are credit markets tone deaf? evidence from credit default ... · we also contribute to a growing...

Are Credit Markets Tone Deaf? Evidence from Credit Default Swaps

Hitesh Doshi a

University of Houston

Saurin Patel b

Western University

Srikanth Ramani c

University of New Brunswick

Matthew Sooy b

Western University

February 19, 2019

Acknowledgements: The authors would like to thank Jeffrey Callen, Craig Dunbar, Partha Mohanram, Stephen

Saap and brownbag seminar participants at the Midwest Finance Conference, University of Houston and Western

University for insightful discussions on this research project. We also thank an anonymous reviewer from FARS

Conference for constructive comments. Patel acknowledges financial support from the Social Sciences and

Humanities Research Council (SSHRC).

a – University of Houston, Bauer College of Business, Houston, TX 77204, USA. at [email protected]

b – Western University, Ivey Business School, London, ON N6G0N1, Canada. [email protected],

[email protected]. Matthew Sooy is corresponding author.

c – University of New Brunswick, Faculty of Business, Saint John, NB, E2L4L5, Canada. [email protected]

Electronic copy available at: https://ssrn.com/abstract=3311776

mailto:[email protected]





ABSTRACT: We examine credit market responses to uncertain linguistic tone in disclosures made

in 10-Q/K fillings, controlling for the information content conveyed in the reports. Examining

windows around quarterly filings, we find that uncertain tone levels are associated with changes

in credit default swap (CDS) spreads and are incremental to spread changes implied via equity

market responses to the same information. We also observe that the magnitude of the relationship

varies according to longitudinal and cross-sectional contract attributes specific to debt, with CDS

spread responses monotonically increasing as contracts approach maturity, and increasing also in

indicators of firms’ default risk. Our results are robust to other sources of risk such as earnings

surprises, management guidance, special firm-specific events firm, and alternative proxies of

uncertainty. Our findings are consistent with market participants linking uncertainty in disclosure

language to firm' default risk, implying that the tone of accounting disclosures provides valuable

incremental information to the CDS markets.

JEL classifications: D81; D83; G12; G14; M41

Keywords: Textual Analysis, Default Risk, Uncertainty, Tone, Accounting Disclosure


1


1. Introduction

This study examines credit market responses to uncertain linguistic tone in 10-Q/K filings

as reflected in credit default swap (‘CDS’) spreads. Credit markets command a large and central

($110 trillion notional, BIS 2018) position in worldwide capital markets, and accounting research

has detailed the value relevance of financial information to credit investors.1 Related research has

begun to also document credit investors’ use of the wealth of textual information conveyed in

accounting reports (Chiu et al. 2018).2 We investigate credit market responses to one such type of

textual information, uncertain linguistic tone (see Loughran and McDonald 2011, hereafter LM

2011). We argue that uncertain tone in 10-Q/K filings reflects management’s lack of confidence

in the firm’s business strategy, competitive position and/or future financial wellbeing.

Consequently, we hypothesize that higher uncertain tone in 10-Q/K disclosures will trigger

uncertainty in credit investors’ valuations leading to greater CDS spreads, and will do so

incremental to other known CDS determinants and other linguistic dimensions. We test this central

hypothesis and extend its logic to time-series and cross-sectional predictions described below.

We use dictionaries developed in LM 2011 to measure uncertain tone. Our proxy of

linguistic uncertainty is the proportion of total words in the 10-Q/K that are listed in the Loughran

and McDonald uncertain word list (e.g., approximate, uncertain, indefinite, possible). We link

uncertain tone to credit default swap spreads, our primary measure of credit market response.

Credit default swaps function similar to credit ‘insurance policies’ that pay off only in the event

1 Recent studies investigating the value relevance of financial information to credit investors include Givoly et al.

(2017), Batta et al. (2016), Correia et al. (2012), Shivakumar et al. (2011), Callen et al. (2009), Das et al. (2009), and

Easton et al. (2009). 2 Recent textual studies investigating equity market responses include Huang et al. (2014), Jegadeesh and Wu

(2013), Loughran and McDonald (2013), Loughran and McDonald (2011), Feldman et al. (2010), and Li (2008).


2

of a target firm’s default and thus reflect default risk almost exclusively (and not, for instance,

choices for risk-free benchmarks, liquidity, taxes, or other frictions - see Ericsson et al. 2009; Chen

et al. 2007; Longstaff et al. 2005; Elton et al. 2001). Additionally, CDS markets are standardized,

highly liquid and available at a daily frequency permitting spreads to reflect updated beliefs

(Augustin et al. 2014).

Using a sample constructed from WRDS-Markit daily CDS pricing data for U.S. firms over

the fifteen-year period from 2001 to 2016, we first examine the impact of uncertain disclosure tone

on CDS spread changes. We measure CDS spreads relative to the median CDS spread for all other

firms with the same debt rating (e.g. ‘AAA’, etc.) to control for systematic sources of uncertainty.

We utilize an event study design with an eleven-day (-5, +5) event window centered on the 10-

Q/K filing date event to minimize the possibility that CDS spreads impound other information

(Callen et al. 2009). We include several market-level and firm-level controls for other determinants

of CDS spreads, including controlling for concurrent equity price changes to ensure that our results

are incremental to any response implied by concurrent equity market price changes (Hilscher et al.

2015; Huang et al. 2014; Lok and Richardson 2011). We find a positive relation between uncertain

disclosure tone scores and changes in CDS spreads. Firms with greater uncertainty tone scores

experience a significant increase in their CDS spreads. This effect is both economically and

statistically significant. We find that a one standard deviation increase in uncertain tone scores

corresponds with an increase in incremental CDS spreads of 1.47 basis points (bps) relative to the

mean change across all firms of 0.47 bps.

Next, we advance predictions about time-series and cross-sectional variation in credit

market responses to uncertain disclosure tone. We follow Duffie and Lando (2001) who theorize

that credit spread responses to uncertainty should be greatest among the bonds closest to maturity


3

and nearest to default. We find that CDS spreads with shorter maturities have greater sensitivity

to uncertain tone than long-term maturities, with uncertain tone sensitivities declining

monotonically in time-to-maturity. We also observe credit spread responses to uncertain

disclosure tone are greater in subsamples with high volatility and with high leverage, reflecting

greater ex-ante default risk. Together, our tests address whether and how credit investors impound

uncertain disclosure tone into CDS spreads, addressing also robustness to other known effects,

including those observed in equity market settings (Huang et al. 2014, Lok and Richardson 2011).

Lastly, we investigate several alternative explanations for our results. First, we construct

two tests to address the possibility that observed effects may be driven by concurrently released

positive or negative financial information that causes managers to change tone and investors to

update beliefs (e.g. Shivakumar et al. 2011). We construct a measure of earnings surprise to

control for new financial information released in the 10-Q/K (Callen et al. 2009). We find that

uncertain tone scores contain incremental information about default risk that is not captured by

earnings surprise. To control for material financial information concurrently released in other

disclosures, we build subsamples that exclude firms with 8-K releases or with management

guidance releases that overlap with our event window. We find no difference in our results for the

revised samples. We also explore the possibility that the findings we attribute to uncertain

disclosure tone reflect other textual factors such as readability of 10-Q/K filings or negative tone

(Loughran and McDonald 2014; Lehavy et al. 2011; Li 2008). We control for readability measures

and negative tone in our baseline regression with no change in results, suggesting that our results

are incremental to other known tone dimensions. Lastly, we demonstrate that our results hold for

other specifications of our research design including an alternatively constructed measure of

uncertain tone, taken by netting negative and positive tone (Loughran and McDonald 2013),


4

alternate window sizes, sample constructions, and time periods. Our results are robust to these

specifications.

Together, our findings suggest that credit markets impound uncertain linguistic disclosure

tone into CDS spreads, and that they do so incremental to other known determinants. We

additionally note two ways in which CDS spreads vary in response to uncertain tone, both

according to maturity length and according to ex-ante default risk. Our results make several

important contributions to our understanding of credit markets and to our understanding of market

consequences of linguistic tone. We document an important influence on credit spreads and

demonstrate its incremental impact after controlling for equity market responses to the same

information. This adds to existing accounting research on credit market responses which has

focused primarily on quantitative financial information in the pricing of CDS contracts (e.g. Batta,

et al. 2016; Arora et al. 2014; Kim et al. 2013; Shivakumar et al. 2011; Callen et al. 2009), but

which has also investigated non-quantitative disclosure of specific risks (Chiu et al. 2018). Our

evidence, which observes cross-sectional and time-series variation in the relationship between

uncertain tone and CDS spreads additionally supports credit theory (e.g. Duffie and Lando 2001).

More practically, our study contributes to regulatory discussions on the apparent gap between

firms facing default/going-concern uncertainty and those which disclose it (OSC 2010). Our

findings on the relationship between linguistic tone and CDS spreads contribute to evidence

suggesting that managers are aware of and implicitly communicate these uncertainties even when

they are unwilling to explicitly convey them (see also Mayew et al. 2014).

We also contribute to a growing textual analysis literature (Loughran and McDonald 2016).

Researchers have documented the usefulness of textual analysis in generating value-relevant

information to investors using tone, content and sentiment in newspaper articles (e.g., Tetlock


5

2007), corporate disclosures (e.g., LM 2011, Li 2008), press releases (e.g., Engelberg 2008) as

well as investor message boards (e.g. Antweiler et al. 2004) and its impact on equity valuations.

