquidity measure and bid-
ask spreads) increases significantly after big baths taken by deceptive CEOs as compared to those
taken by less deceptive CEOs. Thus, investors are able to discern which managers are deceptive
and react accordingly. We believe our findings add to the big-bath literature as well as the
accounting literature in general.
34

The measurement of management deception in this paper relies on the findings in Larcker and
Zakolyukina (2012). By applying textual-analysis methodology to identify managers who are more
likely to engage in such action in order to manage earnings, this paper also contributes to the
literature by introducing a managerial factor that may help explain the inconsistent findings in
prior studies regarding the impact of big baths on information asymmetry. The study further adds
to the literature by studying how management ethics, indicated by the linguistic patterns during
conference calls can affect the information environment. Future research can consider alternative
textual-analysis techniques such as naïve Bayes and Support Vector Machines to test the validity
of our findings and potentially improve the precision in capturing management deception. Finally,
we encourage future research to explore the mechanisms behind the findings in this study in more
detail, possibly using surveys or interview-based approaches, and possibly also using other
institutional settings (e.g., in countries where data on individual investors and their characteristics
are available).
35

REFERENCES
Albert, L. S., J. R. Scott, and B. Turan. 2015. Turning inward or focusing out? Navigating theories
of interpersonal and ethical cognitions to understand ethical decision-making. Journal of
Business Ethics, 130 (2), 467–484.
Allee, K., and M. D. Deangelis. 2015. The structure of voluntary disclosure narratives: evidence
from tone dispersion. Journal of Accounting Research, 53 (2), 241-274.
Amernic, J., R. Craig, and D. Tourish. 2010. Measuring and assessing tone at the top using annual
report CEO letters. Edinburgh: The Institute of Chartered Accountants of Scotland.
Amihud, Y. 2002. Illiquidity and stock returns: cross-section and time-series effects. Journal of
Financial Markets, 5 (1), 31–56.
Amihud, Y., and H. Mendelson. 1986. Asset pricing and the bid-ask spread. Journal of Financial
Economics, 17 (2), 223–49.
Armstrong, C. S., A. D. Jagolinzer, and D. F. Larcker. 2010. Chief executive officer equity
incentives and accounting irregularities. Journal of Accounting Research, 48 (2), 225–271.
Balakrishnan, K., M. B. Billings, B. Kelly, and A. Ljungqvist. 2014. Shaping liquidity: on the
causal effects of voluntary disclosure. The Journal of Finance. 69 (5), 2237–2278.
Baucus, M. 1994. Pressure, opportunity, and predisposition: a multivariate model of corporate
illegality. Journal of Management, 20 (4), 699–721.
Baumhart, R. S. 1961. How ethical are businessmen? Harvard Business Review 39, 6–31.
Bens, D. A., and R. Johnston. 2009. Accounting discretion: use or abuse? An analysis of
restructuring charges surrounding regulator action. Contemporary Accounting Research,
26 (3), 673–699.
Bhattacharya, N., H. Desai, and K. Venkataraman. 2013. Does earnings quality affect information
asymmetry? Evidence from trading cost. Contemporary Accounting Research, 30 (2), 482–
516.
Bhushan, R. 1989. Collection of information about publicly traded firms: theory and evidence.
Journal of Accounting and Economics, 11 (2-3), 183-206.
Boot, A. W. A. and A. V. Thakor. 1993. Security design. The Journal of Finance, 48 (4), 1349–
1378.
Bowen, R., A. Davis, and D. Matsumoto. 2002. Do conference calls affect analysts’ forecasts? The
Accounting Review, 77 (2), 285–316.
Brown, S. and S. A. Hillegeist. 2007. How disclosure quality affects the level of information
asymmetry. Review of Accounting Studies, 12 (2–3), 443–477.
Brown, S., S. A. Hillegeist, and K. Lo. 2004. Conference calls and information asymmetry. Journal
of Accounting and Economics, 37 (3), 343-366.
Brown, S., S. A. Hillegeist, and K. Lo. 2009. The effect of earnings surprises on information
asymmetry. Journal of Accounting and Economics, 47 (3), 208-225.
Bushee, B. J., J. E. Core, W. Guay, and S. J. W. Hamm. 2010. The role of the business press as an
information intermediary. Journal of Accounting Research, 48 (1), 1-19.
Cardinaels, E. 2016. Earnings benchmarks, information systems, and their impact on the degree of
honesty in managerial reporting. Accounting, Organizations, and Society 52: 50-62.
Carson, T. L. 2003. Self-interest and business ethics: some lessons of the recent corporate scandals.
Journal of Business Ethics 43 (4), 389–394.
Chae, J. 2005. Trading volume, information asymmetry, and timing information. The Journal of
Finance, 60 (1), 413-442.
Chen, J. D., D. Cumming, W. X. Hou, and E. Lee. 2016. CEO accountability for corporate fraud:
36

evidence from the Split Share Structure Reform in China. Journal of Business Ethics, 138
(15), 787–806.
Christensen, T. E., G. H. Paik, and E. K. Stice. 2008. Creating a bigger bath using the deferred tax
valuation allowance. Journal of Business Finance & Accounting, 35 (5) & (6), 601–625.
Chung, C. K., and J. W. Pennebaker. 2012. Linguistic Inquiry and Word Count (LIWC):
pronounced “Luke,”... and other useful facts. Applied Natural Language Processing:
Identification, Investigation, and Resolution. Chapter 12. IGI Global.
Ciancia, A. M., and S. E. Kaplan. 2010. The effect of CEO reputation and explanations for poor
performance on investors’ judgments about the company’s future performance and
management. Accounting, Organizations and Society, 35 (4), 478–495.
Craig, R., T. Mortensen, and S. Iyer. 2013. Exploring top management language for signals of
possible deception: the words of Satyam’s Chair Ramalinga Raju. Journal of Business
Ethics, 113 (2), 333–347.
Davis, A. K., W. Ge, and D. Matsumoto. 2014. The effect of manager-specific optimism on the
tone of earnings conference calls. Review of Accounting Studies, 20 (2), 639–673.
Dehejia, R. and S. Wahba. 1999. Causal effects in nonexperimental studies: reevaluating the
evaluation of training programs. Journal of the American Statistical Association, 94 (448),
1053-1062.
Druz, M., A. F. Wagner, R. J. Zeckhauser. 2015. Tips and tells from managers: how analysts and
the market read between the lines of conference calls. NBER Working Paper
Efendi, J., A. Srivastava, and E. Swanson. 2007. Why do corporate managers misstate financial
statements? The role of in-the-money options and other incentives. Journal of Financial
Economics, 85 (3), 667–708.
Ekin, M. G. S., and S. H. Tezölmez. 1999. Business ethics in Turkey: an empirical investigation
with special emphasis on gender. Journal of Business Ethics, 18 (1), 17–34.
Elliott, J. A., and W. H. Shaw. 1988. Write-offs as accounting procedures to manage perceptions.
Journal of Accounting Research, 26, 91–119.
Evans, J. H., R. L. Hannan, R. Krishnan, and D. V. Moser. 2001. Honesty in managerial reporting.
The Accounting Review, 76 (4), 537–559.
Francis, J., J. D. Hanna, and L. Vincent. 1996. Causes and effects of discretionary asset write-offs.
Journal of Accounting Research, 34, 117–134.
Frankel, R., and X. Li. 2004. Characteristics of a firm's information environment and the
information asymmetry between insiders and outsiders. Journal of Accounting Research,
37 (2), 229-259.
Frankel, R., W. Mayew, and Y. Sun. 2009. Do pennies matter? Investor relations consequences of
small negative earnings surprises. Review of Accounting Studies, 15 (1), 220–242.
Geiger, M. A., and D. S. North. 2006. Does hiring a new CFO change things? An investigation of
changes in discretionary accruals. The Accounting Review, 81 (4), 781-809.
Glosten, L., and L, Harris. 1988. Estimating the components of the bid/ask spread. Journal of
Financial Economics, 21 (1), 123-142.
Goyenko, R. Y., C. W. Holden, and C. A. Trzcinka. 2009. Do liquidity measures measure liquidity?
Journal of Financial Economics, 92 (2), 153–181.
Haggard, K. S., J. S. Howe, and A. A. Lynch. 2015. Do baths muddy the waters or clear the air?
Journal of Accounting and Economics, 59 (1), 105–117.
Hambrick, D. C., and P. A. Mason. 1984. Upper echelons: the organization as a reflection of its
top managers. Academy of Management Review, 9 (2), 193–206.
37

