Qualitative Corporate Disclosure and Credit Analysts’ Soft
Rating Adjustments
Zahn Bozanic
The Ohio State University
Pepa Kraft∗
New York University
October 30, 2014
Abstract
Several factors comprise the credit ratings analysts publicly report. While academics have
extensively studied the quantitative models credit analysts employ, far less is known about
the qualitative adjustments analysts make to their models’ outputs (“soft adjustments”). We
examine whether and how credit analysts employ credit-relevant qualitative disclosure in
making their credit risk assessments. Our results indicate that credit analysts impound the
information conveyed by firms’ qualitative disclosure in their publicly-reported ratings. Our
results further suggest that the “soft”, but not “hard”, adjustments are the mechanism by
which the information in qualitative disclosure is impounded into the rating. While disclosure opacity plays a relatively minor role in analysts’ use of soft information, the extent of
numerical disclosure that accompanies qualitative disclosure appears to convey precision and
credibility. We then analyze the informativeness of credit rating changes that involve soft
adjustments following Jorion et al. (2005). We find that credit rating changes that involve
soft adjustments are more informative than those that do not; however, the increase in informativeness diminishes after the repeal of the Regulation Fair Disclosure exemption for credit
rating agencies.
∗
We would like to thank participants of the NYU Stern brownbag for helpful comments and suggestions. We
are grateful to Patrick Pfeifer for research assistance. We appreciate funding from Fisher College of Business Ohio
State University and NYU Stern School of Business.
1
Introduction
Existing research on information intermediaries as users of qualitative disclosure has largely focused on equity analysts at the expense of credit analysts. As a result, our understanding of the
role qualitative disclosure plays in credit risk analysis is limited. The lack of research on the use
of qualitative disclosure by credit analysts is puzzling considering that credit analysts are major
users of corporate disclosures whose ratings have significant implications for firms’ access to and
cost of capital. Prior research on credit analysts focuses on their reliance on quantitative modeling
to produce initial credit ratings. In such models, amongst other determinants of credit quality, the
traditional Altman (1968) financial statement variables appear. More recently, research has extended the purview of how credit rating analysts arrive at their credit risk assessments by examining
analysts’ soft adjustments, that is, qualitative adjustments made to the rating recommendations
produced by their quantitative models (Kraft (2014)). While researchers can reproduce the quantitative models credit analysts rely on to make their initial credit assessments, less understood are
the inputs analysts rely on in making their soft rating assessments. In this study, we examine
the inputs that influence credit rating analysts’ soft rating assessments and the informativeness of
those assessments.
According to the credit ratings agencies, analysts consider several factors in making their soft
adjustments, such as caliber of management, financial transparency, and competitiveness (Standard
& Poor’s (2008); Moody’s (2007)). We focus on a single channel by which credit ratings analysts
inform their soft adjustments, qualitative disclosure, for at least two reasons. First, beyond the
traditional quantitative inputs of credit analysts’ models from financial statements, how managers
contextualize the quantitative information in the financial statements is a likely complementary
information channel. A large, recent body of research on qualitative disclosure lends credibility to
the conjecture that stakeholders beyond equity market investors and equity analysts plausibly use
this form of information.
Second, prior to Regulation FD (“Reg FD”), both equity and credit analysts operated under a
level playing field in that either party was privy to direct access to management and the attendant
1
private information conveyed by managers. However, in a post-Reg FD world, the level playing field
was altered by disallowing equity analysts’ access to managers, yet the access for credit analysts
remains unfettered. Given the lack of ratings timeliness and accuracy that arguably contributed to
the credit crisis as a result of the close relationship between credit analysts and management, this
relationship was questioned and the Dodd-Frank Wall Street Reform and Consumer Protection
Act (“the Dodd-Frank Act”) repealed the exemption Reg FD granted credit analysts. Hence,
credit analysts’ information supply diminished, arguably increasing the search costs of information
acquisition and the processing costs once found. As a result, credit analysts are likely to spend
more time and effort examining firm disclosures for information that cannot be captured by their
quantitative models alone.
We examine several risk-relevant proxies for qualitative information found in firms’ annual reports. While the majority of the proxies are consistent with prior literature, we augment these
proxies by the addition of a proxy likely to reflect the types of information in which credit, as
opposed to equity, analysts are interested. Namely, we first examine the provision for information on debt-related topics in the disclosure, such as liquidity, solvency, and covenants. Second,
consistent with prior studies (e.g., Jennings (1987); Pownall et al. (1993); Skinner (1994); Easton
et al. (2009)), we examine the provision for earnings-related topics, given that credit analysts are
likely to scrutinize earnings disclosures since such disclosures may signal the firm’s ability to service debt, especially during loss periods. Third, in addition to historical trends, credit analysts
are likely interested in managerial outlooks when forming ratings change and outlook decisions,
whether in the form upgrades, downgrades, or additions to watch lists. We therefore examine the
provision for forward-looking disclosure (Li (2010a); Bozanic et al. (2013)). Fourth, some qualitative disclosure may not explicitly mention debt or earnings topics or provide forward-looking
disclosures. However, the disclosure may more generally convey uncertainty. While uncertainty
is a concern for both equity and credit analysts, equity investors are likely to be relatively more
uncertainty-seeking for known risks than debt investors. Hence, we examine how firm disclosures
reflect uncertainty (Loughran and McDonald (2011)).
2
We first investigate whether actual, publicly-reported credit ratings reflect the information conveyed by qualitative disclosure. This step allows us to assess the information content of qualitative
disclosure by examining a standard credit rating agency output, the rating publicly issued by the
rater, which is the traditional metric consumed by investors as well as studied by academics. We
find that ratings tend to be lower when firms provide more disclosure on debt-related topics and
higher when firms provide more disclosure on earnings-related topics. These results are consistent
with managers of firms disclosing debt-related topics that appear to reflect adverse credit conditions and earning-related topics that reflect favorable information for debt investors. We also
find some evidence that ratings tend to be lower when firms provide more forward-looking and
uncertainty-related disclosure. However, given the high collinearity between these two proxies,
forward-looking and uncertainty-related disclosure appear to be substitutes for one another, which
is consistent with the findings in Bozanic et al. (2013).
While these result shed light on whether qualitative information is reflected in reported credit
ratings, they do not speak to the mechanism through which the information flows, that is, whether
the information conveyed by qualitative disclosure merely reflects that contained in credit analysts’
quantitative models or whether the information is uniquely impounded into their soft adjustments.