To our knowledge, this is the first study to apply linguistic tone analysis to credit derivative pricing

and show that disclosure tone of the accounting reports provides value-relevant information to

credit markets. Our research adds also to other work linking positive and negative linguistic tone

in management discussion & analysis (MD&A) to going concern opinions (Mayew et al. 2014).

We document a market-based response to expected default risks, derived from uncertain linguistic

tone that is incremental to responses implied by negative linguistic tone. We further show that

credit investors impound uncertain tone incrementally to other linguistic dimensions such as

readability and negative tone.

This essay proceeds as follows. Section 2 details related research and motivates our

primary predictions. Section 3 describes our sample construction and highlights our research

design. Section 4 reports the results from our empirical tests related to uncertain disclosure tone

and CDS spreads. Section 5 discusses alternative explanations for our results and enumerates

related robustness tests. Section 6 concludes the paper with discussion of our evidence, its

limitations, and opportunities for future research.

2. Background & Hypothesis Development

Background

Credit Markets and Credit Default Swaps

Empirical capital market research has given considerable attention to credit markets, which

are both impressive in magnitude and central to financial markets (BIS 2018). Guided by theory

emphasizing default risks in credit investments, this research has extensively documented

determinants of default risk (Demerjian and Owens 2016; Jacobson et al. 2006; Altman and


6

Saunders 1997; Ohlson 1980; Altman 1968), credit market responses to financial information via

determinants of default risk (Jung et al. 2013; Lok and Richardson 2011; Longstaff 2010; Callen

et al. 2009; Tang 2009), and/or investigating the impact of default risk factors on the amount and

structure of credit issuances (e.g. public/private, covenants, etc.) (Demerjian 2017; Doblas-Madrid

and Minetti 2013; Ball et al. 2008; Barath et al. 2008).

Credit Default Swaps, which serve the function of insuring against default, have enabled

credit market researchers to examine default risks more directly because these instruments reflect

default risk in a very straightforward way (see Ericsson et al. 2009). Recent investigations examine

the relationship between CDS spreads and earning announcements (Callen et al. 2009), cash flow

news in management forecasts (Shivakumar et al. 2011), the quality of internal control and cost of

debt (Tang et al. 2015), the adoption of International Financial Reporting Standards (Bhat et al.

2016), risk factor disclosures in accounting reports (Chiu et al. 2018) among others.3 Separate

from default risk itself, CDS spreads have been theorized and shown to also reflect investor

uncertainties in default risk assessment. Duffie and Lando (2001) model assessment uncertainty

as accounting ‘noise’, reflecting the difference between the current state of the firm and its last

verified state (e.g. audited SEC filings), showing that CDS spreads increase where assessment

uncertainty is greatest such as issuances nearest to maturity and for firms most likely to default.4

We extend this logic to assessment uncertainty stemming from managements’ own beliefs about

the likelihood of future firm states, which can be thought of as the standard deviation of possible

3 See Griffin (2014) and Augustin et al. (2014) for a comprehensive review of accounting and finance research on

CDS. Research also shows that default risk information priced into CDS mechanically link credit and equity markets

(see Lok and Richardson 2011). 4 Researchers also use hybrid models of accounting- and market-based information in pricing financial distress

through CDS contracts (e.g., Correira et al. 2012; Das et al. 2009).


7

future outcomes around a point estimate of firm value. We assert that managers perceive this type

of uncertainty and intentionally or unintentionally convey it in tone as described below.

Textual analysis

Financial reports such as SEC filings are rich in quantitative financial information useful

for assessing default risks, but also convey rich non-quantitative information. While this

information has been critical to market analysts for years, capital markets research has only

recently developed systems for categorizing and analyzing textual communication (see Loughran

and McDonald 2016 and Li 2010b for recent reviews). A growing textual analysis research body

acknowledges that managers embed signals of their knowledge of the firm’s ‘economic reality’

into communication both intentionally and unintentionally, based on the sum of deliberate and

latent language choices (Loughran and McDonald 2016). As it relates to the present study, we

assert that managers’ uncertainty about the future state(s) of the firm will also manifest in

deliberate and/or unintentional language cues identifiable in textual communication. This study

explores in particular a dimension of text that indirectly conveys information – uncertain linguistic

tone. Managers with low confidence in the firm’s business strategy, competitive position within

the industry and/or future financial wellbeing feel constrained in which words they may use to

portray present and future prospects based on reputational and legal risks, resulting in less certain

(more uncertain) language in their disclosures. Managers’ uncertainty exists separately from other

dimensions also embedded into tone such as valence, associated with positive or negative tone.

Several equity market studies examine the effects of various types of linguistic tone,

observing that dimensions of linguistic tone are associated both with market responses to

accounting disclosures and with firms’ current and future performance (Allee and Deangilis 2015;

Mayew et al. 2015; Loughran and McDonald 2013; Davis et al. 2012; Feldman et al. 2010).


8

Research further suggests that managers sometimes use tone to mislead investors, where abnormal

tone is associated with both current positive returns and negative future returns, negative future

performance, and greater future litigation risk (Huang et al. 2014; Rogers et al. 2011). Two studies

link uncertain linguistic tone to equity market responses, observing that ‘ambiguous’ tone is

associated with future stock price risk (Ertugrul et al. 2017), and that uncertain tone is associated

with greater IPO returns and future price volatility (Loughran and McDonald 2015).

Given mechanical linkages between credit and equity markets, care must be taken to ensure

that any observed responses are not mere replications of previously documented effects in equity

markets (Hirschler et al. 2015; Lok and Richardson 2011). We note that credit market valuations

depart from equity valuations in many ways (e.g. emphasizing default risks and corresponding

uncertainties) that would lead credit investors to respond differently and incrementally to uncertain

tone relative to equity investors. As discussed in Section IV, we nevertheless include controls for

concurrent equity market responses to ensure that the credit market responses we observe are

incremental to any response implied by concurrent equity market price changes.

Hypothesis Development

We first make a general prediction about the relationship of managers’ uncertain linguistic

tone to credit market responses. Specifically, we posit that firm managers who are uncertain about

their firms’ future prospects will intentionally and/or unintentionally embed linguistic signals of

their uncertainty into disclosures such as by selecting less certain words. To the extent that SEC

filings offer sufficient flexibility to reflect tone differences and credit investors are sufficiently

sensitive to linguistic cues, uncertain tone will create valuation uncertainty in investors’ minds

resulting in higher credit spreads. Formally put, we first hypothesize:

HYPOTHESIS 1. Firms with higher uncertain disclosure tone have higher changes in CDS

spreads around the disclosure event.


9

We next make specific predictions about time-series and cross-sectional variation in the

relationship between uncertain linguistic tone and CDS spreads. Duffie and Lando (2001) show

that shorter maturity credit spreads (relative to longer) respond more strongly to assessment

uncertainty, controlling for other substantive signals of firm risk. Over longer durations,

assessment uncertainty is dominated by asset evolution risks, leading longer maturity credit

spreads to respond less strongly. That is, uncertainty interacts with the term structure of credit

spreads. This leads us to our second hypothesis:

HYPOTHESIS 2. The impact of uncertain disclosure tone is lower for relatively longer maturity

CDS contracts around the disclosure event, ceteris paribus.

Duffie and Lando (2001) also suggest that the effect of assessment uncertainty on credit

spreads is larger for firms with higher default probabilities as reflected in lower initial asset values.

This logic can be extended to the firm’s underlying volatility and leverage, which are both

theoretically and empirically tied to default risk (Ericsson et al. 2009, Merton 1974). We posit that

the credit spreads of firms with relatively higher asset volatility and those with relatively higher

leverage (which are more likely to default) should be more sensitive to uncertainty signals

compared to firms with comparatively better credit ratings.

HYPOTHESIS 3a. The impact of uncertain disclosure tone is larger for firms with a relatively

higher asset volatility around the disclosure event.

HYPOTHESIS 3b. The impact of uncertain disclosure tone is larger for firms with relatively

greater leverage around the disclosure event.

3. Sample & Research Design

Sample

We test our hypotheses using a sample constructed from CDS data obtained from the

WRDS Markit CDS database for the period 2001-2016. Markit records composite end-of-day CDS

spreads for firms with highly liquid contracts and is widely used both in practice and in research.


10

Our sample begins with the daily data for all single name CDS contracts with 1-, 3-, 5-, 7- and 10-

year maturity. Following prior research, we utilize 5-year CDS contracts for our primary analyses,

as they are most liquid and have the best coverage in the database (Chiu et al. 2018; Bhat et al.

2014; Shivakumar et al. 2011). However, we also use 1-, 3-, 7- and 10-year CDS contracts in our

test of H2, which relates to variation in credit spreads by maturity. To maintain uniformity in

contracts, we limit our sample to CDS spreads for senior unsecured debt with a modified restricting

(MR) clause and denominated in US dollars (see also Chiu et. al. 2018; Jorion and Zhang 2007).

We combine CDS spread data with disclosure tone data collected from WRDS SEC

analytics suite-Readability and Sentiment Analysis database. The database records the lexical

features of the language used in SEC-filed financial reports (e.g. readability and linguistic

complexity) including the uncertainty/weak-model words and total words employed for each

report. We construct our measure of uncertain tone for each quarterly 10-Q/K disclosure using the

count of words from the report that appear in the Loughran-McDonald Financial-Uncertainty

words list (Fin-Unc) and then scale by the total number of words in the same report (see LM 2011).