Healy, P. M. 1985. The effect of bonus schemes on accounting decisions. Journal of Accounting
and Economics 7 (1-3), 85–107.
Hobson, J. K., W. J. Mayew, M. Peecher, M. Venkatachalam. 2017. Improving experienced
auditors’ detection of deception in CEO narratives. Forthcoming, Journal of Accounting
Research.
Hodge, F., P. E. Hopkins, and J. Pratt. 2006. Management reporting incentives and classification
credibility: The effects of reporting discretion and reputation. Accounting, Organizations
and Society, 31 (7), 623–634.
Hou, Y. 2015. The role of diversification in the pricing of accruals quality. Review of Accounting
Studies, 20 (3), 1059–1092.
Javers, E. 2010. Broker, Trader, Lawyer, Spy: the secret world of corporate espionage. New York:
Harper Collins Publishers.
Khanna, V., E. H. Kim, and Y. Lu. 2015. CEO connectedness and corporate fraud. Journal of
Finance 70 (3), 1203-1252.
Kimbrough, M. 2005. The effect of conference calls on analyst and market under-reaction to
earnings announcements. The Accounting Review, 80 (1), 189–219.
Kirschenheiter, M., and H. D. Melumad. 2002. Can big bath and earnings smoothing co-exist as
equilibrium financial reporting strategies? Journal of Accounting Research, 40 (3), 761–
796.
Kohlberg, L. 1981. The philosophy of moral development. San Francisco, CA: Harper & Row.
Kothari, S. P., S. Shu, and P. D. Wysocki. 2009. Do managers withhold bad news? Journal of
Accounting Research, 47 (1), 241–276.
Lang, M., K. V. Lins, and M. Maffett. 2012. Transparency, liquidity, and valuation: international
evidence on when transparency matters most. Journal of Accounting Research, 50 (3), 729–
774.
Larcker, D. F., and A. A. Zakolyukina. 2012. Detecting deceptive discussions in conference Calls.
Journal of Accounting Research, 50 (2), 495-540.
Leuz, C. 2003. IAS versus US GAAP: information asymmetry–based evidence from Germany's
new market. Journal of Accounting Research, 41 (3), 445–472.
Levitt, A. 1998.The numbers game. The CPA Journal, 68 (12), 14-19.
Liu, X. K., A. M. Wright, and Y. J. Wu. 2015. Managers’ unethical fraudulent financial reporting:
the effect of control strength and control framing. Journal of Business Ethics, 129 (2), 295–
310.
Loughran, T., and B. McDonald. 2011. When is a Liability not a Liability? Textual analysis,
Dictionaries, and 10Ks. Journal of Finance, 66 (1), 35-65.
Loughran, T., and B. McDonald. 2013. IPO first-day returns, offer price revisions, volatility, and
form S-1 language. Journal of Financial Economics, 109 (2), 307–326.
Maak, T., and N. M. Pless. 2006. Responsible leadership in a stakeholder society – a relational
perspective. Journal of Business Ethics, 66 (1), 99–115.
Mackey, A. 2008. The effect of CEOs on firm performance. Strategic Management Journal, 29
(12), 1357–1367.
Matsumoto, D., M. Pronk, and E. Roelofsen. 2011. What makes conference calls useful? The
information content of managers' presentations. The Accounting Review, 86 (4), 1383-
1414.
Mayew, M. J., and M. Venkatachalam. 2012. The power of voice: managerial affective states and
future firm performance. The Journal of Finance, 67 (1), 1–43.
38

Mendenhall, R. R., and W. D. Nichols. 1988. Bad news and differential market reactions to
announcements of earlier-quarters versus fourth-quarter earnings. Journal of Accounting
Research, 26, 63-86.
Mercer, M. 2004. How do investors assess the credibility of management disclosures? Accounting
Horizons, 18 (3), 185–196.
Mercer, M. 2005. The fleeting effects of disclosure forthcomingness on management’s reporting
credibility. The Accounting Review, 80 (2), 723–744.
Mihajlov, T. P., and W. F. Miller. 2012. Fraud at Buca di Peppo: a villa in Tuscany? Strategic
Finance, 94 (1), 40–46.
Mohd, E. 2005. Accounting for software development costs and information asymmetry. The
Accounting Review, 80 (4), 1211-1231.
Moore, M. L. 1973. Management changes and discretionary accounting decisions. Journal of
Accounting Research, 11 (1), 100–107.
Murphy, K. J., and J. L. Zimmerman. 1993. Financial performance surrounding CEO turnover.
Journal of Accounting and Economics, 16 (1-3), 273-315.
Newman, M. L., J. W. Pennebaker, D. S. Berry, and J. M. Richards. 2003. Lying words: predicting
deception from linguistic styles. Personality and Social Psychology Bulletin, 29 (5), 665-
675.
Patelli, L., and M. Pedrini. 2015. Is tone at the top associated with financial reporting
aggressiveness? Journal of Business Ethics, 126 (1), 3–19.
Paternoster, R., R. Brame, P. Mazerolle, and A. Piquero. 1998. Using the correct statistics test for
the equality of regression coefficients. Criminology, 36 (4), 859–866.
Pennebaker, J. W., C. K. Chung, M. Ireland, A. Gonzales, and R. J. Booth. 2007. The development
and psychometric properties of LIWC 2007. Austin, TX.
Pennebaker, J. W., R. L. Boyd, K. Jordan, and K. Blackburn. 2015. The development and
psychometric properties of LIWC 2015. Austin, TX: University of Texas at Austin.
Petty, R. E., and J. T. Cacioppo. 1986. The elaboration likelihood model of persuasion. Advances
in Experimental Social Psychology, 19, 123–205.
Pornpitakpan, C. 2004. The persuasiveness of source credibility: a critical review of five decades’
evidence. Journal of Applied Social Psychology, 34 (2), 243–281.
Purda, L., and D. Skillicorn. 2015. Accounting variables, deception, and a bag of words: assessing
the tools of fraud detection. Contemporary Accounting Research, 32 (3), 1193–1223.
Reynolds, S. J. 2006. A neurocognitive model of the decision making process: implications for
study and practice. Journal of Applied Psychology, 91 (4), 737–748.
Riedl, E. J. 2004. An examination of long-lived asset impairment. The Accounting Review, 79 (3),
823-852.
Riske, S. T., and S. E. Taylor. 1991. Social cognition. Second edition. New York: McGraw – Hill.
Shipman, J. E., Q. T. Swanquist, and R. L. Whited. 2017. Propensity score matching in accounting
research. The Accounting Review, 92 (1), 213–244.
Smith, C. W. and R. L. Watts. 1992. The investment opportunity set and corporate financing,
dividend, and compensation policies. Journal of Financial Economics, 32 (3), 263-292.
Solomon, C. R. 1992. Ethics and excellence. Cooperation and integrity in business. New York, NY:
Oxford University Press.
Strong, J. S., and J. R. Meyer. 1987. Asset writedowns: managerial incentives and security returns.
Journal of Finance, 42 (3), 643–661.
Tausczik, Y. R., and J. W. Pennebaker. 2010. The psychological meaning of words: LIWC and
39

computerized text analysis methods. Journal of Language and Social Psychology, 29 (1),
24–54.
Tourish, D., and N. Vatcha. 2005. Charismatic leadership and corporate cultism at Enron: the
elimination of dissent, the promotion of conformity and organizational collapse.
Leadership, 1 (4), 455–480.
Utama, S. and W. M. Cready. 1997. Institutional ownership, differential predisclosure precision
and trading volume at announcement dates. Journal of Accounting and Economics, 24 (2),
129–150.
Vitell, S. J., J. R. Lumpkin, and M. Y. A. Rawwas. 1991. Consumer ethics: an investigation of the
ethical beliefs of elderly consumers. Journal of Business Ethics, 10 (5), 365–375.
Vrij, A. Detecting lies and deceit: pitfalls and opportunities. 2008. Second edition. Chichester, UK:
John Wiley and Sons.
Weber, J. 1990. Managers’ moral reasoning: assessing their response to three moral dilemmas.
Human Relations, 43 (7), 687–702.
Williams, P. A. 1996. The Relation between a prior earnings forecast by management and analyst
response to a current management forecast, 71 (1), 103–115
Zona, F., M. Minoja, and V. Coda. 2013. Antecedents of corporate scandals: CEOs’ personal traits,
stakeholders’ cohesion, managerial fraud, and imbalanced corporate strategy. Journal of
Business Ethics, 113 (2), 265–283.
40