Therefore, we next investigate whether credit rating analysts employ the risk-relevant information
conveyed by qualitative disclosure in making both their soft and hard adjustments, respectively.
We find that credit rating analysts are more likely to make credit risk increasing soft adjustments when firms provide more disclosure on debt-related topics. When managers provide more
discussion on earnings-related topics, credit rating analysts are more likely to make credit risk
decreasing soft adjustments. We also find that both forward-looking and uncertain disclosure increase the likelihood that credit rating analysts make credit risk increasing soft adjustments. While
the results on the relation between credit rating analysts’ soft adjustments and qualitative disclosure largely mirror those related to the reported rating, the economic significance of the impact of
qualitative disclosure on soft adjustments is greater. Further, the results hold after controlling for
an extensive set of variables, such as size, leverage, profitability, cash flow, tangibility, scale of di-
3
versification, corporate governance, competition, return volatility, and financial reporting quality.
In stark contrast, we do not find that any of our proxies for qualitative disclosure are related to
hard rating adjustments, i.e., those arising strictly from analysts’ adjustments to the data underlying their quantitative modeling. Taken together, the results suggest that qualitative disclosure
is useful to credit rating analysts when making their risk assessments and soft adjustments are
the conduit through which qualitative information is impounded into the actual, publicly-reported
credit rating.
The results on the proxies for qualitative disclosure examined thus far focus on disclosure
content or “what firms say” leaving the style or “how firms say it” of the disclosure unexamined. Therefore, following prior literature (Li (2008); Lundholm et al. (2014)), we next examine
managerial disclosure style. Namely, we assess whether the level of obfuscation of qualitative disclosure, as well as the level of numerical intensity that complements the disclosure, impacts how
analysts impound qualitative disclosure in their soft rating adjustment assessments. We find that
credit rating analysts are more likely to make credit risk decreasing soft adjustments for opaque,
earnings-related qualitative disclosure. This finding is consistent with managers attempting to
obfuscate performance (Li (2008)). In terms of numerical disclosure, we find that credit rating
analysts are more likely to make credit risk increasing soft adjustments for debt-related qualitative
disclosure accompanied by high numerical intensity, which is consistent with numerical disclosure
adding precision to credit analysts’ beliefs regarding such topics. However, we find that credit
rating analysts are more likely to make credit risk decreasing soft adjustments for forward-looking
disclosure accompanied by high numerical intensity. This finding is consistent with numerical disclosure adding precision and credibility to forward-looking disclosures provided by management
(Hutton et al. (2003)).
Lastly, while we have established that credit ratings analysts appear to use qualitative information to make their soft ratings adjustments, it is unclear if the efforts expended affect the
informativeness of the rating. Following Jorion et al. (2005), we examine the relative informativeness of credit rating changes for firms whose credit ratings receive more soft adjustments. Then, we
4
extend Dimitrov et al. (2014) to investigate how the repeal of the Reg FD exemption promulgated
by the Dodd-Frank Act impacts the relative informativeness of credit ratings in the post-DoddFrank era. We find that the credit rating changes from analysts that make soft adjustments are
more informative than changes from analysts that do not make soft adjustments. However, consistent with Dimitrov et al. (2014), we find that the increase in informativeness diminishes in the
post-Dodd-Frank era for firms receiving soft rating adjustments.
Our paper contributes to at least two broad literatures. First, we contribute to the literature on
information intermediaries. Distinct from the decades of research on equity analysts, we provide
new insights on how credit analysts employ risk-relevant qualitative information in generating their
primary output, credit ratings. Critically, instead of merely linking qualitative information to the
actual, publicly-reported rating, which relies on both quantitative and qualitative assessments, we
are able to link qualitative information to the soft adjustments themselves, i.e., precisely where the
qualitative information is likely to have the largest impact. Second, we contribute to the growing
literature on qualitative disclosure which has largely focused on how investors and equity analysts
consume and utilize qualitative information. We extend this line of research to demonstrate the
role qualitative information plays in credit markets. Our results should be of interest to academics
in devising their credit risk models, regulators in creating new standards that may cater to debt
clienteles, and managers of firms in choosing how best to convey risk-relevant information to
influence the beliefs of credit analysts.
2
Hypothesis Development
A substantial literature on bankruptcy and credit risk prediction uses both accounting and stock
price data (Horrigan (1966); West (1970): Pogue and Soldofsky (1969); Kaplan and Urwitz (1979),
Blume et al. (1998)), yet corporate disclosures contain a large amount of unstructured textual information, such as the notes to the financial statements in firms’ 10-K filings. Textual, or qualitative,
disclosures are a rich information source as they contain information about the underlying data
5
generating functions, such as firms’ accounting policies and reporting incentives (Li (2010b)). A
growing body of literature on textual disclosure finds that such disclosure conveys information
content with respect to both firm fundamentals and stock price reactions that is incremental to
standard quantitative measures of information content (Tetlock et al. (2008); Engelberg (2008); Li
(2008); Davis et al. (2012)). Further, given the evidence from existing research that textual disclosures incrementally predict future performance (Li (2008); Li (2010a)), it is likely that textual
disclosures contain incremental information content that predicts credit risk. Therefore, our initial
prediction is that rating analysts impound the information contained in qualitative disclosure into
their actual, publicly-reported ratings. Since textual disclosures may be generic “boilerplate,” a
test of our initial prediction allows us to rule out this possibility. However, such a test does not
provide us with an understanding of the mechanism by which credit analysts impound qualitative
disclosure into their reported ratings, that is, whether qualitative disclosure influences analysts’
hard, or quantitative, adjustments versus their soft, or qualitative, adjustments, which is our
primary research question.