The Loughran-McDonald Financial-Uncertainty words list includes 285 words that denote

uncertainty through emphasis on imprecision rather than risk, particularly in business/financial

context (e.g. approximate, uncertain, depends, unpredictable and indefinite).5 Other readability

measures such as Flesch-Kincaid index, Fog index, Coleman-Liau index, Harvard general inquirer

negative index used in robustness analyses are also taken from WRDS SEC analytics suite. We

match the daily CDS data with the disclosure tone data on the SEC filing date by linking permco,

gvkey, cik unique identifiers. Our initial match contains 809 firms and 29,688 firm-quarter

observations.

5 The complete word list is available at http://www.nd.edu/~mcdonald/Word_Lists.html.


http://www.nd.edu/~mcdonald/Word_Lists.html

11

We also construct a number of control variables using firm-specific and economy-wide

data. We rely on COMPUSTAT for firm-specific control variables such as size, leverage, and

return on assets, merged with our sample using gvkey. We use the CRSP database to obtain equity

prices, daily stock return and the number of shares outstanding. With this data, we calculate the

event period equity return (EPER), and realized volatility (Rvol), matched using cusip. Our final

sample has 27,655 firm-quarter observations and 798 firms.

Lastly, we build several economy-wide control variables using data obtained from two

sources. We use the Federal Reserve Economic Data (FRED) database of the Federal Reserve

Bank of St. Louis to determine the Risk-free rate (Rf) using the three-month treasury-bill yield,

Term Spread (TS) measured as the difference between the yields of a ten-year and one-year

government bond, and Default Spread (DS) measured as the difference between the yield of

Moody’s Baa corporate bond and the yield of the ten-year constant maturity treasury bond. We

use data from the Chicago Board Options Exchange (CBOE) website to construct measures of

aggregate investors’ risk appetite or market uncertainty by using a measure of implied volatility

from S&P 500 index option prices with 30-day maturity, better known as VIX.

Empirical Design

We examine the impact of disclosure tone on the changes in CDS spreads using a short-

window event study design. Event studies use firms as their own control in the non-window

sample, mitigating a number of research challenges including heteroscedasticity concerns (Callen

et al. 2009). We measure CDS spread changes by subtracting the median change experienced by

all other firms with the same credit rating to control for systematic market responses. This enables

our study to associate otherwise unexplained credit market responses in a short event window with

the linguistic tone conveyed within accounting reports released during the window. Changes


12

designs also represent a more conservative test of CDS spread determinants (Ericsson et al. 2009).

Our baseline regression specification is the following:

∆𝐶𝐷𝑆 𝑆𝑝𝑟𝑒𝑎𝑑 = 𝛽0 + 𝛽1𝑈𝑁𝐶𝑇𝑂𝑁𝐸 + 𝛽2∆𝑆𝑖𝑧𝑒 + 𝛽3∆𝐿𝑒𝑣 + 𝛽4∆𝑅𝑂𝐴 + 𝛽5𝐸𝑃𝐸𝑅

+𝛽6𝑅𝑣𝑜𝑙 + 𝛽7∆𝑉𝐼𝑋 + 𝛽8∆𝑅𝑓 + 𝛽9∆𝑇𝑆 + 𝛽10∆𝐷𝑆 + 𝐹𝐸𝑠 + 𝜀 (1)

The dependent variable, ΔCDS Spread, is the unexpected change in CDS spreads over an

eleven-day window [-5, +5] centered on the 10-Q/K disclosure date. The primary test variable in

the regression is uncertain tone (UNCTONE), which reflects the risk qualitatively disclosed in 10-

Q/K filling. As noted previously, we measure uncertain disclosure tone using uncertain words

(e.g., approximate, uncertain, indefinite, possible) in 10-Q/K filings (see LM 2011). Firm-specific

controls include changes in firm size (ΔSize), leverage (ΔLev), and return on assets (ΔROA),

measured relative to their value in the previous quarter.6 We also include changes in market-wide

variables that occur during the [-5, +5] event window using variables described in Section III.A:

risk-free rate (ΔRf), Implied Volatility of S&P 500 Index options (ΔVIX), Term Spread (ΔTS), and

Default Spread (ΔDS). Our regression model includes terms for the interaction of time and industry

fixed effects, and includes clustered standard errors at the firm level. All variable definitions also

available in Appendix.

IV. Results

Summary Statistics

Table 1 presents descriptive statistics for all variables employed in our analysis. In Panel

A, we report descriptive statistics for levels variables. We note that the mean (median) CDS spread

on the 10-Q/K filing date is 197 (87) basis points. Firms covered in our sample are both large

6 In untabulated analysis, we substitute Book-to-Market (B/M) ratio in place of Leverage. All results hold for this

alternate specification. We do not include both B/M and Lev in the same regression because of multicollinearity

issues.


13

(average market value of $20.7B) and highly skewed (median market value is roughly 1/3 of the

mean, $7.7B). Given skewness, we use the natural logarithm of firm size in all empirical

specifications, including in measures of firm size changes. Firm-specific controls include leverage

(leverage), and return on assets (ROA) which reflects firm profitability. The mean (median)

leverage in our sample is 0.27 (0.23). The mean quarterly ROA is 1.02%, with somewhat large

variation (25th percentile value of 0.35% and a 75th percentile value of 1.95%). Equity market

responses are captured with cumulative event period equity return (EPER) and annualized realized

volatility of the firm’s daily equity returns for the past year (RVOL). EPER has a mean (median)

of 100.5 basis points (100.3) while RVOL has a mean (median) value of 0.345 (0.287). Macro

controls include measures of general market conditions including the volatility index (VIX) of S&P

500 options and the three-month risk-free rate (Rf). We observe that S&P 500 options have a mean

(median) of 20.02 basis points (17.58) and that the mean (median) three-month risk-free rate (Rf)

is 1.42 (0.31) percent. Lastly, we report also macro-level controls relating to debt markets

specifically such as the term spread (TS) and default spread (DS), which are defined respectively

as the ten-year U.S. treasury bond rate minus the risk-free rate, and Moody’s Baa Corporate bond

rate relative to the ten-year treasury bond rate. We observe that TS has a mean (median) of 2.05

(2.23) percent and that DS has a mean (median) of 2.67 (2.66) percent.

In Panel B, we report descriptive statistics for changes variables. Our primary dependent

variable, ΔCDS Spread, measures the unexpected change in CDS spreads around the disclosure

event after subtracting median changes for all firms with the same credit rating. The mean change

in CDS spreads in our sample is around one basis point, but varies significantly in our sample (25th

and 75th percentile values of approximately -3 and +3 basis points respectively). Our main

independent variable is uncertain disclosure tone (UNCTONE). The mean value of the disclosure


14

uncertainty tone is 1.38%. The 25th and 75th percentile values for UNCTONE are 1.14% and 1.59%

respectively. The summary statistics for UNCTONE suggest that it is not highly skewed and do

not suffer from the presence of outliers.

In Panel C, we report a univariate correlation matrix of our primary research variables. The

correlation coefficient between our primary dependent variable, ΔCDS Spread, and our primary

independent variable, UNCTONE, is 0.011. The correlation between ΔCDS Spread and equity

returns (EPER) -0.158, consistent with CDS markets sharing some information linkages with

equity markets. The correlation between ΔCDS Spread and change is firm size (ΔSize) is -0.154.

UNCTONE demonstrates no correlation of 0.10 or greater in absolute terms with other variables.

Hypothesis Tests

Hypothesis 1 – Credit Spread Response to Uncertain Tone

We first examine the relationship between uncertain disclosure tone in 10-Q/K filings and

the change in five-year CDS spreads measured in short windows around disclosure event dates.

We estimate the panel regression in equation (1). The dependent variable ΔCDS Spread, is the

change in CDS spreads over an eleven-day window [-5, +5] centered on the 10-Q/K disclosure

date. The change in firm-specific variables are relative to the prior quarter. Market-level change

variables reflect the change in market condition around the disclosure event date. In all our

specifications, we include industry-quarter fixed effects and the clustered standard errors at the

firm level. We present the results of the panel regression in Table 2.

[ INSERT TABLE 2 HERE ]

The specification in column (A) presents univariate results. The coefficient on our measure

of uncertain tone (UNCTONE) is positive and statistically significant, indicating that the CDS

spreads increase for firms that increase uncertain disclosure tone around the disclosure event. The


15

specification in column (B) includes all firm-level controls. The coefficient on the uncertain tone

measure is greater in magnitude compared to column (A) and continues to be positive and

significant. Examining firm-level controls, CDS spreads increase around the disclosure event date

for firms that have recently experienced a decline in firm size.

The specification in column (C) includes also market-level controls. We control for the

portion of CDS spread changes that are implied by equity market responses with cumulative daily

equity returns (EPER) of the firm during the [-5, +5] period and the realized volatility of the firm’s

daily equity return (Rvol) over the event day [-252, -6] relative to the disclosure date. We note

that the coefficient on the event period cumulative return (EPER) indicates that firms with lower

cumulative daily returns have a larger change in CDS spreads. These results are consistent with

the intuition that lower equity returns and drops in firm equity relate to elevated default risk as

reflected by an increase in CDS spreads. The market level controls do not affect the magnitude or

significance of the uncertain tone measure but result in some improvement in R2. Our measure of

uncertain tone continues to be positive and statistically significant. The magnitude and significance

of the firm controls also continue to be similar to the specification in column (B). The coefficient

on the change in default spread is positive and significant; suggesting that firm CDS spreads

increase in conjunction with aggregate default risk.