Appendix A: How We Compute Big Baths
(1) Accruals Approach
We use two standards in recognizing whether a firm is taking a big bath: (1) if the firm has
extreme negative discretionary accruals, estimated from the approach proposed in Kothari, Leone,
and Wasley (2005); (2) the firm’s incentive of taking a big bath is discernible. Specifically, this
standard is applied to distinguish whether an earnings bath is happening when the current-year
financial performance is relatively poor compared to its industry peers, i.e. big bath incentive, or
when the company achieves superior financial performance and thus inclines to smooth earnings
into the future, i.e. earnings smooth incentives. Bens and Johnston (2009) discuss the two different
managerial incentives behind large accounting write-offs in their study of restructuring charges
and earnings management.
Kothari et al. (2005) suggest that the ROA (return on assets) matched discretionary accrual
derived from Jones model is a viable measure for earnings management, and this performance-
matched approach performs better than incorporating a performance variable in the discretionary
accruals regression. Applying this approach, we first estimate discretionary accruals by running
the Jones model (illustrated below) cross-sectionally each year using all firm-year observations in
the same two-digit SIC code. We require a minimum observations of ten for each two-digit SIC
industry and year combinations.
TAit = β0 + β1 (1/ASSETSit-1) + β2 ∆SALESit + β3 PPEit + Ɛit
TAit is total accruals, defined as the difference between income before extraordinary items and
operating cash flows, scaled by lag of total assets. ASSETSit-1 represents lagged total assets.
∆SALESit is change in sales scaled by lagged total assets. PPEit is the net property, plant, and
equipment scaled by lagged total assets. All data are obtained from COMPUSTAT.
41

After obtaining the Jones-model discretionary accrual, we match each firm-year observation
with another from the same industry (two-digit SIC code) and year with the closest current year
ROA, which is calculated as net income divided by total assets. Kothari et al. (2005) demonstrate
that matching based on the current year ROA is superior to matching on the prior-year ROA. We
calculate our performance-matched discretionary accrual, ACCRUAL, as a firm’s Jones-model
discretionary accrual minus the matched firm’s Jones-model discretionary accrual.
The first standard in defining big baths is whether a firm-year is associated with extreme
negative accruals. In achieving this objective, we rank performance matched discretionary accruals
into quintile and identify observations satisfying standard one if these observations fall under the
bottom quintile rank of discretionary accruals. The second standard in defining big baths is whether
a firm is experiencing a financial downturn in the current year, relative to other firms in the same
industry. We rank firms’ basic income, calculated as income before extraordinary items minus
special items, into tercile at industry-year level (industry represented by two-digit SIC code) and
standard two is met for observations belong to the bottom tercile of the basic income rank.
Indicator variable BATH is defined if a firm-year observation satisfies both of these two standards.
(2) Definition of Special Items – Special Items Approach
In Compustat, special itmes (SPI) are defined as unusual and/or nonrecurring items considered
special items by the company, including: (1) Adjustments applicable to prior years; (2) After-tax
adjustments to net income for the purchase portion of net income of partly pooled companies; (3)
Any significant nonrecurring items; (4) Bad debt expense/Provisions for doubtful
accounts/Allowance for losses if non-recurring; (5) Current year’s results of discontinued
operations and operations to be discontinued; (6) Flood, fire, and other natural disaster losses; (7)
42

Interest on tax settlements; (8) Items specifically called “Restructuring/Reorganization”, “Special,”
or “Non-recurring” regardless of the number of years they are reported; (9) Inventory writedowns
when separate line item or called non-recurring; (10) Nonrecurring profit or loss on the sale of
assets, investments, and securities; (11) Profit or loss on the repurchase of debentures; (12)
Recovery of allowances for losses if original allowance was a special item; (13) Relocation and
moving expense; (14) Severance pay when a separate line item; (15) Special allowance for
facilities under construction; (16) Transfers from reserves provided for in prior years; (17) Write-
downs or write-offs of receivables and intangibles; (18) Year 2000 expenses regardless of the
number of years they are reported.
43

Appendix B: Brief Summary of the Linguistic Approach in Larcker and Zakolyukina (2012)27
Larcker and Zakolyukina (2012) conduct linguistic analysis on quarterly earnings-conference
calls during the period of September 2003 to May 2007, and calculate word frequencies for all the
word categories that have been shown by previous psychological and linguistic research to be
related to deception. Larcker and Zakolyukina (2012) further regress financial deception indicators
on these word frequencies using logit regression and find that deceptive CEOs use significantly
more references to general knowledge words, more extreme positive emotion words, fewer
references to shareholder value words, and fewer anxiety words.
Each conference call is labeled as “truthful” or “deceptive” if it is associated with a
restatement: contains a disclosure of a material weakness, an auditor change, a late filing, or a
Form 8-K filing; relates to an irregularity as described in Hennes, Leone, and Miller (2008);
involves accounting issues that elicit a significant negative market reaction such as those described
in Palmrose, Richardson, and Scholz (2004) and Scholz (2008); involves a formal SEC
investigation that leads to an issuance of an Accounting and Auditing Enforcement Release
(AAER).
Word categories related to deception are selected based on Vrij (2008), which provides the
theoretical framework of explaining an individual’s nonverbal behavior during deception,
including emotions, cognitive effort, attempted control, and lack of embracement. To construct
deception related word categories, Larcker and Zakolyukina (2012) use LIWC, with some word
categories expanded by adding synonyms from a lexical database of English WordNet. Larcker
and Zakolyukina (2012) also establish word categories specific to conference call setting, namely,
references to general knowledge words, shareholder value words, and value creation words.
27
Please see Appendix C for definitions of deceptive words.
44

Appendix C: Definitions of Deceptive Words and Key Variables
Category Abbreviation Content

you know, you guys know, you folks know, you well know, you long know,
you would agree, everybody knows, everybody well knows,
everybody long knows, everybody would agree, everyone knows, everyone
well knows, everyone long knows, everyone would agree,
Reference others know, others well know, others long know, others would agree, they
to general GenKnlRef know, they well know, they long know, they would agree,
knowledge investors know, investors well know, investors long know, investors would
agree, shareholders know, shareholders well know,
shareholders long know, shareholders would agree, stockholders know,
stockholders well know, stockholders long know,
stockholders would agree
amaz*, A-one, astonish*, awe-inspiring, awesome, awful, bang-up, best,
bless*, brillian*, by all odds, careful*, challeng*, cherish*,
confidence, confident, confidently, convinc*, crack, cracking, dandy, deadly,
definite, definitely, delectabl*, delicious*, deligh*,
deucedly, devilishly, dynam*, eager*, emphatically, enormous, excel*,
excit*, exult, fab, fabulous*, fantastic*, first-rate, flawless*,
genuinely, glori*, gorgeous*, grand, grande*, gratef*, great, groovy, hero*,
Extreme huge, illustrious, immense, in spades, in truth,
positive PosEmoExtr incredibl*, insanely, inviolable, keen*, luck, lucked, lucki*, lucks, lucky,
emotions luscious, madly, magnific*, marvellous, marvelous, neat*,
nifty, outstanding, peachy, perfect*, phenomenal, potent, privileg*, rattling,
redoubtable, rejoice, scrumptious*, secur*, sincer*,
slap-up, smashing, solid, splend*, strong*, substantial, succeed*, success*,
super, superb, superior*, suprem*, swell, terrific*,
thankf*, tiptop, topnotch, treasur*, tremendous, triumph*, truly, truth*,
unassailable, unbelievable, unquestionably, vast,
wonderf*, wondrous, wow*, yay, yays, very good
shareholder value, shareholder welfare, shareholder well-being, value for our
shareholders, value for shareholders, stockholder
Shareholder
ShareValue value, stockholder welfare, stockholder well-being, value for our
value
stockholders, value for stockholder, investor value, investor
welfare, investor well-being, value for our investors, value for investors
LIWC category “anx”: worried, fearful, nervous, etc. Prior research:
Anxiety Anxiety Bachenko, Fitzpatrick, and Schonwetter (2008), Bond and Lee (2005),
Knapp, Hart, and Dennis (1974), Newman et al. (2003), Vrij (2008)
45