Standard & Poor’s performs “quantitative, qualitative, and legal analyses” in forming a rating
opinion (Standard & Poor’s 2008, p.4). Standard & Poor’s supplements the financial statement
analysis with quantitative and qualitative considerations. Hard or quantitative adjustments include adjustments to financial statements. Soft or qualitative considerations include assessments of
the caliber of management, financial transparency, and a company’s competitiveness (Standard &
Poor’s 2008, p.4, 22). Standard & Poor’s claims that the extent of information risk influences the
level of the rating, and in cases of extreme information risk, whether a rating is assigned (Standard
& Poor’s 2008, p.40). Moody’s rating process is similar in that it includes both quantitative and
qualitative aspects. In addition to its hard adjustments, Moody’s examines the following qualitative factors for its soft adjustments: industry structure, financial policy, management quality,
contingent liabilities, aggressive accounting, weak accounting controls, governance risk, and event
risk (Moody’s 2007, slide 43). To the extent our proxies for qualitative disclosure reflect the information credit analysts take into consideration when making their soft adjustments, our next
6
prediction is that credit analysts impound the information contained in qualitative disclosure into
their soft, but not hard, rating adjustments. If we find that our proxies for qualitative disclosure
are related to soft adjustments but are unrelated to hard adjustments, the latter test will corroborate the findings of the former test by ruling out the possibility that the risk-relevant information
conveyed by soft disclosure merely reflects analysts’ adjusted model inputs.
Qualitative information is difficult to process. If a substantial portion of credit-relevant information is contained in qualitative disclosures, then information processing costs are higher for
users of such information. In other words, processing costs may be a function not only of “what
firms say” but also of “how firms say it” or the style by which managers choose to frame the
content of their firms’ disclosures. Disclosure opacity has been extensively studied (Li (2008);
Lehavy et al. (2011); Loughran and McDonald (2014), to name a few); existing literature generally
finds that managers tend to obfuscate firm disclosure to bury details that reflect poor current and
future performance. Further, qualitative disclosure accompanied by numerical disclosure has been
argued to be more credible when “specific enough to be compared with subsequent realizations”
(Hutton et al. (2003)). Therefore, in supplemental analyses, we assess whether the level of opacity
of qualitative disclosure, as well as the level of numerical intensity that complements the disclosure,
impacts how analysts impound qualitative disclosure in their soft ratings adjustment assessments
(Petersen (2004)). If credit analysts are superior information processors, then we expect their soft
adjustments to be unaffected by disclosure opacity. And if analysts perceive qualitative disclosure
to be more precise and credible in the presence of numerical disclosure, then we expect the observed
baseline results to be stronger when disclosures are accompanied by numerical disclosure. While
our supplemental analyses may further our understanding of how managerial style impacts how
credit rating analysts interpret the credit-relevant information in qualitative disclosure, they do
not shed light on our second main research question on whether analysts’ soft rating adjustments
are informative to firm stakeholders, which we turn to next.
On October 23, 2000, the Securities and Exchange Commission (SEC) implemented Regulation
Fair Disclosure (“Reg FD”). Reg FD was adopted to prevent selective disclosure to those who would
7
be reasonably expected to trade securities on the basis of the information or provide others with
advice about securities trading (Ohlson et al. (2010)). However, an exemption was granted to
rating agencies. Reg FD does not apply to disclosures made to rating agencies provided that the
ratings are made publicly available (Carbone (2010)). On September 29, 2010, the SEC amended
Reg FD to remove the express exemption for disclosures of material non-public information for
credit rating agencies (former Rule 100(b)(2)(iii) of Regulation FD). This was required by Section
939B of the Dodd-Frank Wall Street Reform and Consumer Protection Act (“the Dodd-Frank
Act”). This amendment became effective on October 4, 2010 (Ohlson et al. (2010)).
Analysts expend effort in making their soft adjustments to assess credit risk from qualitative
factors that cannot be captured by their quantitative modeling. Hence, we predict that credit rating
changes that involve soft rating adjustments will be more informative than those that do not. To
test this prediction, we use standard event study methodology following Jorion et al. (2005). We
then extend Dimitrov et al. (2014) by comparing the equity market reaction to rating changes
that involve soft rating adjustments before and after the Dodd-Frank Act’s repeal of the Reg FD
exemption. On the one hand, credit analysts’ information supply diminished, arguably increasing
the search costs of information acquisition and the processing costs once found. As a result, credit
analysts are likely to spend more time and effort examining firm disclosures for information that
cannot be captured by their quantitative models alone. If the search and processing costs are not
too great, credit rating informativeness may at best remain unchanged in the post-Dodd-Frank era.
Further, the NRSROs that have publicly addressed the removal of the exemption under Regulation
FD do not believe that it will change the way issuers share material non-public information with the
rating agencies (Carbone (2010)). As a consequence, the informativeness of credit rating changes
that involve soft rating adjustments is ultimately an empirical question and, therefore, we make
no prediction for this test.
8
3
Data and descriptive statistics
3.1
Sample
We derive our sample of soft and hard rating agency adjustments from Moody’s Financial Metrics.
We retrieve all industry methodology reports for the period 2002 through 2014. Our sample
commences with 2002 because it is the first year in which we have data from Moody’s Financial
Metrics. We derive our sample of annual reports from the Securities and Exchange Commission’s
EDGAR database. We retrieve all Form 10-K filings for the same period corresponding to the
availability of the Moody’s Financial Metrics data. To merge the samples, we first match the
firms from Moody’s Financial Metrics to Compustat GVKEY and CIK firm identifiers. We then
match the CIK code on Form 10-K filings to the Compustat GVKEY. We obtain all credit rating
announcements during the sample period from RatingsXpress. We collect all announcements
of long-term entity ratings by Standard & Poor’s (issuer ratings). Rating levels are numerical
transformations of the alphanumerical rating codes issued by Standard & Poor’s, from 1 to 21
(AAA to C). Stock prices are obtained from CRSP and are used to calculate cumulative abnormal
returns surrounding announcements of rating changes.
Hard rating agency adjustments (HARD) are calculated as the difference between the rating
produced from an unadjusted quantitative model (RAT IND REP) and a hypothetical rating based
on a model using adjusted financial statements (RAT IND ADJ). Greater values of HARD imply
greater credit risk assessment arising from hard factors. Soft rating agency adjustments (SOFT)
are calculated as the difference between the actual reported rating (RAT) and a hypothetical rating
based on a model using adjusted financial statements (RAT IND ADJ). Greater values of SOFT
imply greater credit risk assessments arising from soft factors. Rating changes (RCHANGE)
are calculated as the difference between the numerical values assigned to the 21 rating codes.
A downgrade from AAA to AA+ results in a numerical value of 1. Thus, greater values imply
increases in assessed credit risk. Cumulative abnormal return (CAR) is calculated as the cumulative
stock return over the two day period following the rating change less the corresponding return on
9
the CRSP value-weighted index.