Overall, the results in columns (A) to (C) support our first hypothesis. We conclude that

firms with relatively higher uncertain disclosure tone have a higher change in five-year CDS

spreads. This result is also economically significant. The coefficient in column (B) indicates that

an increase in uncertain disclosure tone from 25th to 75th percentile (an increase of 0.45%) is

associated with a 1.91 basis points higher change in our median-adjusted CDS spreads around the

event window.


16

Hypothesis 2 - Term Structure of CDS Spreads

We next investigate hypothesis two, which predicts that uncertain disclosure tone affects

the credit spreads of short-term maturity contracts more than long-term maturity contracts

following Duffie and Lando (2001). We use information from the term structure of CDS spreads,

examining changes in 1-, 3-, 7-, and 10-year CDS spreads to estimate equation (1). We include

both firm-specific and market-level controls for all maturities, including controls for equity market

returns (EPER) and volatility (Rvol). Table 3 presents the results of our analysis.

Columns (A) to (D) of Table 3 present the regressions results for 1-, 3-, 7- and 10-year

maturity contracts respectively. Descriptively, we observe a positive and significant coefficient on

uncertain disclosure tone (UNCTONE) in all four columns, indicating that firms with higher

uncertain tone experience relatively greater changes in spreads across all maturities. We also

observe that the coefficient on uncertain disclosure tone decreases monotonically from columns

one to four, as the maturity increases, as does the statistical significance of these coefficients. As

uncertain tone measure moves from 25th to 75th percentile, the change in CDS spread for 1-year

maturity contracts averages 3.9 basis points, but averages 1.6 basis points for 10-year maturity

contracts. These observations are consistent with Duffie and Lando (2001) and with H2.

[INSERT TABLE 3 HERE]

In column (E), we formally test Hypothesis 2. We estimate the relationship between tone

uncertainty and the change in the slope of the term structure measured using the difference between

10-year and 1-year credit spreads. As noted previously, the slope change is computed after

subtracting the median slope change for the CDS contracts with the same credit rating as the

disclosing firm. Consistent with the monotonic decline in the uncertainty coefficient with maturity,

we find a statistically significant change in the slope negatively relates to the tone uncertainty. We


17

conclude that hypothesis two is supported, reflecting greater reactions to uncertainty signals for

contracts with relatively shorter maturity.

Hypothesis 3 - Effect of Volatility and Leverage

Lastly, we investigate hypothesis three. Duffie and Lando (2001) also show that

accounting precision, which can be thought of as assessment uncertainty, should affect the short-

term credit spreads of firms closer to insolvency more than those further from insolvency. This

implies that the impact of the uncertain disclosure tone on CDS spread changes should be larger

for firms with relatively greater volatility and greater leverage, because these firms are more to

likely to default, ceteris paribus. We test this hypothesis in two ways, reported in Table 4. We first

perform subsample regressions, splitting our sample at the median value of volatility using Rvol

(H3a, columns (A) and (B) in Table 4) and leverage (H3b, columns (D) and (E)), to ensure

balanced sample sizes. Subsample tests provide insight into the comparative weighting of

disclosure tone for firms with high versus low volatility and leverage. Second, we perform tests

on our entire sample, using an interaction term to formally derive a test statistic. In column (C),

which investigates volatility, we assign an indicator variable equal to 1 for observations assigned

to the high volatility subsample, 0 otherwise. In column (F), which investigates leverage, we

assign an indicator variable equal to 1 for observations in the high leverage subsample (0

otherwise).

[ INSERT TABLE 4 HERE ]

Table 4 presents results from our tests of H3a and H3b, examining the relationship between

uncertain disclosure tone and CDS spreads in the sample partitions. Consistent with our

hypotheses, we observe that uncertain linguistic tone is a statistically significant determinant of

CDS spreads for our high volatility subsample (column (A)) but not statistically significant in our


18

low volatility subsample (column (B)). Likewise, we observe that uncertain linguistic tone is also

significantly related to CDS spreads for our high leverage subsample (column (D)) but not in our

low leverage subsample (column (E)). In our formal tests (columns (C) and (G)), we observe that

both the subsample indicator variables and their interactions terms (interacted with UNCTONE)

are significantly related to CDS spreads, although the differential credit market response to

uncertain linguistic for highly leveraged firms is only marginally significant (p=0.093, two-

tailed).7 We conclude that H3a and H3b are supported, suggesting that credit markets

differentially respond to uncertain linguistic tone depending on cross-sectional differences in ex-

ante default risk, and do so incrementally to other known cross-sectional determinants of CDS

spreads (Ericsson et al. 2009).

5. Robustness & Alternative Explanations

In robustness analysis, we explore alternative explanations for the results we observe

broadly related to the possibility that our evidence reflects material non-textual information

released concurrently with 10-Q/K filings or other textual features that are not reflecting uncertain

tone per se.

Material Non-Textual Information

One potential explanation for the credit market response we observe is that firms may

disclose other financial results and/or announcements concurrently with (or in) 10-Q/K filings that

lead credit markets to update CDS spreads and, separately, also lead firm managers to employ

different tone. We construct two tests to better understand if our results reflect concurrent

information releases. First, to account for new financial information conveyed in the 10-Q/K report

7 In untabulated analysis, we also perform pooled regressions specified as in Column 3 (volatility) and Column 6

(leverage) but including also a control variable for Rvol, as employed in tests of our other hypotheses. Our results

are inferentially identical when including this additional control, and Rvol is not statistically significant in either

regression.


19

we add a control variable to our primary regression specification (1), reflecting earnings surprise

(Surprise). For our analysis, Surprise is computed by taking the difference between the actual

earnings per share and the median analyst estimate standardized by the price of the stock. Second,

we construct alternate samples excluding sources of material information outside of the 10-Q/K

that might impact CDS spreads (e.g., Shivakumar et al. 2011). We do this by constructing

subsamples of firm-quarters without 8-K and/or management guidance releases that occur during

the corresponding event window.


Table 5 presents the results of these tests. We observe in column (A) that our measure of

uncertain disclosure tone remains statistically significant and of similar magnitude after controlling

for earnings surprise. In columns (B) through (D), we observe also that uncertain tone remains

statistically significant in all limited concurrent release subsamples. We conclude that the

relationship between uncertain tone and CDS spreads is not driven by concurrent releases of other

financial information.

Other Textual Dimensions

Another potential explanation for our evidence is that credit investors are responding to

other textual dimensions of 10-Q/K filings that we may be attributing to our measure of uncertain

tone. Research observes that firms with unfavorable news issue more complex disclosures (Lo et

al. 2017; Li 2010; Bloomfield 2008), which are associated with higher search costs and hence

greater riskiness (Lehavy et al. 2011; Li 2008). Thus, differences in uncertain tone could plausibly

correspond with differences in readability and negative or positive tone. To control for the

possibility that readability, rather than uncertain tone, drives our results we re-run our primary

regression with various measures of readability/complexity as additional controls. We consider


20

four measures of readability/complexity advanced in Li (2008) and Loughran and McDonald

(2014), which include the log of file size, Kincaid index, Fog index and Coleman-Liau index. We

include each of these indices separately as an additional control together with all firm and market

controls. Table 6 presents our results of our regression analysis including these additional controls.

The relationship we observe between uncertain tone and CDS spreads remains statistically

significant in all four regressions.


To control for the possibility that negative tone drives our results we rerun our primary

analysis with two measures of negative tone (NEGTONE), taken from LM 2011, or from the

Harvard Inquirer Dictionary. We present the results from these regressions in columns (A) through

(C) of Table 7. We observe that uncertain tone remains statistically significant, suggesting that

our results exist incrementally to negative tone also conveyed. We also acknowledge that

Loughran and McDonald (2013) argue that negative and weak modal words may reflect similar

underlying ambiguity as the uncertain words. Following this logic, we build an alternative proxy

for uncertainty based on the difference between negative and positive tone in the 10-Q/K filings.

Because this measure reflects uncertainty less directly than our primary measure, built from a

dictionary explicitly created to reflect uncertainty, we expect that it should load significantly in

regression specifications without our primary measure of uncertain tone but should not load

significantly in a regression that includes UNCTONE. Results from regressions employing the

alternative measure of uncertainty are presented in columns (D) through (G) of Table 7. We

observe that NEGTONE and NET are positive and statistically significant in the regressions

without UNCTONE, but not significant in the regressions that includes UNCTONE. We conclude

that our results are robust to alternative measurement of uncertain tone.


21


Other Robustness

Lastly, we replicate our results under several alternative specifications to ensure that our

results are not somehow driven by idiosyncrasies in our research design. We first rerun our

baseline regression with varying lengths of event window centered around the 10-Q/K disclosure

filing date such as (i) three-day window ([-1, +1]); (ii) seven-day window ([-3, +3]); (iii) twenty-

one-day window ([-10, +10]); and (iv) six-day window ([0, +5]). Table 8, columns 1 through 4

present results of these regressions. Our conclusions remain unchanged after these modifications.