Appendix C: Definitions of Deceptive Words and Key Variables (continued)
Variables Definition
TREAT Indicator variable - equals one for each treatment bath
POST Indicator variable - equals one if in the post bath period
Indicator variable - equals one if a CEO is in the high deception
DECEPTION
group
Monthly mean of the daily absolute return divided by dollar volume:
1,000,000 × |ret|÷ (prc × vol). Log of one plus this ratio is used in the
AMIHUD
regressions. Daily CRSP data (variables ret, prc, and vol) are used to
calculate the ratio
Monthly mean of the daily bid-ask spread, which is calculated as 100
× (ask − bid)/[(ask + bid)/2]. Daily closing bid and ask data from
SPREAD
CRSP (variables ask and bid) is used, with crossed quotes (negative
spreads) excluded
TURNOVER Monthly total trading volume divided by shares outstanding
DOL_VOL Monthly average of log dollar trading volume
SIZE The natural logarithm of total assets
BTM Book value of equity divided by market value of equity
REVENUE Revenue scaled by total assets
INCOME Net income scaled by total assets
LEVERAGE Book value of debt divided by market value of equity
The natural logarithm of 1 plus the number of analysts issuing
ANALYST earnings forecasts for any horizon during the fiscal period. 0 for
any period in which no data are available on I/B/E/S
The percentage of shares held by institutional investors during the
INSTOWN fiscal period; 0 for any period in which no data are available in the
13-F filings
SALEGROWTH The percentage change in sales from the previous year
Expected dollar change in CEO wealth for a 1% change in stock
sensitivity (Delta) price (using entire portfolio of stocks and
DELTA options) computed as in Core and Guay (2002). Data of DELTA
are obtained from the website of Lalitha Naveen:
https://sites.temple.edu/lnaveen/data/. The data span 1992-2014
Expected dollar change in CEO wealth for a 1% change in stock
return volatility (using entire portfolio of options) computed as in
VEGA Guay (1999). Data of VEGA are obtained from the website of Lalitha
Naveen: https://sites.temple.edu/lnaveen/data/. The data span 1992-
2014
46

DELTA_M Indicator variable equals one if data on DELTA is missing
VEGA_M Indicator variable equals one if data on VEGA is missing
Absolute value of earnings shock, which is calculated as actual
EARNING_SHOCK earnings minus analysts’ consensus earnings forecast. Data are
obtained from I/B/E/S
Indicator variable equals one if actual earnings minus consensus
BEAT
analysts’ earnings forecast lies in the range of 0 to 0.01
RETURN_A Annual cumulative stock return.
Annual stock turnover (total trading volume divided by shares
TURNOVER_A
outstanding).
AMIHUD_A Log of one plus annual mean of daily Amihud ratio.
Average word frequency of GenKnlRef for the prior conference
GENKNLREF_PRIOR
calls
Average word frequency of PosEmoExtr for the prior conference
POSEMOEXTR_PRIOR
calls
Average word frequency of ShareValue for the prior conference
SHVALUE_PRIOR
calls
ANX_PRIOR Average word frequency of Anxiety for the prior conference calls
47

Appendix D: Examples of Big-Bath Related Discussions in Conference Calls
Analyst (or Investor) Questions Followed by Company Executive Responses
Company Name Analyst Name Question

Call Date Executive Name Answer
Okay, great. And then can you give us a better sense of this $10 million to
Jeff Hammond
$15 million restructuring? How does that flush out by quarter? Does that
- KeyBanc
kind of exacerbate the downside in the second quarter or does that flush
Capital Markets
out through the year may be just a little more?
Actuant Corp. Yes, it is definitely not a big bath, the way you could book restructuring
Dec 18, 2008 reserves in the old days. I mean this will be coming in over the balance of
the year. I would say, it's going to be more in the third and fourth quarter
Andy Lampereur
than what you would see in the second, but there will be some in the
second. The more facility oriented projects will be back half of the year,
and that's where the bigger dollars are.
Philip Ng Free cash flow has been somewhat depressed the last few years with a
- Jefferies & bigger reinvestment cycle and restructuring. So when I look out, going
Company, Inc., forward, you guys kind of alluded to restructuring. Will that be a big
Owens-Illinois Research Division headwind again on cash flow going forward?
Jul 26, 2012 ...But we will certainly try to avoid what we did in 2010, where we took a
Stephen P. significant haircut in the global free cash flow number and kind of took a
Bramlage big bath approach on the restructuring all at once. It will be much more of
a moderated, consistent approach.
Peter Costa The impairment of intangible assets this quarter, presumably that was
- FTN Equity mostly related to Illinois and Texas and not really related to the PBM, is
Capital Markets that accurate concerning if you’re talking about a gain for the PBM?
The vast majority of that was actually related to the PBM and the revenue.
WellPoint, Inc.
The way you look at the calculations is to base it on revenue, not
Oct 28, 2009
Wayne S. necessarily the operating earnings, the way the GAAP accounting rules
Deveydt work. So it’s the membership associated that was being driven over to the
PBM has actually gone away. That actually takes the majority of that
writeoff.
Yes, I have two questions; the first is very simple. Do you see any more
John Ibis
writedowns or writeoffs as in the example of this last quarter of $15
- Private Investor
million...
Sure, I think I’ll take them all as you asked questions. The amount that we
Anworth have written off relative to our investment of Belvedere actually exceeds
Joseph E.
Mortgage Asset our economic exposure to Belvedere by the $7 or $8 million that we
McAdams
Corp discussed before, so we do not anticipate any additional writedowns
Mar12, 2008 relative to Belvedere.
John Ibis But are there any other writedowns coming down the pike that you know
- Private Investor of right now?
Joseph E. No, and again, as we mentioned if anything there should be a reversal of
McAdams that $7 to $8 million writedown upon the dissolution of Belvedere.
Stan Meyers And final question, I guess -- you guys -- Rich, you mentioned some
- Piper Jaffray product writeoffs in the quarter. Is there an estimate what that number was,
Companies, incremental?
Research Division
DreamWorks
Yes, I don't know if we'd actually disclose that number. It is -- it
Animation SKG
traditionally flows to you cost of good lines having, probably, 1% to 2%
Jul 29, 2014
of margin impact in the quarter. We haven't actually sized that. And as you
Rich Sullivan
know, we assess our development slate every quarter. So we take these
types of write-offs on our occasion. This quarter just happened to be larger
than it was last quarter.
48

Got you. And then the last question. I was wondering, you still are burning
Stan Manoukian
inventory, writing down a lot of inventory. What is the reason for this?
- Independent
And when do you think it will stop? And does it have anything to do with
Credit Research
the chloric-acid plant? Were there inefficiencies in the process related to
LLC
Molycorp Inc. this, or what is it?
Nov 6, 2014 Well, first of all, most of the writedown does indeed relate to Mountain
Pass. And the biggest chunk of it, as you see in our release, our cost per
Michael F.
kilogram was $33-plus, but the market price is significantly less than that.
Doolan
So we have to write the inventory down to net realizable value, lower of
cost to market, which accounts for most of it.
Jeff Klinefelter For the next fiscal quarter then we're not anticipating any sort of an
- Piper Jaffray inventory writedown or a margin hit, given that sales have improved?
No. In fact if anything, none of us like to see this type of economic
Oxford
situation that we have right now, but this is nothing new to us. We have
Industries Inc.
had to manage through this type of situation before. Really, this is when
Oct 9, 2007 Doug Wood
you tighten down your inventories and really make sure you're not too
aggressive on your sales projections and don't get too far out there ahead
of yourself. So I don't expect anything like that.
If I recall Mike on your fourth quarter call, you didn’t anticipate an
increase to your bad debt preserves, and you guys took about $41 million
Charles Grom here in the first quarter. Looking at your guidance, it doesn’t suggest that
- J.P. Morgan you’re going to take anymore reserves over the balance of the year, so I’m
just wondering if you could flush it out where you’re getting comfortable
on that for us.
The overall bad debt expense was up $41 million. The actual increase in
the reserve for the quarter was about 23-24. Back in the fourth quarter,
Charles, our point of view was that we thought unemployment was
Nordstrom, Inc.
somewhere in the 9 plus range and that the look of that curve was not
May 14, 2009
going to be as long in terms of the expected length of the unemployment,
but I think more recent data over the last several months and then increase
Michael G.
in our delinquency rates, certainly it changed our thinking. What we’re
Koppel
currently basing our reserves on is an unemployment factor in the 10% to
10.5% range which is a big driver of our writeoffs and an expectation that
really is not going to flatten out till sometime in the beginning of next year,
so with those two changes in point of view, and information that we were
seeing out there, it just felt appropriate and prudent to make those
adjustments in this first quarter.
49