Our textual proxies for qualitative disclosure found in annual reports are intended to capture
managerial narrative that conveys information related to credit risk. Following prior literature,
we use textual analysis to extract firm-specific measures of managers’ perceptions of risk-relevant
topics likely to be of interest to credit analysts. That is, from firms’ 10-K’s, we extract proxies of
qualitative disclosure related to debt, profitability, managerial outlooks, and managerial perceptions of uncertainty. While the former two are likely to be related to current risks, the latter two
are likely to be related to future risks.
First, we proxy for qualitative disclosure related to debt (DEBT wc) using the number of
references to a debt-related topic in a firm’s 10-K scaled by 10-K total word count. Second,
we proxy for qualitative disclosure related to earnings (EARNINGS wc) using the number of
references to an earnings-related topic in a firm’s 10-K scaled by 10-K total word count. Third,
we proxy for qualitative disclosure related to prospective disclosure (FLS sc) using the number of
forward-looking statements in a firm’s 10-K scaled by 10-K total sentence count. If a sentence
contains at least one forward-looking term, it is considered to be a forward-looking statement (Li
(2010a); Bozanic et al. (2013)). Fourth, we proxy for qualitative disclosure related to uncertainty
(UNCERTAINTY wc) using the number of references to an uncertainty-related topic in a firm’s
10-K scaled by 10-K total word count (Loughran and McDonald (2011)). See Appendix B for lists
of debt-related, earnings-related, forward-looking, and uncertainty-related terms.
For our cross-sectional tests on managerial style of disclosure framing, we also construct a
measure of readability based on the Fog Index and a measure of numerical intensity. The Fog
Index is calculated as .4[(words/sentences)+100(three or more syllable words/words)] and is meant
to capture the level of disclosure obfuscation. Numerical intensity is calculated as the number of
numbers contained in the narrative sections of a firm’s 10-K scaled by 10-K total word count. It is
meant to capture managers’ preference for providing precise and credible (i.e., verifiable) disclosure
in contrast to mere ”soft talk” or puffery (Hutton et al. (2003); Lundholm et al. (2014)).
Size (SIZE) is measured by the logarithm of total assets. Leverage (LEV) is measured as
10
the sum of short-term debt and long-term debt divided by total assets. Profitability (ROA) is
measured as the ratio of operating income to total assets. Cash flow liquidity (CFO) is measured
as the ratio of operating cash flow to the sum of short-term and long-term debt. Tangibility
(TANG) is measured as the ratio of the sum of inventory and property, plant, and equipment to
total assets. Interest coverage (INTCOV) is measured as the ratio of operating income to interest
expense. LOSSni is a dummy variable equal to one if net income before extraordinary items is
negative, and zero otherwise. Business segments (BSEG) is the number of business segments
and geographic segments (GSEG) is the number of geographic segments. Institutional ownership
(IO) is the number of shares held by institutional investors at the end of the quarter and then
averaged within the year. The Herfindahl index (HHI) is a measure of industry concentration based
on market share. Return volatility (RETVOL) is stock return volatility calculated as standard
deviation of daily stock returns. Market-book-ratio (MB) calculated as market value of equity
divided by book value of total assets. To measure financial statement quality, we use the FSD
Score (FSD) which is based on the level of financial statement divergence from Benford’s Law
following Amiram et al. (2014) where greater value implies lower financial statement quality.
To conduct our analyses, we require that firms have sufficient data to calculate rating agency
adjustments and sufficient financial data to calculate measures of leverage and profitability. The
final sample is 3,055 firm-years representing 927 firms. The sample size drops to 2,509 firm-years
when we require sufficient data to calculate stock market volatility. Interest coverage is winsorized
at 0 and 100 following Blume et al. (1998). All remaining variables are winsorized at the 0.5%
and 99.5% level. With respect to the analysis on rating informativeness, we eliminate firms with
extreme rating changes. Specifically, we eliminate firms with rating changes greater than 6 or less
than 6 notches.
3.2
Descriptive Statistics
Table 1, Panel A presents descriptive statistics for the sample used in our initial analysis on soft
rating adjustments and qualitative disclosure. Table 1, Panel B presents descriptive statistics for
11
the sample used in our secondary analysis on the informativeness of credit ratings changes, which
we discuss in detail in Section 4.3. The average value of SOFT is 0.52, which implies that on
average, credit analysts lower the rating by approximately half a notch due to qualitative risk
factors. In contrast, the average value of HARD is 0.26, which implies that on average, credit
analysts lower the rating by approximately quarter of a notch due to quantitative risk factors. The
average value of RAT is 11 which corresponds to a BB+ (or Ba1) rating.
In terms of the qualitative disclosure proxies, on average, 1.4% of a firms’ 10-K disclosures
discuss debt-related topics, 0.5% discuss earnings-related topics, and 1.3% discuss uncertaintyrelated topics. Of the statements firms make in their 10-K’s, roughly one quarter can be considered
as forward-looking. 2.3% of the average firm’s 10-K narrative disclosure contains numerical content
and the average firm’s disclosure opacity is 20.1 (3.0 measured on a natural log scale). The average
10-K filing has roughly 74,000 words and 2,300 sentences (untabulated).
Regarding the controls for firm characteristics that are likely factors analysts use in making their
hard adjustments, the average firm has assets of 13.4 billion, leverage of 0.37, interest coverage of
7.60, return on assets of 0.09, and market-to-book ratio is 0.93. Cash flow from operations (scaled
by total debt) is 0.46, the fraction of tangible assets is 0.48, and 19% firm-years have negative
net income. In terms of the controls that are likely factors analysts use in making their soft
adjustments, the average firm has 6.4 business segments, 7.6 geographic segments, institutional
holdings of 176 million shares, a Herfindahl index of 0.30, and stock return volatility is 0.03. Lastly,
the measure of financial reporting quality based on Benford’s Law is 0.03, which is consistent with
the results in Amiram et al. (2014).
Table 2 presents the Pearson correlations of SOFT with the proxies for qualitative disclosure
and firm characteristic variables. SOFT has a significant negative association with EARNINGS wc
and significant positive associations with FLS sc and UNCERT wc. This implies that references
to forward-looking statements and uncertainty-related disclosure are correlated with worse (i.e.,
credit risk increasing) SOFT adjustments whereas references to earnings-related disclosure are
correlated with better (i.e., credit risk decreasing) SOFT adjustments.