Second, we run our analyses excluding financial firms, who tend to be more regulated, with

differing capital structures than non-financial firms, and who often act as the dealers and

counterparties in CDS contracts. Column (E) of Table 8 reports these results. We find that

excluding financial firms from our sample does not impact our inferences. Lastly, we re-run our

analyses with a subset of periods that omits the 2007-2009 financial crisis period, when all firms

moved closer to their default points, leading credit markets to react differently (Lok and

Richardson 2011). We remove the crisis period from December 2007 to June 2009 from our sample

data and re-run our primary regression. Column (F) of Table 8 reports the results when we exclude

the financial crisis period. The coefficient of interest is positive and statistically significant which

is very similar to column (C) of Table 2.


6. Discussion & Conclusion

In this study we report results suggesting that credit markets respond to uncertain linguistic

tone in 10-Q/K filings, and that they do so incrementally to other known determinants including

equity market responses to the same information. We predict and find that CDS spreads increase


22

in uncertain tone, and additionally observe that CDS spread changes are greater for firms theorized

to have the greatest default risk assessment uncertainty (Duffie and Lando 2001). Specifically, we

find that the CDS spreads of short-term maturities respond more than those of long-term maturities,

and we also find that the CDS spreads of firms with greater volatility and greater leverage respond

more than those with lower volatility and lower leverage. In robustness analysis, we present

additional evidence suggesting that our results reflect uncertain tone rather than concurrently

released financial information, other textual dimensions, or design idiosyncrasies. Taken together,

these results underscore the importance of the linguistic tone of accounting disclosures, which

influences investors’ assessments about the firms’ future credit risks.

Our results have implications for managers as well as regulators. Because investors pay

close attention to not only the quantitative information but also to how managers express their

views in the disclosures, managers should be extremely careful in articulating firm-related

information. The managers can significantly reduce valuation risks by simply choosing the right

words. From a regulatory perspective, our results show the incremental value relevance of required

qualitative information. Regulators can encourage firms to disclose more nonfinancial information

that can improve the price discovery mechanism in the market. To the extent that CDS spreads

reflect default risk, our results additionally suggest that managers are sensitive to and implicitly

convey default uncertainties, even if they are unwilling to explicitly acknowledge going-concern

risks (Mayew et al. 2014, OSC 2010). Lastly, our study also builds on textual analysis research by

documenting credit market responses to uncertain tone, and demonstrating their incremental

response to other textual dimensions.

Our results raise several additional questions which we leave for future research. Foremost,

future research could investigate further whether uncertain tone was related to greater default


23

frequencies, which would be useful in understanding if markets over/under react to uncertain tone.

Relatedly, future research can also explore distinctions between expected tone, which can be

thought of as increasing transparency in the Duffie and Lando (2001) model, versus abnormal

tone, which can be thought of as attempts by managers to persuade rather than inform and is thus

transparency decreasing. Lastly, future research could investigate further exactly how and why

credit market responses differ from equity markets.


24

References

Allee, K., and M. DeAngelis. 2015. The structure of voluntary disclosure narratives: Evidence

from tone dispersion. Journal of Accounting Research 53, 241-274.

Altman, E. 1968. Financial ratios, discriminant analysis and the prediction of corporate

bankruptcy. Journal of Finance 23, 589-609.

Altman, E. I., & Saunders, A. 1997. Credit risk measurement: Developments over the last 20 years.

Journal of banking & finance, 21(11-12): 1721-1742.

Antweiler, W., & Frank, M. Z. 2004. Is all that talk just noise? The information content of internet

stock message boards. Journal of Finance 59, 1259–1294.

Arora, N., S. Richardson, and I. Tuna. 2014. Asset reliability and security prices: Evidence from

credit markets. Review of Accounting Studies 19, 363-395.

Augustin, P., M. Subrahmanyam, D. Tang, and S. Wang. 2014. Credit default swaps: A survey,

Foundations and Trends in Finance 9, 1-196.

Augustin, P., M. Subrahmanyam, D. Tang, and S. Wang. 2016. Credit default swaps: Past, present,

and future. Annual Review of Financial Economics 8, 175-196.

Ball, R., Robin, A., & Sadka, G. 2008. Is financial reporting shaped by equity markets or by debt

markets? An international study of timeliness and conservatism. Review of accounting

studies, 13(2-3): 168-205.

Bharath, S. T., Sunder, J., & Sunder, S. V. 2008. Accounting quality and debt contracting. The

Accounting Review, 83(1), 1-28.

Batta, G. 2011. The direct relevance of accounting information for credit default swap pricing.

Journal of Business Finance and Accounting 38, 1096-1122.

Batta, G., J. Qiu, and F. Yu. 2016. Credit derivatives and analyst behavior. The Accounting Review

91, 1315-1343.

Bhat, G., J.L. Callen, and D. Segal. 2014. Credit risk and IFRS: The case of credit default swaps.

Journal of Accounting, Auditing & Finance 29, 129-162.

Bank for International Settlements (BIS). 2018. BIS Quarterly Review: International banking

and financial market developments. September 2018. Accessed December 2018.

Available at: https://www.bis.org/publ/qtrpdf/r_qt1809.pdf

Callen, J.L., J. Livnat, and D. Segal. 2009. The impact of earnings on the pricing of credit default

swaps. The Accounting Review 84, 1363-1394.

Chen, L., Lesmond, D. A., & Wei, J. 2007. Corporate yield spreads and bond liquidity. The Journal

of Finance, 62(1): 119-149.

Chiu, T. T., Guan, Y., & Kim, J. B. 2018. The Effect of Risk Factor Disclosures on the Pricing of

Credit Default Swaps. Contemporary Accounting Research.


https://www.bis.org/publ/qtrpdf/r_qt1809.pdf

25

Collin-Dufresne, P., and B. Goldstein. 2001. Do credit spreads reflect stationary leverage ratios?

Journal of Finance 56, 1929-1957.

Collin-Dufresne, P., R. Goldstein, and J. Martin. 2001. The determinants of credit spread changes,

Journal of Finance 56, 2177-2207.

Correia, M., S. Richardson, and I. Tuna. 2012. Value investing in credit markets. Review of

Accounting Studies 17, 572-609.

Das, S.R., P. Hanouna, and A. Sarin. 2009. Accounting-based versus market-based cross-sectional

models of CDS spreads. Journal of Banking and Finance 33, 719-730.

Davis, A., J. Piger, and L. Sedor. 2012. Beyond the numbers: Measuring the information content

of earnings press release language. Contemporary Accounting Research 29, 845-868.

Davis, A., and I. Tama-Sweet. 2012. Managers’ use of language across alternative disclosure

outlets: earnings press release language versus MD&A. Contemporary Accounting

Research 29, 804-837.

Demerjian, P. R. 2017. Uncertainty and debt covenants. Review of Accounting Studies, 22(3):

1156-1197.

Demerjian, P. R., & Owens, E. L. 2016. Measuring the probability of financial covenant violation

in private debt contracts. Journal of Accounting and Economics, 61(2-3): 433-447.

Demers, E., and C. Vega. 2011. Linguistic Tone in Earnings Announcements: News Or Noise?

Working paper, INSEAD.

Doran, J., D. Peterson, and S. Price. 2012. Earnings conference call content and stock price: The

case of REITs. The Journal of Real Estate Finance and Economics 45, 402-434.

Doblas-Madrid, A., & Minetti, R. 2013. Sharing information in the credit market: Contract-level

evidence from US firms. Journal of financial Economics, 109(1): 198-223.

Duffee, G. 1999. Estimating the price of default risk. Review of Financial Studies 12, 197–226.

Duffie, D. and D. Lando. 2001. Term structures of credit spreads with incomplete accounting

information. Econometrica 69, 633-664.

Durnev, A., Mangen, C. 2011. The Real Effects of Disclosure Tone: Evidence from Restatements.

Working Paper.

Easton, P. D., Monahan, S. J., & Vasvari, F. P. 2009. Initial evidence on the role of accounting

earnings in the bond market. Journal of Accounting Research, 47(3): 721-766.

Elkamhi, R., K. Jacobs, H. Langlois, and C. Ornthanalai. 2012. Accounting information releases

and CDS spreads. Working Paper, University of Toronto.

Elton, E. J., Gruber, M. J., Agrawal, D. and Mann, C. 2001. Explaining the Rate Spread on

Corporate Bonds. Journal of Finance, 56: 247-277.


26

Engelberg, Joseph. 2008. Costly Information Processing: Evidence from Earnings

Announcements. January 18, 2008. AFA 2009 San Francisco Meetings Paper. Available

at SSRN: https://ssrn.com/abstract=1107998

Ericsson, J., K. Jacobs, and R. Oviedo. 2009. The determinants of credit default swap premia,

Journal of Financial and Quantitative Analysis 44, 109-132.

Ertugrul, M., Lei, J., Qiu, J., & Wan, C. 2017. Annual report readability, tone ambiguity, and the

cost of borrowing. Journal of Financial and Quantitative Analysis, 52(2): 811-836.

Feldman, R., S. Govindaraj, J. Livnat, and B. Segal. 2010. Management’s tone change, post

earnings announcement drift and accruals. Review of Accounting Studies 15, 915-953.

Givoly, D., Hayn, C., & Katz, S. 2017. The changing relevance of accounting information to debt

holders over time. Review of Accounting Studies, 22(1): 64-108.