Table 1. Propensity-Score Matching Sample Selection
Accruals Approach Special Items Approach
Bath identified 4,557 14,209

Treatment bath identified 3,180 6,174
Treatment bath with propensity score 1,198 3,324
Treatment bath with conference call data 327 1,137
Treatment bath after propensity-score matching 254 442
This table illustrates how the sample size for treatment baths changes after propensity-score matching.
50

Table 2. Propensity-Score Matching Logit Regression
(1) (2)
Accruals Approach Special Items Approach
VARIABLES BATH BATH
SIZE_LAG -0.361*** 0.0964***

(-16.67) (9.463)
BTM_LAG 0.136*** 0.103***
(3.438) (3.938)
REVENUE_LAG -0.0775* 0.220***
(-1.787) (8.057)
INCOME_LAG -2.473*** -1.157***
(-23.02) (-14.67)
RETURN_A_LAG -0.0851* -0.212***
(-1.677) (-7.391)
TURNOVER_A_LAG 0.0672*** 0.0270***
(5.956) (3.767)
AMIHUD_A_LAG -0.0926*** -0.0958***
(-2.675) (-4.137)
LEVERAGE 0.177*** 0.142***
(11.68) (13.16)
SALEGROWTH -0.148*** -0.0788**
(-2.917) (-2.439)
INSTOWN -0.424*** 0.504***
(-4.127) (10.07)
ANALYST 0.138*** -0.129***
(3.773) (-6.791)
Observations 33,585 32,854

Industry FE Yes Yes
Year FE Yes Yes
ROC 82.36% 70.76%
Pseudo R2 0.163 0.0959
This table represents results of logit regressions for propensity-score matching, under both the Accruals Approach and
Special Items Approach. Dependent variable is the big bath indicator. Explanatory variables include lag value of SIZE
(log of total assets), BTM (book to market ratio), REVENUE (revenue scaled by total assets), INCOME (net income
scaled by total assets), RETURN_A (annual return), TURNOVER_A (annual turnover), and AMIHUD_A (annual
Amihud); explanatory variables also include LEVERAGE (leverage), SALEGROWTH (sales growth year over year),
INSTOWN (institutional ownership), ANALYST (analyst coverage). Industry fixed effects and year fixed effects are
included. ROC represents the area under the ROC curve. Z-statistics in parentheses *** p<0.01, ** p<0.05, * p<0.1.
51

Table 3. Panel A: Descriptive Statistics – Accruals Approach
TREAT VARIABLE MEAN SD P25 P50 P75
0 SIZE 6.09 1.59 4.94 5.94 7.13

BTM 0.71 0.76 0.28 0.52 0.98
INCOME -0.06 0.20 -0.12 0.01 0.06
RETURN_A_LAG 0.02 0.52 -0.35 -0.01 0.26
LEVERAGE 0.93 1.87 0.00 0.18 0.78
SALEGROWTH 0.13 0.49 -0.08 0.06 0.22
INSTOWN 0.54 0.38 0.21 0.57 0.88
ANALYST 2.06 0.99 1.79 2.25 2.71
GENKNLREF 0.05 0.08 0.00 0.03 0.07
POSEMOEXTR 0.70 0.34 0.46 0.64 0.90
SHVALUE 0.01 0.03 0.00 0.00 0.00
ANX 0.10 0.12 0.00 0.06 0.13
1 SIZE 6.22 1.77 4.90 6.07 7.50

BTM 0.54 0.72 0.16 0.45 0.82
INCOME -0.23 0.25 -0.33 -0.13 -0.06
RETURN_A_LAG 0.00 0.53 -0.33 -0.07 0.20
LEVERAGE 0.91 1.75 0.01 0.21 0.85
SALEGROWTH 0.08 0.48 -0.15 0.01 0.23
INSTOWN 0.55 0.39 0.20 0.58 0.87
ANALYST 2.05 1.16 1.39 2.30 2.89
GENKNLREF 0.05 0.08 0.00 0.02 0.07
POSEMOEXTR 0.75 0.36 0.50 0.71 0.94
SHVALUE 0.01 0.03 0.00 0.00 0.00
ANX 0.10 0.12 0.00 0.07 0.15
This table exhibits descriptive statistics, under the Accruals Approach, of key variables in the bath-year (or matched
year for non-bath group) for the final sample after propensity-score matching. Descriptive statistics for control and
treatment group are displayed separately in the top and bottom half of the table. Variables listed include SIZE (log of
total assets), BTM (book to market ratio), INCOME (net income scaled by total assets), RETURN_A_LAG (lag of
annual return), LEVERAGE (leverage), SALEGROWTH (sales growth year over year), INSTOWN (institutional
ownership), and ANALYST (analyst coverage). Also listed are descriptive statistics of word frequencies for deceptive
words GenKnlRef (reference to general knowledge), PosEmoExtr (extreme positive emotions), ShareValue
(shareholder value), and ANX (anxiety).
52

Table 3. Panel B: Descriptive Statistics – Special Items Approach
TREAT VARIABLE MEAN SD P25 P50 P75
0 SIZE 7.16 2.16 5.55 7.21 8.60

BTM 0.49 0.50 0.19 0.37 0.63
INCOME -0.01 0.23 0.00 0.04 0.09
RETURN_A_LAG 0.25 0.54 -0.06 0.15 0.43
LEVERAGE 0.45 1.07 0.00 0.11 0.50
SALEGROWTH 0.17 0.43 0.00 0.09 0.22
INSTOWN 0.64 0.37 0.32 0.72 0.94
ANALYST 2.38 1.09 1.95 2.48 3.26
GENKNLREF 0.06 0.11 0.00 0.03 0.08
POSEMOEXTR 0.75 0.41 0.47 0.66 0.94
SHVALUE 0.01 0.03 0.00 0.00 0.00
ANX 0.11 0.12 0.00 0.08 0.16
1 SIZE 7.26 1.99 5.80 7.19 8.56

BTM 0.54 0.49 0.25 0.44 0.72
INCOME -0.02 0.16 -0.05 0.02 0.06
RETURN_A_LAG 0.31 0.56 -0.04 0.19 0.51
LEVERAGE 0.48 0.85 0.04 0.22 0.59
SALEGROWTH 0.17 0.43 0.00 0.09 0.21
INSTOWN 0.65 0.37 0.40 0.75 0.95
ANALYST 2.34 1.11 1.95 2.64 3.14
GENKNLREF 0.05 0.09 0.00 0.03 0.07
POSEMOEXTR 0.74 0.36 0.51 0.70 0.93
SHVALUE 0.01 0.04 0.00 0.00 0.00
ANX 0.10 0.12 0.00 0.07 0.15
This table exhibits descriptive statistics, under the Special Items Approach, of key variables in the bath-year (or
matched year for non-bath group) for the final sample after propensity-score matching. Descriptive statistics for
control and treatment group are displayed separately in the top and bottom half of the table. Variables listed include
SIZE (log of total assets), BTM (book to market ratio), INCOME (net income scaled by total assets),
RETURN_A_LAG (lag of annual return), LEVERAGE (leverage), SALEGROWTH (sales growth year over year),
INSTOWN (institutional ownership), and ANALYST (analyst coverage). Also listed are descriptive statistics of word
frequencies for deceptive words GenKnlRef (reference to general knowledge), PosEmoExtr (extreme positive
emotions), ShareValue (shareholder value), and ANX (anxiety).
53