12
4
Methodology and Empirical Results
4.1
Hard versus Soft Adjustments
As a preliminary test of whether our proxies for qualitative disclosure are likely to contain any
risk-relevant information for credit analysts, we initially examine their relation to the actual ratings
Moody’s publicly reports (RAT). This is a critical first step towards understanding the mechanism
by which the information contained in qualitative disclosure is impounded into the actual rating.
In that sense, this test in part validates our qualitative disclosure proxies before isolating the
possible mechanism, i.e., soft versus hard adjustments, by which the information flows into the
actual rating. We estimate an OLS regression of the following form:
RATi,t+1 = α + β1 DEBT wci,t + β2 EARN IN GS wci,t + β3 F LS sci,t
X
+ β4 U N CERT wci,t + β5 RAT IN D REPi,t +
βj HARD F IRM Controlsi,t
X
+
βk SOF T F IRM Controlsi,t + (1)
We include three sets of controls. First, we include RAT IND REP to control for the estimates
produced by analysts’ quantitative models prior to financial statement adjustments. Second, we
include controls for firm characteristics that likely mirror the variables that serve as inputs to analysts’ quantitative models and are the focus of their hard adjustments (HARD FIRM CONTROLS):
SIZE, LEV, INTCOV, ROA, CFO, TANG, LOSSni, and MB. Third, we include controls likely to
capture firm complexity, monitoring, competition, asset volatility, and financial statement quality (SOFT FIRM CONTROLS): BSEG, GSEG, IO ln, HHI, RETVOL, and FSD. These variables
represent the issues analysts consider when making their soft adjustments. Standard errors are
clustered by firm and year fixed effects are included. Further, given the high collinearity between
FLS sc and UNCERT wc, we include separate regressions that augment the baseline regression
with specifications that exclude each proxy individually.
13
The results from estimating Equation 1 can be found in Table 3. Column 1 includes both
FLS sc and UNCERT wc, whereas Column 2 removes UNCERT wc and Column 3 removes FLS sc
to address the collinearity issue. In all three columns, the full set of controls is included. We find
in Column 1 that the coefficient on DEBT wc is 81.049 and the coefficient on EARNINGS wc is 260.715; both coefficients are statistically significant at the 1% level or better. In terms of economic
magnitudes, a one standard deviation increase in DEBT wc (EARNINGS wc) is associated with a
0.30 notch increase (0.35 notch decrease) in RAT, or 2.7% (3.2%) of the mean value of RAT.
These results imply that actual ratings tend to be lower when firms provide more disclosure
on debt-related topics and higher when firms provide more disclosure on earnings-related topics.
When FLS sc and UNCERT wc are included in the same regression (Column 1), only FLS sc is
statistically significant. However, when we exclude UNCERT wc (Column 2) and exclude FLS sc
(Column 3), both proxies are statistically significant. In terms of economic magnitudes, a one
standard deviation increase in FLS wc (UNCERT wc) is associated with a 0.01 notch increase
(0.82 notch) increase in RAT, or 0.1% (7.4%) of the mean value of RAT. (In column 3, where the
coefficent of UNCERT wc is statistically significant, a one standard deviation increase is associated
with 2.32 notch increase, or a 21.1% of the mean value of RAT.) We therefore view UNCERT wc
and FLS sc as partial substitutes for one another, where increases in either tend to lower actual
credit ratings. Taken together, we view these results as supporting the prediction that publiclyreported credit ratings reflect the information conveyed by qualitative disclosure incremental to
the hard financial statement inputs of analysts’ quantitative models as well as the soft factors that
credit analysts are known to consider in making their soft adjustments.
We next turn to our primary research question: whether qualitative disclosure influences the
soft rating adjustments credit rating analysts make. In order to answer this question, we modify
14
Equation 1 as follows:
SOF Ti,t+1 = α + β1 DEBT wci,t + β2 EARN IN GS wci,t + β3 F LS sci,t
X
+ β4 U N CERT wci,t + β5 RAT IN D ADJi,t +
βj HARD F IRM Controlsi,t
X
+
βk SOF T F IRM Controlsi,t + (2)
Aside from investigating a new dependent variable, Equation 2 is similar to Equation 1 with one
important exception: we now control for the indicated adjusted rating, that is, the rating produced
from credit analysts’ models after the underlying financial statement data used to estimate the
models have undergone hard, but not soft, adjustments.
The results from estimating Equation 2 can be found in Table 4, which follows a similar
format to that found in Table 3. We find in Column 1 that the coefficient on DEBT wc is 43.365
and the coefficient on EARNINGS wc is -214.973; both coefficients are statistically significant
at the 1% level or better. In terms of economic magnitudes, a one standard deviation increase
in DEBT wc (EARNINGS wc) is associated with a 0.16 notch increase (0.34 notch decrease) in
SOFT, or 31% (66%) of the mean value of SOFT. These results suggest that credit rating analysts
are more likely to make credit risk increasing soft adjustments when firms provide more disclosure
on debt-related topics. These results further suggest that when managers provide more discussion
on earnings-related topics, credit rating analysts are more likely to make credit risk decreasing
soft adjustments. Looking across Columns 1-3, similar to before, we also find that both forwardlooking and uncertain disclosure increase the likelihood that credit rating analysts make credit risk
increasing soft adjustments, yet FLS sc appears to be the dominant driver. In terms of economic
magnitudes, a one standard deviation increase in FLS wc (UNCERT wc) is associated with a
0.01 notch increase (1.12 notch) increase in SOFT, or 1.4% (228%) of the mean value of SOFT.
While the results on the relation between credit rating analysts’ soft adjustments and qualitative
disclosure largely mirror those related to the reported rating, the economic significance of the
15
impact of qualitative disclosure on soft adjustments is greater. Collectively, these results support
our prediction that credit rating analysts’ soft adjustments reflect the risk-relevant information
conveyed by qualitative disclosure.