Griffin, P.A. 2014. The market for credit default swaps: new insights into investors' use of

accounting information? Accounting & Finance 54, 847-883.

Griffin, P.A., H.A. Hong, and J.B. Kim. 2016. Price discovery in the CDS market: the

informational role of equity short interest. Review of Accounting Studies 21, 1116-1148.

Guay, W., D. Samuels, and D. Taylor. 2016. Guiding through the Fog: Financial statement

complexity and voluntary disclosure. Journal of Accounting and Economics 62, 234-269.

Hanley, K.W. and G. Hoberg. 2012. Litigation risk, strategic disclosure and the underpricing of

initial public offerings. Journal of Financial Economics 103, 235-254.

Henry, E. 2008. Are investors influenced by how earnings press releases are written? Journal of

Business Communication 45, 363–407.

Henry, E. and A. Leone. 2015. Measuring qualitative information in capital markets research:

comparison of alternative methodologies to measure disclosure tone. The Accounting

Review 91, 153-178.

Huang, X., S. Teoh, and Y. Zhang. 2014. Tone management. The Accounting Review 89, 1083-

1113.

Hilscher, J., J.M. Pollet, and M. Wilson. 2015. Are Credit Default Swaps a Sideshow? Evidence

That Information Flows from Equity to CDS Markets. Journal of Financial and

Quantitative Analysis 50(3), 543-567.

Jacobson, T., Lindé, J., & Roszbach, K. 2006. Internal ratings systems, implied credit risk and the

consistency of banks’ risk classification policies. Journal of Banking & Finance, 30(7):

1899-1926.

Jung, B., Soderstrom, N., & Yang, Y. S. 2013. Earnings smoothing activities of firms to manage

credit ratings. Contemporary Accounting Research, 30(2): 645-676.

Kearney, C. and S. Liu. 2014. Textual sentiment in finance: A survey of methods and models.

International Review of Financial Analysis 33, 171-185.


27

Kim, S., P. Kraft, and S. Ryan. 2013. Financial statements comparability and credit risk. Review

of Accounting Studies 18, 783-823.

Kothari, S., X. Li, and J. Short. 2009. The effect of disclosures by management, analysts, and

business press on cost of capital, return volatility, and analyst forecasts: A study of using

content analysis. The Accounting Review 84, 1639-1670.

Lehavy, R., Li, F., & Merkley, K. 2011. The effect of annual report readability on analyst following

and the properties of their earnings forecasts. Accounting Review, 86, 1087–1115.

Li, F. 2008. Annual Report Readability, Current Earnings, and Earnings Persistence. Journal of

Accounting and Economics 45, 221–47.

Li, F. 2010a. The information content of forward looking statements in corporate filings—A naïve

Bayesian machine learning approach. Journal of Accounting Research 48, 1049-1102.

Li, F. 2010b. Survey of the Literature. Journal of accounting literature, 29: 143-165.

Li, F., Lundholm, R. and Minnis, M. 2013. A Measure of Competition Based on 10‐K Filings.

Journal of Accounting Research 51, 399-436.

Li, J.Y. and D. Tang. 2016. The leverage externalities of credit default swaps. Journal of Financial

Economics 120, 491-513.

Lok, S., and S. Richardson. 2011. Credit markets and financial information. Review of Accounting

Studies 16, 487-500.

Longstaff, F. A. 2010. The subprime credit crisis and contagion in financial markets. Journal of

financial economics, 97(3): 436-450.

Longstaff, F.A., S. Mithal, and E. Neis. 2005. Corporate yield spreads: Default risk or liquidity?

New evidence from the credit default swap market. Journal of Finance 60, 2213-2253.

Loughran, T. and B. McDonald. 2011. When is a liability not a liability? Textual analysis,

dictionaries, and 10‐Ks. Journal of Finance 66, 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, 307-326.

Loughran, T. and B. McDonald. 2014. Measuring readability in financial disclosures. Journal of

Finance 69, 1643-1671.

Loughran, T. and B. McDonald, B. 2016. Textual analysis in accounting and finance: A survey.

Journal of Accounting Research 54, 1187-1230.

Mayew, W.J., M. Sethuraman, and M. Venkatachalam. 2014. MD&A Disclosure and the Firm's

Ability to Continue as a Going Concern. The Accounting Review 90, 1621-1651.

Merton, R. 1974. On the pricing of corporate debt: The risk structure of interest rates. The Journal

of Finance 29, 449-470.

Oehmke, M. and A. Zawadowski. 2017. The anatomy of the CDS market. Review of Financial

Studies 30, 80-119.


28

Ohlson, J.A. 1980. Financial Ratios and the Probabilistic Prediction of Bankruptcy. Journal of

Accounting Research 18, 109-131.

Ontario Securities Commission. 2010. OSC Staff Notice 52-719: Going Concern Disclosure

Review.

Price, S., J. Doran, D. Peterson, and B. Bliss. 2012. Earnings conference callas and stock returns:

The incremental informativeness of textual tone. Journal of Banking and Finance 36, 992-

1011.

Rogers, J. L., A. Van Buskirk, and S. Zechman. 2011. Disclosure tone and shareholder litigation.

The Accounting Review 86, 2155-2183.

Shivakumar, L., O. Urcan, F.P. Vasvari, and L. Zhang. 2011. The debt market relevance of

management earnings forecasts: Evidence from before and during the credit crisis. Review

of Accounting Studies 16, 464.

Tan, H.T., E. Ying Wang, and B. Zhou. 2014. When the use of positive language backfires: The

joint effect of tone, readability, and investor sophistication on earnings judgments. Journal

of Accounting Research 52, 273-302.

Tang, D. Y., F. Tian, and H. Yan. 2015. Internal Control Quality and Credit Default Swap Spreads.

Accounting Horizons 29, 603-629.

Tang, D.Y. and H. Yan. 2010. Market conditions, default risk and credit spreads. Journal of

Banking & Finance 34, 743-753.

Tang, T. T. 2009. Information asymmetry and firms’ credit market access: Evidence from Moody's

credit rating format refinement. Journal of Financial Economics, 93(2): 325-351.

Tetlock, P. C. 2007. Giving content to investor sentiment: The role of media in the stock market.

Journal of Finance 62, 1139–1168.

Yu, F. 2005. Accounting transparency and the term structure of credit spreads. Journal of

Financial Economics 75, 53–84.

Zhang, B.Y., H. Zhou, and H. Zhu. 2009. Explaining credit default swap spreads with the equity

volatility and jump risks of individual firms. Review of Financial Studies 22, 5099-5131.

Zhang, G., and S. Zhang. 2013. Information efficiency of the U.S. credit default swap market:

Evidence from earnings surprises. Journal of Financial Stability 9, 720-730.


29

APPENDIX A

Variable Description

Variable Name Description

CDS Spread CDS Spread obtained from WRDS-Markit on the event day (basis points)

UNCTONE

The number of Loughran-McDonald Financial-Uncertainty words in the

document divided by the total number of words in the document that occur

in the master dictionary (percentage)

Size Logarithm of the market value of the firm’s equity

Leverage (LEV) 𝐵𝑜𝑜𝑘 𝑉𝑎𝑙𝑢𝑒 𝑜𝑓 𝐷𝑒𝑏𝑡

𝑀𝑎𝑟𝑘𝑒𝑡 𝑉𝑎𝑙𝑢𝑒 𝑜𝑓𝐸𝑞𝑢𝑖𝑡𝑦 .

ROA Net Income (NI) plus income from Extraordinary items and discontinued

operations (XIDO) divided by the dollar amount of assets in the firm(AT)

Turnover Logarithm of daily turnover over the previous 12 months as calculated

from CRSP

Event period equity return (EPER) Cumulative daily returns of the firm during the event period obtained from

CRSP

Realized volatility (Rvol) Realized volatility (annualized) of the firm's

daily equity returns over the past year

VIX Volatility index obtained from Chicago Board Options Exchange (CBOE)

Risk-free rate (Rf) Three-month U.S. treasury bill rate obtained from Federal Reserve

Economic Data (FRED) (basis points)

Term spread (TS) Ten year U.S. treasury bond rate minus the three-month treasury bill rate

obtained from FRED (basis points)

Default Spread (DS) Moody's Baa corporate bond yield relative to yield on ten-year treasury

bond rate obtained from FRED (basis points)

Fog Index 0.4*(average number of words per sentence + percent of complex words)

Flesch Kincaid 0.39*(number of words/number of sentences) + 11.8*(number of

syllables/number of words) – 15.59

Coleman-Liau 0.0588*(average number of letters per 100 words) – 0.296*(average

number of sentences per 100 words) – 15.8


30

TABLE 1

Descriptive Statistics

Panel A: Distribution of Variables (levels)

Variable Name Mean STD P25 Median P75 N

CDS Spread (bps) 196.825 495.871 44.995 87.376 192.615 26,003

CDS Excess Spread (bps) 0.0033 0.0299 -0.0019 0.0000 0.0030 26003

UNCTONE (%) 1.378 0.349 1.139 1.358 1.587 26,002

Size 8.973 1.412 8.042 8.959 9.863 25,377

Leverage 0.2728 0.1928 0.1271 0.2335 0.3809 26003

ROA 0.010 0.039 0.004 0.011 0.0195 26,003

Turnover -4.8440 0.7023 -5.2581 -4.8533 -4.4352 24,980

EPER (bps) 100.47 12.46 97.16 100.29 103.30 25,277

Rvol 0.345 0.221 0.209 0.287 0.403 24,984

VIX 20.021 9.368 13.690 17.580 22.810 25,935

Risk free rate (Rf) (bps) 142.338 171.769 9.000 31.000 247.000 26,001

Term Spread (TS) (bps) 205.501 111.692 149.000 223.000 283.000 26,001

Default Spread (DS) (bps) 267.464 85.154 205.000 266.000 306.000 26,001

Panel B: Distribution of Variables (changes)