Table 4. Panel A: Variable Differences after Propensity-Score Matching - Accruals Approach
Matching Variable Difference T-test (P-value) K-S (P-value)
SIZE_LAG -0.15 0.34 0.11

BTM_LAG 0.07 0.23 0.14
REVENUE_LAG 0.03 0.61 0.09
INCOME_LAG 0.06 0.00 0.00
RETURN_A_LAG 0.02 0.67 0.25
TURNOVER_A_LAG -0.12 0.63 0.14
AMIHUD_A_LAG -0.01 0.84 0.62
LEVERAGE 0.02 0.91 0.14
SALEGROWTH 0.05 0.23 0.04
INSTOWN 0.00 0.95 1.00
ANALYST 0.01 0.90 0.37
PROPENSITY-SCORE 0.00 1.00 1.00
This table illustrates the difference of key matching variables between treatment group and control group after
propensity-score matching, for the Accruals Approach. Matching variables include lag value of SIZE (log of total
assets), BTM (book to market ratio), REVENUE (revenue scaled by total assets), INCOME (net income scaled by
total assets), RETURN_A (annual return), TURNOVER_A (annual turnover), and AMIHUD_A (annual Amihud);
key matching variables also include LEVERAGE (leverage), SALEGROWTH (sales growth year over year),
INSTOWN (institutional ownership), ANALYST (analyst coverage). The difference in propensity-score is also
provided. In testing the difference of these matching variables between treatment and control group, t-test and
Kolmogorov-Smirnov test are performed, with p-values provided in the table.
54

Table 4. Panel B: Variable Differences after PSM - Special Items Approach
Matching Variable Difference T-test (P-value) K-S (P-value)

SIZE_LAG -0.04 0.78 0.53
BTM_LAG 0.00 0.97 0.08
REVENUE_LAG -0.13 0.01 0.00
INCOME_LAG -0.03 0.01 0.23
RETURN_A_LAG -0.06 0.08 0.26
TURNOVER_A_LAG 0.01 0.93 0.17
AMIHUD_A_LAG -0.05 0.06 0.86
LEVERAGE -0.03 0.65 0.00
SALEGROWTH 0.00 0.93 0.81
INSTOWN -0.01 0.69 0.64
ANALYST 0.04 0.62 0.34
PROPENSITY-SCORE 0.00 1.00 1.00
This table illustrates the difference of key matching variables between treatment group and control group after
propensity-score matching, for the Special Items Approach. Matching variables include lag value of SIZE (log of total
assets), BTM (book to market ratio), REVENUE (revenue scaled by total assets), INCOME (net income scaled by
total assets), RETURN_A (annual return), TURNOVER_A (annual turnover), and AMIHUD_A (annual Amihud);
key matching variables also include LEVERAGE (leverage), SALEGROWTH (sales growth year over year),
INSTOWN (institutional ownership), ANALYST (analyst coverage). The difference in propensity-score is also
provided. In testing the difference of these matching variables between treatment and control group, t-test and
Kolmogorov-Smirnov test are performed, with p-values provided in the table.
55

Table 5. Pearson Correlation Table – Accruals Approach
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
AMIHUD_M (1)
SPREAD_M (2) 0.87
TREAT (3) 0.02 0.04
POST (4) 0.00 0.01 0.00
DECEPTION (5) -0.01 -0.01 0.02 0.00
SIZE (6) -0.39 -0.48 0.03 0.00 -0.03
BTM (7) 0.12 0.14 -0.11 -0.05 -0.03 0.00
INCOME (8) -0.08 -0.16 -0.33 0.11 0.12 0.34 0.11
LEVERAGE (9) 0.02 0.04 -0.01 -0.02 -0.09 0.33 -0.11 0.04
RETURN_A_LAG (10) -0.12 -0.17 -0.02 0.00 0.04 0.03 -0.10 0.08 -0.11
INSTOWN (11) -0.28 -0.38 0.01 -0.01 0.05 0.29 -0.13 0.13 -0.11 0.14
ANALYST (12) -0.35 -0.44 -0.01 0.01 0.03 0.44 -0.17 0.11 -0.06 0.07 0.63
EARNING_SHOCK (13) -0.08 -0.09 0.10 -0.09 -0.06 0.26 -0.03 0.03 0.27 -0.01 0.12 0.20
ROA_CHANGE (14) 0.10 0.10 -0.07 0.00 0.02 -0.11 -0.02 0.02 -0.03 0.02 -0.01 -0.04 -0.03
BEAT (15) -0.03 -0.04 -0.03 -0.01 -0.01 -0.02 -0.02 0.04 -0.05 0.03 0.05 0.04 -0.05 0.03
DELTA (16) -0.05 -0.06 -0.09 0.00 0.09 -0.07 -0.07 0.11 -0.09 0.09 0.11 0.03 -0.05 0.01 0.05
VEGA (17) -0.03 -0.03 0.02 0.00 0.08 -0.07 -0.05 0.06 -0.06 0.02 0.06 0.03 -0.03 -0.01 -0.01 0.71
This table presents the correlations among regression variables for the main regressions under the Accruals Approach. Numbers that are significant at least at 10%
level are displayed in bold format.
56

Table 6. Panel A: Main Regression – Accruals Approach
(1) (2) (3) (4)

AMIHUD AMIHUD SPREAD SPREAD
VARIABLES THREE SIX THREE SIX
MONTH MONTH MONTH MONTH
TREAT×POST×DECEPTION 0.0957*** 0.105*** 0.196*** 0.191**

(2.844) (3.118) (2.640) (2.291)
TREAT 0.132*** 0.134*** 0.311*** 0.282***

(3.004) (3.246) (2.832) (2.722)
POST 0.00431 0.00813 0.0650** 0.0505
(0.352) (0.619) (2.008) (1.420)
DECEPTION 0.0775* 0.0723* 0.168* 0.149
(1.713) (1.780) (1.674) (1.534)
TREAT×POST -0.0311 -0.0570** -0.120** -0.104*
(-1.503) (-2.251) (-2.331) (-1.725)
TREAT×DECEPTION -0.188*** -0.180*** -0.338** -0.326**
(-3.142) (-3.209) (-2.367) (-2.407)
POST×DECEPTION -0.0605*** -0.0598*** -0.147*** -0.122**
(-2.744) (-2.796) (-3.262) (-2.448)
SIZE -0.104*** -0.101*** -0.306*** -0.297***

(-6.980) (-7.325) (-8.655) (-8.958)
BTM 0.0740*** 0.0623*** 0.207*** 0.188***
(3.318) (3.214) (4.047) (4.054)
INCOME 0.0617 0.0717 -0.0161 -0.0352
(0.891) (1.164) (-0.0926) (-0.213)
LEVERAGE 0.0289*** 0.0270*** 0.0941*** 0.0845***
(3.315) (3.231) (4.463) (3.993)
RETURN_A_LAG -0.0378* -0.0358* -0.104* -0.115**
(-1.915) (-1.961) (-1.847) (-2.269)
INSTOWN -0.149*** -0.131*** -0.499*** -0.465***
(-3.412) (-3.010) (-4.531) (-4.252)
ANALYST -0.00984 -0.0116 -0.0458 -0.0473
(-0.537) (-0.666) (-1.019) (-1.076)
EARNING_SHOCK -0.00726 -0.00749 -0.0101 -0.0100
(-0.652) (-0.725) (-0.354) (-0.369)
ROA_CHANGE 0.00321** 0.00289** 0.00738** 0.00732**
(1.976) (1.985) (2.038) (2.111)
BEAT -0.0975* -0.0974* -0.211 -0.191
(-1.748) (-1.942) (-1.631) (-1.553)
DELTA -0.0348 -0.0401 -0.103 -0.120
(-0.835) (-1.112) (-0.950) (-1.270)
57