Despite the initial results that support our predictions, we cannot conclude that soft adjustments are the mechanism by which analysts impound the information found within credit-relevant
qualitative disclosure. That is, we have yet to rule other whether or not the information conveyed
by qualitative disclosure is uniquely impounded into analysts’ soft rating adjustments or if the information merely reflects information similar to their hard rating adjustments. In order to answer
this question, we modify the dependent variable in Equation 1 as follows:
HARDi,t+1 = α + β1 DEBT wci,t + β2 EARN IN GS wci,t + β3 F LS sci,t
X
+ β4 U N CERT wci,t + β5 RAT IN D REPi,t +
βj HARD F IRM Controlsi,t
X
+
βk SOF T F IRM Controlsi,t + (3)
The results from estimating Equation 3 can be found in Table 5, which follows a similar format
to that found in Table 4. Across Columns 1-3, we find no evidence that risk-relevant qualitative
disclosure influences credit analysts’ hard ratings adjustments. We view this result as supporting
those found in Table 4 on soft rating adjustments which suggest that soft ratings adjustments, as
opposed to hard ratings adjustments, are the mechanism by which the information contained in
qualitative disclosure makes its way into the actual rating publicly reported by the credit rating
agency.
4.2
Cross-Sectional Tests
In this section, we expand on the baseline regressions on soft ratings adjustments to consider how
disclosure style affects the relation between soft rating adjustments and qualitative disclosure.
Having established that credit analysts’ soft adjustments reflect the information conveyed in qual16
itative disclosure, it may be the case that the information impounded is impacted by the level of
disclosure opacity and the amount of numerical disclosure that accompanies the textual disclose.
We augment Equation 2 to include interaction terms for disclosures that possess above sample
median opacity (FOG high) and above sample median numerical intensity (NT high). The results
of these estimations can be found in Tables 6 and 7, respectively.
In Table 6, Column 1 we find that the coefficient on the interaction term Earnings wc*FOG high
is -117.873 and is statistically significant at the 5% level or better. This suggests that credit rating
analysts are more likely to make credit risk decreasing soft adjustments for opaque, earnings-related
qualitative disclosure, which is consistent with managers successfully obfuscating firm performance
(Li (2008)). No other proxies for qualitative disclosure are impacted by varying the level of disclosure obfuscation, suggesting that either managers do not attempt to obfuscate disclosure related
to the other credit-relevant topics or analysts see through such attempts by managers.
In Table 7, Column 1 we find that the coefficient on the interaction term Debt wc*NT high is
-38.556 and is statistically significant at the 10% level or better. This suggests that credit rating
analysts are more likely to make credit risk increasing soft adjustments for debt-related qualitative
disclosure accompanied by high numerical intensity, which is consistent with numerical disclosure
adding precision to credit analysts’ beliefs regarding such topics. In addition, we find in Column 2
that the coefficient on the interaction term FLS sc*NT high is -3.712 and is statistically significant
at the 5% level or better. This suggest that credit rating analysts are more likely to make credit risk
decreasing soft adjustments for forward-looking disclosure accompanied by high numerical intensity, which is consistent with numerical disclosure adding credibility to forward-looking disclosures
provided by management (Bozanic et al. (2013)). No other proxies for qualitative disclosure are
consistently impacted by varying the level of numerical intensity.
17
4.3
4.3.1
Informativeness of Ratings Changes
Full Period Sample
In our final set of analyses, we examine the informativeness of rating changes that involve soft rating
adjustments. To do so, we obtain a sample that consists of 415 downgrades and 336 upgrades (for
a total of 751 rating changes) that involve soft rating adjustments. Each issuer rating change
constitutes one sample observation. Our sample selection also requires the availability of daily
stock returns data for the sample firms in order to compute abnormal stock returns. The average
downgrade lowers the rating by 1.3 notches whereas the average upgrade increases the rating by
1.1 notches. Less than ten percent of changes, whether in the form of upgrades or downgrades,
involve revisions from investment grade to speculative grade or vice versa. Following Jorion et al.
(2005), we estimate an OLS regressions of the following form:
CARi,t = α0 + β1 RCHAN GEi,t + β2 N onZeroSOF Ti,t
+ β3 IGRADEi,t + β3 DAY S lni,t + β4 RAT priori,t + (4)
CAR is the cumulative stock return over the two day period following the rating change less the
corresponding return on the CRSP value-weighted index. RATCHANGE is the size of the rating
change, which is the difference between the numerical value of the new rating and that of the old
rating of the same issuer. NonZeroSOFT is a dummy variable equal to one if the absolute value
of SOFT is greater than zero, and zero otherwise. IGRADE is a dummy variable equal to one if
a rating is revised from investment grade to speculative grade or vice versa, and zero otherwise.
DAYS is the natural log of the number of days since the previous rating change. RAT prior is the
prior rating before the change.
The variable of interest is NonZeroSOFT. If rating changes are more informative for firms
with non-zero SOFT adjustments, then we would expect the coefficient on NonZeroSOFT to be
negative. Ceteris paribus, the larger is the RCHANGE variable, the larger the stock price response
18
should be. Consistent with Jorion et al. (2005), we expect the slope coefficient on RCHANGE to
be negative for rating changes.1
Table 8 presents the results of estimating Equation 4. Results are presented for all ratings
changes and then separately for ratings downgrades and upgrades. We present several specifications
that vary the inclusion of our test and control variables. However, since the results are consistent
across the specifications, we focus our discussion on Columns 4, 8, and 12.
In Column 4, where we use all rating changes, the coefficient on RCHANGE is -0.009 and is
statistically significant at the 1% level or better. In Column 8, where we focus on downgrades, the
coefficient on RCHANGE is -0.019 and is statistically significant at the 10% level or better. We
find no statistically significant results for upgrades, consistent with the market not finding credit
rating upgrades to convey new or credible information content. In line with prior literature, the
results collectively suggest that the market finds credit rating downgrades to be informative.
Building on prior literature, we next examine our variable of interest, NonZeroSOFT. We find
that the coefficient on NonZeroSOFT in Column 4 is -0.012 and is statistically significant at the
10% level or better. We further find in Column 8 that the relation is driven by credit rating
downgrades. The coefficient on NonZeroSOFT in Column 8 is -0.021 and is statistically significant
at the 5% level or better. However, the coefficient on NonZeroSOFT is not statistically significant
in column 12. Collectively, our results suggest that credit rating downgrades that involve soft
adjustments are more informative than credit rating changes that do not.