Variable Name Mean STD P25 Median P75 N

∆CDS Spread (bps) 0.466 109.866 -3.229 0.000 3.236 25,742

∆Size 0.010 0.292 -0.070 0.026 0.113 24,632

∆Leverage -0.000 0.065 -0.018 -0.001 0.015 25297

∆ROA 0.000 0.048 -0.005 0.000 0.005 25,297

∆Turnover 0.012 0.175 -0.062 0.011 0.086 24236

∆VIX 0.250 3.700 -1.870 -0.080 1.920 25,935

∆Rf (bps) -2.020 16.288 -3.000 16.288 3.000 25,935

∆TS (bps) -0.756 21.445 -13.000 -2.000 11.000 25,935

∆DS (bps) 1.357 12.933 -5.000 2.000 8.000 25,935


31

Panel C: Correlation Table

∆CDS

Spread UNCTONE ∆Size ∆Leverage ∆ROA EPER Rvol ∆VIX ∆Rf ∆TS

UNCTONE 0.011 -

∆Size -0.154 0.014 -

∆Leverage 0.053 0.012 -0.566 -

∆ROA 0.001 0.003 0.039 -0.047 -

EPER -0.158 -0.003 0.131 -0.209 0.013 -

Rvol -0.008 0.080 0.025 -0.053 0.013 0.087 -

∆VIX 0.055 0.015 -0.095 0.077 -0.004 -0.161 -0.058 -

∆Rf -0.034 -0.024 0.132 -0.102 0.018 0.073 -0.044 -0.201 -

∆TS -0.001 -0.013 -0.019 0.012 -0.003 0.018 0.0401 -0.176 -0.500 -

∆DS 0.072 0.005 -0.144 0.113 -0.001 -0.071 -0.141 0.324 -0.219 -0.278

This table reports the descriptive statistics of all the important variables used in regressions. CDS Spread is the five-year CDS spread on the event day. UNCTONE

is the proportion of uncertainty words to the total number of words in the document as measured by Loughran and McDonald (2011). Size is the logarithm of the

market capitalization of the firm. Leverage is the book value of short-term and long-term debt divided by the Market Value of the firm's equity. ROA is the return

on assets. Turnover is the logarithm of daily turnover over the previous 12 months as calculated from CRSP. Event-period equity return is the cumulative daily

equity returns calculated over the event period. Rvol is the realized volatility of the firm's daily equity returns over the past year. VIX is the volatility index obtained

from Chicago Board Options Exchange (CBOE). Risk free rate is the three-month U.S. Treasury bill rate. Term Spread is the difference between ten-year U.S

Treasury bond rate and the three-month U.S. Treasury bill rate. Default Spread is the Moody's corporate bond yield relative to the yield on ten-year Treasury bond

rate. All the firm specific change variables represent the change in value from the previous quarter. ∆CDS Spread, ∆VIX, ∆Rf, ∆TS, and ∆DS represent the change

in value over [-5, +5] event period. ∆CDS Spread is the change in five-year maturity CDS spreads over the event period relative to the median change in CDS

spreads in the same credit rating group across the same period.


32

TABLE 2:

The Effect of Uncertain Tone on the Change in CDS Spreads

DV: Change in CDS Spreadsi,[-5,+5]

A B (H1)

C (H1)

UNCTONEi,t 3.132**

4.222***

4.255***

(0.023) (0.007) (0.006)

∆Sizei,t -67.835*** -67.650***

(0.009) (0.010)

∆Leverage,t -150.583

-148.660

(0.152) (0.158)

∆ROAi,t 9.897

9.702

(0.707) (0.712)

EPERi,t -1.134*

(0.084)

Rvoli,t -3.310

(0.870)

∆VIXt 0.589*

(0.096)

∆Rft 0.022

(0.840)

∆TSt 0.036

(0.699)

∆DSt 0.655***

(0.000)

Ind x Qtr FE Yes Yes Yes

Adj R2 0.009 0.050 0.053

N 25,243 24,458 24,458

This table reports the effect of uncertain tone in 10-Q/K statements on the changes in CDS spreads around the event

window [-5, +5] days of the disclosure. The dependent variable is the change in five-year maturity CDS spreads from

a week before the disclosure to a week after the disclosure [-5, +5] relative to the median change in CDS spreads in

the same credit rating group across the same time period. The independent variable of interest is UNCTONE the

uncertainty word proportion as defined in the Appendix. The other independent variables consist of firm controls:

∆Size is the change in size; ∆Leverage is the change in leverage; ∆ROA is the change in return on assets (ROA); EPER

is the event period return; realized volatility (Rvol), and marketwide controls: change in VIX; change in risk-free rate

(Rf); change in term spread (∆TS); change in default spread (∆DS). All the firm specific change variables represent

the change in value from the previous quarter. ∆CDS Spread, ∆Rf, ∆TS, and ∆DS represent the change in value over

[-5, +5] event period. Standard errors clustered at firm level and p-values reported below coefficient estimates.

Industry X quarter fixed effects included in all regressions. Sample period is from 2001 to 2016. Statistical significance

levels of 1%, 5% and 10% are indicated by ***, **, and * respectively.


33

TABLE 3:

Term Structure of CDS Spreads: Uncertain Tone and the Change in CDS Spreads


1 Yr

(A)

3 Yr

(B)

7 Yr

(C)

10 Yr

(D)

10 Yr-1 Yr

(E)

UNCTONEi,t 7.685***

5.605***

3.641**

3.293**

-4.129**

(0.004) (0.001) (0.016) (0.026) (0.049)

∆Sizei,t -108.210***

-73.839**

-51.828**

-49.525**

56.875**

(0.010) (0.012) (0.022) (0.018) (0.019)

∆Leverage,t -395.279*

-208.259

-119.868

-117.736

276.753*

(0.079) (0.174) (0.254) (0.154) (0.077)

∆ROAi,t -33.554

-17.716

-6.083

-12.075

17.447

(0.442) (0.544) (0.789) (0.569) (0.552)

EPERi,t -3.678***

-2.418***

-2.374***

-2.221***

1.456***

(0.001) (0.000) (0.000) (0.001) (0.002)

Rvoli,t -17.884

-12.737

1.691

7.976

19.823

(0.630) (0.594) (0.928) (0.640) (0.457)

∆VIXt -0.487

0.010

0.293

0.271

0.709***

(0.305) (0.977) (0.348) (0.423) (0.005)

∆Rft -0.014

0.087

0.169

0.075

0.084

(0.943) (0.483) (0.122) (0.495) (0.489)

∆TSt 0.100

0.114

0.169*

0.121

-0.003

(0.589) (0.319) (0.095) (0.221) (0.981)

∆DSt 0.549**

0.593***

0.650***

0.644***

0.149

(0.044) (0.001) (0.000) (0.000) (0.362)

Ind x Qtr FE Yes Yes Yes Yes Yes

Adj R2 0.069 0.071 0.081 0.080 0.033

N 22,203 23,327 23,889 22,558 21,579

This table reports the effect of uncertainty tone in 10-Q/K statements on the changes in CDS spreads around the event

window [-5, +5] days of the disclosure. The dependent variable is the change in CDS spreads from a week before the

disclosure to a week after the disclosure [-5, +5] relative to the median change in CDS spreads in the same credit

rating group across the same time period. The other independent variables are the same as in Table 2. All the firm

specific change variables represent the change in value from the previous quarter. ∆CDS Spread, ∆VIX, ∆Rf, ∆TS, and

∆DS represent the change in value over [-5, +5] event period. All columns have firm and market controls. We present

the results for various maturity contracts (1, 3, 7 and 10 year). 10 Yr – 1 Yr represent the change in 10-year CDS

spreads minus the change in 1-year CDS spreads in excess of the median change in the same rating group. Standard

errors clustered at firm level and p-values reported below coefficient estimates. Industry X quarter fixed effects

included in all regressions. Sample period is from 2001 to 2016. Statistical significance levels of 1%, 5% and 10% are

indicated by ***, **, and * respectively.