VEGA -0.102 -0.0940* -0.211 -0.183
(-1.562) (-1.770) (-1.309) (-1.405)
DELTA_M 0.0419 0.0930 0.431** 0.395**
(0.443) (1.230) (2.092) (2.174)
VEGA_M -0.0994 -0.135* -0.534*** -0.473***
(-1.007) (-1.710) (-2.646) (-2.634)
GENKNLREF_PRIOR -0.0656 -0.0815 -0.132 -0.131
(-0.243) (-0.316) (-0.201) (-0.203)
POSEMOEXTR_PRIOR 0.0629 0.0734 0.181 0.193
(0.888) (1.043) (1.025) (1.082)
SHVALUE_PRIOR 0.844 0.788 2.364 2.245
(0.778) (0.801) (0.966) (0.980)
ANX_PRIOR -0.137 -0.171 -0.527 -0.522
(-0.739) (-1.024) (-1.316) (-1.349)
Observations 3,048 6,096 3,048 6,096

Adjusted R2 0.324 0.311 0.425 0.418
Year FE Yes Yes Yes Yes
Industry FE Yes Yes Yes Yes
Cluster Firm Firm Firm Firm
This table represents results for the main OLS regressions under the Accruals Approach. Amihud and Spread are
dependent variables, with regressing results shown both the three-month and the six-month as pre and post event
window length. Explanatory variables include TREAT, POST, DECEPTION, interaction terms of the three variables,
triple interaction term of TREAT×POST×DECEPTION. Control variables include SIZE (log of total assets), BTM
(book to market ratio), INCOME (income scaled by total assets), RETURN_A_LAG (lag of annual return),
LEVERAGE (leverage), SALEGROWTH (sales growth year over year), INSTOWN (institutional ownership),
ANALYST (analyst coverage), EARNING_SHOCK (absolute earnings shock), ROA_CHANGE (change in ROA
year over year), BEAT (indicator variable of beating analyst consensus forecast), DELTA (CEO compensation delta),
VEGA (CEO compensation vega), DELTA_M (indicator variable of whether delta is missing), VEGA_M (indicator
variable of whether vega is missing), GENKNLREF_PRIOR (average word frequency of “reference to general
knowledge” words in prior conference calls), POSEMOEXTR_PRIOR (average word frequency of “extreme positive
emotions” words in prior conference calls), SHVALUE_PRIOR (average word frequency of “shareholder value”
words in prior conference calls), and ANX_PRIOR (average word frequency of “anxiety” words in prior conference
calls). Industry fixed effects and year fixed effects are included. Standard errors are clustered at firm level. Robust t-
statistics in parentheses *** p<0.01, ** p<0.05, * p<0.1
58

Table 6. Panel B: Main Regression – Special Items Approach
(1) (2) (3) (4)

TREAT×POST×DECEPTION 0.0280** 0.0333** 0.0645** 0.0590**

(2.210) (2.524) (2.440) (2.113)
TREAT 0.0267 0.0265* 0.0206 0.0177

(1.644) (1.745) (0.508) (0.482)
POST 0.0133* 0.0138* 0.0282* 0.0281*
(1.687) (1.807) (1.755) (1.861)
DECEPTION -0.00215 0.00618 -0.0280 -0.00951
(-0.141) (0.409) (-0.625) (-0.230)
TREAT×POST -0.0156* -0.0195** -0.0323 -0.0196
(-1.721) (-2.097) (-1.607) (-1.000)
TREAT×DECEPTION -0.0234 -0.0251 -0.00305 -0.0145
(-1.046) (-1.150) (-0.0521) (-0.265)
POST×DECEPTION -0.0136 -0.0194** -0.0412** -0.0467**
(-1.494) (-2.124) (-2.180) (-2.364)
SIZE -0.0397*** -0.0382*** -0.138*** -0.132***

(-6.711) (-6.688) (-10.76) (-10.76)
BTM 0.0619** 0.0629** 0.123** 0.132**
(2.143) (2.269) (2.004) (2.311)
INCOME 0.0598* 0.0626* 0.0102 0.0257
(1.662) (1.894) (0.116) (0.312)
LEVERAGE 0.0259* 0.0260* 0.109*** 0.0933***
(1.916) (1.872) (2.954) (2.804)
RETURN_A_LAG -0.0354*** -0.0338*** -0.0739*** -0.0697***
(-3.877) (-4.058) (-2.915) (-3.265)
INSTOWN -0.0947*** -0.0920*** -0.347*** -0.342***
(-3.922) (-3.888) (-5.831) (-6.173)
ANALYST -0.0173** -0.0184** -0.0524** -0.0509**
(-2.112) (-2.236) (-2.321) (-2.431)
EARNING_SHOCK -0.00226 -0.00267 0.0104 0.0148
(-0.226) (-0.284) (0.341) (0.573)
ROA_CHANGE 0.000435 0.00184 0.00180 0.00422
(0.215) (0.886) (0.425) (1.012)
BEAT -0.00418 -0.00446 0.0283 0.0344
(-0.296) (-0.350) (0.776) (1.065)
DELTA -0.00891*** -0.00853*** -0.0303*** -0.0285***
(-3.266) (-3.035) (-4.222) (-3.876)
VEGA -0.106* -0.105* -0.335* -0.355**
59

(-1.681) (-1.806) (-1.962) (-2.254)
DELTA_M 0.0654 0.0408 0.183 0.121
(1.173) (0.716) (1.026) (0.628)
VEGA_M -0.0946* -0.0729 -0.224 -0.168
(-1.665) (-1.254) (-1.241) (-0.860)
GENKNLREF_PRIOR -0.124 -0.117 -0.312 -0.276
(-1.499) (-1.456) (-1.504) (-1.397)
POSEMOEXTR_PRIOR 0.0292 0.0378 0.0513 0.0666
(0.996) (1.327) (0.741) (1.044)
SHVALUE_PRIOR -0.668* -0.637* -0.848 -0.674
(-1.789) (-1.713) (-0.995) (-0.768)
ANX_PRIOR -0.0713 -0.0625 -0.111 -0.0834
(-0.949) (-0.896) (-0.568) (-0.475)
Observations 5,304 10,608 5,304 10,608

Adjusted R2 0.259 0.254 0.434 0.427
This table represents results for the main OLS regressions under the Special Items Approach. Amihud and Spread are
dependent variables, with regressing results shown both the three-month and the six-month as pre and post event
window length. Explanatory variables include TREAT, POST, DECEPTION, interaction terms of the three variables,
triple interaction term of TREAT×POST×DECEPTION. Control variables include SIZE (log of total assets), BTM
(book to market ratio), INCOME (income scaled by total assets), RETURN_A_LAG (lag of annual return),
LEVERAGE (leverage), SALEGROWTH (sales growth year over year), INSTOWN (institutional ownership),
ANALYST (analyst coverage), EARNING_SHOCK (absolute earnings shock), ROA_CHANGE (change in ROA
year over year), BEAT (indicator variable of beating analyst consensus forecast), DELTA (CEO compensation delta),
VEGA (CEO compensation vega), DELTA_M (indicator variable of whether delta is missing), VEGA_M (indicator
variable of whether vega is missing), GENKNLREF_PRIOR (average word frequency of “reference to general
knowledge” words in prior conference calls), POSEMOEXTR_PRIOR (average word frequency of “extreme positive
emotions” words in prior conference calls), SHVALUE_PRIOR (average word frequency of “shareholder value”
words in prior conference calls), and ANX_PRIOR (average word frequency of “anxiety” words in prior conference
calls). Industry fixed effects and year fixed effects are included. Standard errors are clustered at firm level. Robust t-
statistics in parentheses *** p<0.01, ** p<0.05, * p<0.1
60

Table 7. Alternative Information Asymmetry Measures
(1) (2) (3) (4)

TURNOVER TURNOVER DOL_VOL DOL_VOL
TREAT×POST×DECEPTION -0.154* -0.183** -0.375*** -0.424***

(-1.933) (-2.177) (-3.092) (-3.199)
Observations 3,048 6,096 3,048 6,096

Adjusted R2 0.420 0.423 0.761 0.765
Controls Yes Yes Yes Yes
This table represents results for the OLS regressions under the Accruals Approach, using alternative information
asymmetry proxies. Turnover and Dollar Volume are dependent variables, with regressing results shown both the
three-month and the six-month as pre and post event window. Explanatory variables include TREAT, POST,
DECEPTION, interaction terms of the three variables, triple interaction term of TREAT×POST×DECEPTION.
Control variables include SIZE (log of total assets), BTM (book to market ratio), INCOME (income scaled by total
assets), RETURN_A_LAG (lag of annual return), LEVERAGE (leverage), SALEGROWTH (sales growth year over
year), INSTOWN (institutional ownership), ANALYST (analyst coverage), EARNING_SHOCK (absolute earnings
shock), ROA_CHANGE (change in ROA year over year), BEAT (indicator variable of beating analyst consensus
forecast), DELTA (CEO compensation delta), VEGA (CEO compensation vega), DELTA_M (indicator variable of
whether delta is missing), VEGA_M (indicator variable of whether vega is missing), GENKNLREF_PRIOR (average
word frequency of “reference to general knowledge” words in prior conference calls), POSEMOEXTR_PRIOR
(average word frequency of “extreme positive emotions” words in prior conference calls), SHVALUE_PRIOR
(average word frequency of “shareholder value” words in prior conference calls), and ANX_PRIOR (average word
frequency of “anxiety” words in prior conference calls). Industry fixed effects and year fixed effects are included.
Standard errors are clustered at firm level. Robust t-statistics in parentheses *** p<0.01, ** p<0.05, * p<0.1
61