4.3.2
Repeal of the Reg FD Exemption
We conclude by examining how the repeal of the Reg FD exemption impacted the informativeness
of credit ratings that involve soft adjustments. Following Dimitrov et al. (2014), we augment
1
In untabulated analysis, we find that the characteristics of firms with NonZeroSOFT equal to one are largely
indistinguishable from firms with NonZeroSOFT equal to zero.
19
Equation 4 to include an interaction term for the repeal of the Reg FD exemption (RepealRegFD).
CARi,t = α0 + β1 RCHAN GEi,t + β2 RepealRegF Di,t + β3 N onZeroSOF Ti,t
+ β4 N onZeroSOF T ∗ RepealRegF Di,t + β5 IGRADEi,t + β6 DAY S lni,t
+ β7 RAT priori,t + epsilon
(5)
The variable of interest in Equation 5 is NonZeroSOFT*RepealRegFD, which is the interaction
between NonZeroSOFT and RepealRegFD. It is meant to capture the relative informativeness of
credit rating changes that involve soft adjustments after the repeal of Reg FD. We focus our
discussion on Columns 2, 4, and 6 since the results without RAT prior are similar. While we
continue to find results similar to those found in Table 8 for NonZeroSOFT, consistent with
Dimitrov et al. (2014), we find that the increase in informativeness diminishes after the repeal of
Reg FD for ratings changes that involve soft rating adjustments.
5
Conclusion
In this paper, we study the inputs that influence credit analysts’ soft rating adjustments and the
informativeness of those adjustments. While much is known about how analysts quantitatively
model their credit rating assessments, little is known about the soft adjustments analysts make
prior to publicly releasing a credit rating. We focus on a single channel by which analysts’ soft
adjustments are likely to be influenced: credit-relevant qualitative disclosure. We first validate
our proxies for qualitative disclosure by showing their relation to the actual ratings analysts publicly disclose. Controlling for credit analysts’ hard adjustments, firm characteristics that likely
influence hard adjustments, and other variables analysts are likely to consider in making their soft
adjustments, we find that qualitative disclosure influences analysts’ soft credit ratings adjustments.
Specifically, we find that credit rating analysts are more likely to make credit risk increasing soft
20
adjustments when firms provide more disclosure on debt-related topics. When managers provide
more discussion on earnings-related topics, credit rating analysts are more likely to make credit
risk decreasing soft adjustments. We also find that both forward-looking and uncertainty-related
disclosure increase the likelihood that credit rating analysts make credit risk increasing soft adjustments. We further rule out that qualitative disclosure is reflected in credit analysts’ hard
adjustments in order to isolate the mechanism by which qualitative disclosure flows through to the
publicly-reported rating.
In cross-sectional tests, we then examine several qualitative disclosure characteristics meant to
capture how managers frame their disclosure. Namely, we assess whether the level of opacity of
qualitative disclosure, as well as the level of numerical intensity that complements the disclosure,
impacts how analysts impound qualitative disclosure in their soft ratings adjustment assessments.
We find that credit rating analysts are more likely to make credit risk decreasing soft adjustments
for opaque, earnings-related qualitative disclosure. This finding is consistent with managers attempting to obfuscate performance (Li (2008)). In terms of numerical disclosure, we find that credit
rating analysts are more likely to make credit risk decreasing soft adjustments for forward-looking
disclosure accompanied by high numerical intensity, which is consistent with numerical disclosure
adding credibility to forward-looking disclosures provided by management (Hutton et al. (2003)).
Finally, following Jorion et al. (2005), we examine the informativeness of credit rating changes
that involve soft adjustments. After confirming Jorion et al.’s results in our sample, we find that
credit rating changes that involve soft adjustments are more informative than those that do not,
and this relation is driven by credit rating downgrades. We then extend the Jorion et al. (2005)
study in a manner similar to Dimitrov et al. (2014) to examine the informativeness of credit rating
changes that involve soft adjustments after the repeal of the Reg FD exemption previously afforded
to credit rating agencies. We find that the increase in informativeness diminishes after the repeal
of Reg FD for ratings changes that involve soft rating adjustments.
Our paper contributes to our understanding of how important information intermediaries, credit
analysts, employ qualitative disclosure in forming their soft risk assessments. While much is known
21
about how equity analysts and equity market investors use and process qualitative information,
little is known about how credit analysts use this information. In this paper, we are able to link
qualitative information to the soft adjustments themselves, i.e., precisely where the qualitative
information is likely to have the largest impact. In so doing, we are able to directly examine
the mechanism by which qualitative disclosure is reflected in the final, publicly-reported output
traditionally studied by academics and consumed by firm stakeholders. Our results should be of
interest to academics in devising their credit risk models, regulators in creating new standards that
may cater to debt clienteles, and managers of firms in choosing how best to convey risk-relevant
information to influence the beliefs of credit analysts.
22
References
Altman, E. I. (1968). Financial ratios, discriminant analysis and the prediction of corporate
bankruptcy. The Journal of Finance 23 (4), pp. 589–609.
Amiram, D., Z. Bozanic, and E. Rouen (2014). Financial statement irregularities: Evidence
fromt he distributional properties of financial statemet numbers. Working Paper.
Blume, M. E., F. Lim, and A. C. Mackinley (1998). The declining credit quality of U.S.
corporate debt: Myth or reality? Journal of Finance 53 (4), 1389–1413.
Bozanic, Z., D. T. Roulstone, and A. V. Buskirk (2013). Management earnings forecasts and
forward-looking statements. Working Paper, Ohio State University.
Carbone, D. (2010). The impact of the Dodd-Frank Act’s credit rating agency reform of public
companies. The Corporate & Securities Law Advisor .
Davis, A. K., J. M. Piger, and L. M. Sedor (2012). Beyond the numbers: Measuring the
information content of earnings press release language. Contemporary Accounting
Research 29 (3), 845–868.
Dimitrov, V., D. Palia, and L. Tang (2014). Impact of the Dodd-Frank Act on credit ratings.
Journal of Financial Economics, forthcoming.
Easton, P., S. J. Monahan, and F. P. Vasvari (2009). Initial evidence on the role of accounting
earnings in the bond market. Journal of Accounting Research 47 (3), 721–766.
Engelberg, J. (2008). Costly information processing: Evidence from earnings announcements.
Working Paper.
Horrigan, J. O. (1966). The determination of long-term credit standing with financial ratios.
Journal of Accounting Research 4, 44–62.