34

TABLE 4:

Default Risk and CDS Spreads: Uncertain Tone and the Change in CDS Spreads

Volatility Partition Leverage Partition

High Vol

(A)

Low Vol

(B)

Pooled

(C)

High Lev

(D)

Low Lev

(E)

Pooled

(F)

UNCTONEi,t 7.795***

(0.004)

-0.233

(0.476)

1.421*

(0.061)

1.767*

(0.051)

11.685*

(0.054)

1.750**

(0.017)

VOL_DUMi,t -9.050***

(0.009)

UNCTONE *

VOL_DUM 5.489**

(0.026)

LEV_DUMi,t

-8.268**

(0.020)

UNCTONE *

LEV_DUM 4.910*

(0.069)

∆Sizei,t -99.433***

(0.009)

-0.325

(0.859)

-67.683***

(0.010)

-24.588**

(0.020)

-93.750**

(0.040)

-67.687***

(0.010)

∆Leveragei,t -213.400

(0.139)

4.006

(0.419)

-148.520

(0.155)

0.449

(0.982)

-262.505

(0.197)

-147.774

(0.157)

∆M/Bi,t 18.852

(0.568)

-1.469

(0.790)

9.559

(0.718)

1.356

(0.853)

3.877

(0.922)

9.601

(0.717)

EPERi,t -1.188*

(0.092)

-0.288***

(0.000)

-1.138*

(0.087)

-1.205***

(0.001)

-1.269

(0.109)

-1.139*

(0.087)

∆VIXt 0.855*

(0.061)

0.183***

(0.004)

0.589*

(0.098)

0.029

(0.875)

2.198**

(0.013)

0.590*

(0.097)

∆Rft -0.047

(0.783)

0.038

(0.115)

0.023

(0.830)

0.081

(0.148)

-0.205

(0.611)

0.021

(0.849)

∆TSt 0.030

(0.839)

0.002

(0.857)

0.037

(0.694)

0.078*

(0.088)

-0.104

(0.771)

0.037

(0.692)

∆DSt 0.845***

(0.000)

0.199***

(0.000)

0.650***

(0.000)

0.291***

(0.000)

1.746***

(0.004)

0.655***

(0.000)

Industry x Qtr FE Yes Yes Yes Yes Yes Yes

Adj R2 0.062 0.035 0.053 0.071 0.064 0.053

N 12,148 12,310 24,458 18,680 5,778 24,458

This table reports the effect of uncertain tone on 10-Q and 10-K statements on the changes in CDS spreads around the

event window [-5, +5] days of the disclosure. The dependent variable is ∆CDS the change in five-year maturity CDS

spreads from a week before the disclosure to a week after the disclosure [-5, +5] relative to the median change in CDS

spreads in the same credit rating group across the same time period. The independent variable of interest is UNCTONE

the uncertainty word proportion as defined in the Appendix. The other independent variables are the same as in Table

2. Standard errors clustered at firm level and p-values reported below coefficient estimates. Industry X quarter fixed

effects included in all regressions. Sample period is from 2001 to 2016. Statistical significance levels of 1%, 5% and

10% are indicated by ***, **, and * respectively. High Vol indicates the part of sample with realized volatility (Rvol)

higher than the median and Low Vol indicates the sample with Rvol lower than the median. The columns titled Pooled

include a dummy variable (vol_Dum) which is 1 if Rvol is above the median and 0 otherwise.


35

TABLE 5:

Earning Surprise and Special Firm-Specific Events


Earning

Surprise

(A)

Excluding 8-K

Statements

(B)

Excluding Management

Guidance (MG)

(C)

Excluding 8k

and MG

(D)

UNCTONEi,t 3.859***

2.827**

4.819***

3.323** (0.007) (0.034) (0.004) (0.022)

Surprisei,t -1.528

(0.963)

All Controls Yes

Yes

Yes

Yes

Ind x Qtr FE Yes Yes Yes Yes

Adj R2 0.063 0.075 0.054 0.032

N 23,086 15,380 21,439 13,501

This table reports the effect of uncertain tone on 10-Q/K statements on the changes in CDS spreads around the event

window [-5, +5] days of the disclosure. The dependent variable is ∆CDS the change in five-year maturity CDS spreads

from a week before the disclosure to a week after the disclosure [-5, +5] relative to the median change in CDS spreads

in the same credit rating group across the same time period. The independent variable of interest is the uncertain tone

as defined in the Appendix. Surprise is Earnings Surprise defined as the actual earnings minus the median analyst

estimate standardized by price of the stock. The other independent variables are the same as in Table 2. All the firm

specific change variables represent the change in value from the previous quarter. ∆CDS Spread, ∆VIX, ∆Rf, ∆TS, and

∆DS represent the change in value over the event period. Standard errors clustered at firm level and p-values reported

below coefficient estimates. Industry X quarter fixed effects included in all regressions. Sample period is from 2001

to 2016. Statistical significance levels of 1%, 5% and 10% are indicated by ***, **, and * respectively.


36

TABLE 6:

Report Readability, Uncertain Tone and CDS Spreads


ln(File Size)

(A)

Fog Index

(B) Flesch-Kincaid

(C)

Coleman-Liau

(D)

UNCTONEi,t 4.282*** 4.327*** 4.324***

3.837***

(0.008)

(0.006)

(0.006) (0.008)

Readabilityi,t -0.003 0.425 0.483

0.207

(0.996)

(0.349)

(0.312) (0.733)

All Controls Yes

Yes

Yes Yes

Ind x Qtr FE Yes Yes Yes Yes

Adj R2 0.053 0.053 0.053 0.079

N 24,458 24,458 24,009 24,009

This table reports the effect of uncertainty tone on 10-Q/K statements on the changes in CDS spreads around the event

window [-5, +5] days of the disclosure. The dependent variable ∆CDS is the change in five-year maturity CDS spreads

from a week before the disclosure to a week after the disclosure [-5, +5] relative to the median change in CDS spreads

in the same credit rating group across the same time period. The independent variable of interest is UNCTONE,

uncertain tone defined in the Appendix. Readability measures include: log of 10-Q/K filings size (in megabytes), as

defined by Loughran and McDonald (2014); Fog Index is 0.4*(average number of words per sentence + percent of

complex words); Flesch-Kincaid Index is 0.39*(number of words/number of sentences) + 11.8*(number of

syllables/number of words) – 15.59; Coleman- Liau Index is 0.0588*(average number of letters per 100 words) –

0.296*(average number of sentences per 100 words) – 15.8. The other independent variables are the same as in Table

2. All the firm specific change variables represent the change in value from the previous quarter. ∆CDS Spread, ∆VIX,

∆Rf, ∆TS, and ∆DS represent the change in value over the event period. Standard errors clustered at firm level and p-

values reported below coefficient estimates. Industry X quarter fixed effects included in all regressions. Sample period

is from 2001 to 2016. Statistical significance levels of 1%, 5% and 10% are indicated by ***, **, and * respectively.


37

TABLE 7:

Other Word Lists and CDS Spreads


(A) (B) (C) (D) (E) (F) (G)

UNCTONEi,t

3.836** 4.276*** 4.847*** 3.839** (0.011) (0.008) (0.003) (0.010)

NEGTONE 1.388* 0.950

(0.147) (0.307)

WEAK MODAL 3.808 -1.390

(0.310) (0.756)

Net (NEG - POS) 1.355*

0.962

(0.148) (0.286)

Harvard IV NEG -3.814 (0.958)

All Controls Yes Yes Yes Yes Yes Yes Yes

Ind X Qtr FE Yes Yes Yes Yes Yes Yes Yes

Adj. R2 0.053 0.053 0.053 0.053 0.053 0.053 0.053

N 24,458 24,458 23,995 24,458 24,458 24,458 24,458

This table reports the effect of other word lists on the uncertain Tone and changes in CDS spreads relation around the

event window [-5, +5] days of the disclosure. The dependent variable ∆CDS is the change in five-year maturity CDS

spreads from a week before the disclosure to a week after the disclosure [-5, +5] relative to the median change in CDS

spreads in the same credit rating group across the same time period. The independent variable of interest is UNCTONE,

uncertain tone defined in the Appendix. Other word lists are from Loughran and McDonald (2011) and include:

NEGTONE, defined as percentage of negative words in 10-Q/K filings; WEAK MODAL, defined as percentage of

weak modal words in 10-Q/K filings. Net (NEG – POS) is the difference in percentage of negative and positive words

in 10-Q/K filings. Harvard IV NEG is the list of negative words classified by Harvard Inquirer Dictionary. The other

independent variables are the same as in Table 2. The standard errors are clustered at firm level and the p-values are

reported below the coefficient estimates. The sample period is from 2001 to 2016. Statistical significance levels of

1%, 5% and 10% are indicated by ***, **, and * respectively.


38

TABLE 8:

Robustness Checks - Alternate Event Window and Other Specifications

DV: Changes in CDS Spreads

Alternate Window specification

Non-

Financial

Non-

Crisis

[-1, +1]

[-3, +3]

[-10, +10]

[0, +5]

(A) (B) (C) (D) (E) (F)

UNCTONEi,t 1.618* 4.118** 4.860*** 3.574*** 3.552** 3.373**

(0.072) (0.015) (0.006) (0.001) (0.022) (0.026)

All Controls Yes

Yes

Yes

Yes

Yes

Yes

Ind x Qtr FE Yes Yes Yes Yes Yes Yes

Adj R2 0.022 0.078 0.098 0.061 0.046 0.038

N 24,466 24,027 23,900 24,458 20.293 21,501

This table reports the effect of uncertain tone on 10-Q/K on the changes in CDS spreads around the event window.

The dependent variable ∆CDS is change in five-year maturity CDS spreads during the event window relative to the

median change in CDS spreads in the same credit rating group across the same time period. The independent variable

of interest is UNCTONEi,t defined in the Appendix. The other independent variables are the same as in Table 2. The

standard errors are clustered at firm level and the p-values are reported below the coefficient estimates. The sample

period is from 2001 to 2016. Statistical significance levels of 1%, 5% and 10% are indicated by ***, **, and *

respectively.


are credit markets tone deaf? evidence from credit default ... · we also contribute to a growing...

Documents