Table 8. Mechanism Tests
(1) (2) (3) (4)

AMIHUD AMIHUD AMIHUD AMIHUD
LOW HIGH LOW HIGH
VARIABLES THREE THREE SIX SIX
Institutional Ownership
TREAT×POST×DECEPTION 0.0257 0.174*** 0.0538 0.161***
(0.635) (3.207) (1.374) (3.021)
Observations 1,524 1,524 3,048 3,048

Adjusted R2 0.358 0.400 0.333 0.390
Analyst Coverage
TREAT×POST×DECEPTION 0.0554 0.144*** 0.0777* 0.142***
(1.347) (2.788) (1.917) (2.748)
Observations 1,524 1,524 3,048 3,048

Adjusted R2 0.368 0.410 0.351 0.385
Press Coverage
TREAT×POST×DECEPTION 0.00171 0.212*** 0.0114 0.224***
(0.0368) (4.315) (0.267) (4.427)
Observations 1,524 1,524 3,048 3,048

Adjusted R2 0.321 0.425 0.304 0.417
Bath Magnitude
TREAT×POST×DECEPTION 0.0642 0.128** 0.0830** 0.121**
(1.604) (2.099) (2.214) (2.008)
Observations 1,524 1,524 3,048 3,048

Adjusted R2 0.265 0.434 0.243 0.426

This table represents a summary of the cross-sectional test results, with institutional ownership, analyst coverage,
press coverage, and bath magnitude as partition variables. Amihud is the dependent variable, with both three-month
and six-month as event windows. Treatment baths are identified following the Accruals Approach. Control variables,
industry fixed effects, and year fixed effects are included. Standard errors are clustered at firm level. Robust t-statistics
in parentheses *** p<0.01, ** p<0.05, * p<0.1
62

Table 9. Separation of Prepared Remarks and Q&A
(1) (2) (3) (4)

PR Q&A PR Q&A
VARIABLES SIX MONTH SIX MONTH SIX MONTH SIX MONTH
TREAT×POST×DECEPTION_PR 0.0147 0.0755

(0.394) (0.644)
TREAT×POST×DECEPTION_QA 0.0744** 0.165*
(2.196) (1.939)
Observations 6,096 6,096 6,096 6,096

Adjusted R2 0.304 0.310 0.413 0.419
This table represents results using prepared remarks and Q&A section in conference call, separately, to indicate
management deception. DECEPTION_PR is an indicator variable if a CEO is identify as deceptive based on linguistic
features in prepared remarks; DECEPTION_QA is an indicator variable if a CEO is identify as deceptive based on
linguistic features in Q&A. Amihud and Spread are used as dependent variables, with six-month as the event window.
Control variables, industry fixed effects, and year fixed effects are included. Standard errors are clustered at firm level.
Robust t-statistics in parentheses *** p<0.01, ** p<0.05, * p<0.1
63

Table 10. Partition on Prior Deception Level
(1) (2) (3) (4)

AMIHUD AMIHUD AMIHUD AMIHUD
PRIOR PRIOR PRIOR PRIOR
TRUTHFUL DECEPTIVE TRUTHFUL DECEPTIVE
VARIABLES THREE THREE SIX SIX
TREAT×POST×DECEPTION 0.143** 0.0489 0.144*** 0.0325

(2.298) (1.060) (3.055) (0.680)
Observations 1,524 1,524 3,048 3,048

Adjusted R2 0.378 0.338 0.365 0.337
This table presents results of tests partition on prior CEO deception level. Prior truthful and prior deceptive are related
to prior deception level of a CEO, identified based on the linguistic features in all prior conference calls. Amihud is
the dependent variable, with both three-month and six-month as event windows. Control variables, industry fixed
effects, and year fixed effects are included. Standard errors are clustered at firm level. Robust t-statistics in parentheses
*** p<0.01, ** p<0.05, * p<0.1
64

S06-Management Deception PDF

Uploaded by

Copyright:

Available Formats

S06-Management Deception PDF

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

S06-Management Deception PDF

Uploaded by

Copyright:

Available Formats

Management Deception, Big-Bath Accounting, and Information Asymmetry:

Evidence from Linguistic Analysis

February 26, 2018

Electronic copy available at: https://ssrn.com/abstract=3003259

Evidence from Linguistic Analysis

management ethics. Specifically, we investigate whether managers’ truthfulness (or conversely,

conference calls to measure managerial deception and employing a difference-in-differences

research design with propensity-score matching, we find that information asymmetry is

by less deceptive CEOs.

Keywords: Management Deception; Big-Bath Accounting; Linguistic Analysis; Information

Electronic copy available at: https://ssrn.com/abstract=3003259

Evidence from Linguistic Analysis

conversely, deceptiveness) affects investors’ perceptions of big baths.

Electronic copy available at: https://ssrn.com/abstract=3003259

responsibility to the firm’s stakeholders.

Electronic copy available at: https://ssrn.com/abstract=3003259

manager so as to display an improved financial performance in future (Moore 1973).

market, idiosyncratic information risk is priced when information is subject to managers’

calls, sector sentiment, and overall market direction.3

Electronic copy available at: https://ssrn.com/abstract=3003259

is built up by applying a well-developed and frequently used psychosocial dictionary – LIWC.

evidence substantiating the usefulness of CEOs’ linguistic patterns in signifying deception is

Electronic copy available at: https://ssrn.com/abstract=3003259

decision making, we focus on the linguistic pattern of CEOs.

In our primary analyses we employ a difference-in-differences research design coupled with

asymmetry is affected accordingly. Specifically, we find that information asymmetry (proxied

specifications and other robustness tests.

its associated capital-market consequences in the setting of big-bath taking, an economically

Electronic copy available at: https://ssrn.com/abstract=3003259

2. Prior Literature and Hypothesis Development

2.1 Big-Bath Literature

economic consequences, reflect decisions by corporate management. Kirschenheiter and Melumad

Electronic copy available at: https://ssrn.com/abstract=3003259

to increase compensation or maintain their positions in firms.

favorable-resolution hypothesis in which write-offs reflect managers’ acknowledgement of

often interprets large write-offs positively.

Electronic copy available at: https://ssrn.com/abstract=3003259

Electronic copy available at: https://ssrn.com/abstract=3003259

in DICTION, namely Certainty, Realism, and Commonality.

2.3 Linguistic Analysis and the Conference-Call Literature

forecasts and stock recommendations.

Importantly, linguistic analysis can be applied to assess the likelihood of management

Subjectivity is also removed by relying on well-established dictionaries. In a well-cited study,

Electronic copy available at: https://ssrn.com/abstract=3003259

speech during quarterly earnings-conference calls.9

employed other textual-analysis techniques) in relation to big-bath accounting.

2.4 Hypothesis Development

Electronic copy available at: https://ssrn.com/abstract=3003259

Research in psychology indicates that the credibility or reputation of a source influences

environment and management’s reporting reputation affect the importance of incentive

Electronic copy available at: https://ssrn.com/abstract=3003259

existence of asymmetric information can enable market participants to engage in deleterious

strategic behaviors that would be impossible in a world of complete information. Hidden

Electronic copy available at: https://ssrn.com/abstract=3003259

3. Data and Research Design

3.1 Big Baths

proxies for big baths.

we employ the performance-adjusted version of Kothari, Leone, and Wasley’s discretionary