Hutton, A. P., G. S. Miller, and D. J. Skinner (2003). The role of supplementary statements with
management earnings forecasts. Journal of Accounting Research 41 (5), 867–890.
Jennings, R. (1987). Unsystematic security price movements, management earnings forecasts,
and revisions in consensus analyst earnings forecasts. Journal of Accounting Research 25 (1),
90–110.
Jorion, P., Z. Liu, and C. Shi (2005). Informational effects of Regulation FD: Evidence from
rating agencies. Journal of Financial Economics 76 (2), 309–330.
Kaplan, R. S. and G. Urwitz (1979). Statistical models of bond ratings: A methodological
inquiry. Journal of Business 52 (2), 231–61.
Kraft, P. (2014). Rating agency adjustments to GAAP financial statements and their effect on
ratings and credit spreads. The Accounting Review , forthcoming.
23
Lehavy, R., F. Li, and K. Merkley (2011). The effect of annual report readability on analyst
following and the properties of their earnings forecasts. The Accounting Review 86 (3),
1087–1115.
Li, F. (2008). Annual report readability, current earnings, and earnings persistence. Journal of
Accounting and Economics 45 (23), 221 – 247.
Li, F. (2010a). The information content of forward-looking statements in corporate filings—A
naive Bayesian machine learning approach. Journal of Accounting Research 48 (5), 1049–1102.
Li, F. (2010b). Textual analysis of corporate disclosures: A survey of the literature. Journal of
Accounting Literature 29, 143–165.
Loughran, T. and B. McDonald (2011). When is a liability not a liability? Textual analysis,
dictionaries, and 10-Ks. The Journal of Finance 66 (1), 35–65.
Loughran, T. and B. McDonald (2014). Measuring readability in financial disclosures. The
Journal of Finance LXIX (4), 1643–1671.
Lundholm, R., R. Rogo, and J. L. Zhang (2014). Restoring the tower of Babel: How foreign firms
communicate with US investors. The Accounting Review , forthcoming. Forthcoming.
Moody’s (2007). Financial reporting and credit ratings (Presentation by Greg Jonas), CARE
conference, Napa California.
Ohlson, J. F., B. J. Lane, R. O. Mueller, A. L. Goodman, and A. Fabens (2010). Repeal of credit
ratings agency exemption from regulation fd. Gibson Dunn.
Petersen, M. A. (2004). Information: Hard and soft. Working Paper, Northwestern University.
Pogue, T. F. and R. M. Soldofsky (1969). What’s in a bond rating. The Journal of Financial
and Quantitative Analysis 4 (2), 201–228.
Pownall, G., C. Wasley, and G. Waymire (1993). The stock price effects of alternative types of
management earnings forecasts. The Accounting Review 68 (4), 896–912.
Skinner, D. J. (1994). Why firms voluntarily disclose bad news. Journal of Accounting
Research 32 (1), 38–60.
Standard & Poor’s (2008). Corporate ratings criteria.
Tetlock, P. C., M. Saar-Tsechansky, and S. Macskassy (2008). More than words: Quantifying
language to measure firms’ fundamentals. Journal of Finance LXIII (3), 1437–1467.
West, R. R. (1970). An alternative approach to predicting corporate bond ratings. Journal of
Accounting Research 8 (1), 118–125.
24
APPENDIX A: VARIABLE DEFINITIONS
Variable
Definition
Credit rating variables
SOFT
Actual rating less indicated adjusted rating (RAT_IND_ADJ). Greater value implies greater credit risk
assessment.
HARD
Indicated adjusted rating (RAT_IND_ADJ) less indicated reported rating (RAT_IND_REP). Greater value
implies greater credit risk assessment.
RAT
Actual issuer rating by Moody's.
RAT_IND_REP
Indicated reported issuer rating by Moody's from quantitative modeling before quantitative adjustments to
underlying financial statement inputs.
RAT_IND_ADJ
Indicated adjusted issuer rating by Moody's after quantitative adjustments to underlying financial statement
inputs.
Qualitative disclosure proxies
DEBT_wc
The number of references to a debt-related topic in a firm's 10-K scaled by 10-K total word count. See
Appendix B for a list of debt-related terms.
EARNINGS_wc
The number of references to an earnings-related topic in a firm's 10-K scaled by 10-K total word count. See
Appendix B for a list of earnings-related terms.
FLS_sc
The number of forward-looking statements in a firm's 10-K scaled by 10-K total sentence count. See
Appendix B for a list of forward-looking terms. If a sentence contains at least one forward-looking term, it is
considered to be a forward-looking statement (FLS).
UNCERT_wc
The number of references to an uncertainty-related topic in a firm's 10-K scaled by 10-K total word count. See
Appendix B for a list of uncertainty-related terms.
NT_wc
The number of numbers contained in the narrative (i.e., non-tabulated) sections of a firm's 10-K scaled by 10K total word count.
NT_high
Dummy variable equal to one if NT_wc is greater than sample median, and zero otherwise.
FOG_ln
Log of the Fog Index, which is calculated as .4[(words/sentences)+100(three or more syllable words/words)].
Note that a higher Fog Index indicates the disclosure is less readable.
FOG_high
Dummy variable equal to one if FOG_ln is greater than sample median, and zero otherwise.
Firm characteristics
SIZE
AT
LEV
ROA
CFO
TANG
INTCOV
LOSSni
MB
BSEG
GSEG
IO
IO_ln
HHI
RETVOL
FSD
Log of total assets.
Total assets.
(Short-term debt + long-term debt) / total assets.
Operating income / total assets.
Operating cash flow / (short-term debt + long-term debt).
(Inventory + property, plant, and equipment) / total assets.
Operating income / interest expense on debt winsorized at 0 and 100.
Dummy variable equal to one if net income before extraordinary items is negative, and zero otherwise.
Market-book-ratio calculated as market value of equity divided by book value of total assets.
Number of business segments.
Number of geographic segments.
Institutional ownership. Number shares held by institutional investors at the end of the quarter and then
averaged within the year.
Ln of institutional ownership.
Herfindahl index. Greater value implies a more monopolistic industry (less competition).
Stock return volatility calculated as standard deviation of daily stock returns.
Index of financial reporting quality based on the level of financial statement divergence (FSD) from Benford's
Law following Amiram et al. (2014). Greater value implies lower financial statement quality.
Price response analysis
RCHANGE
The difference between the new rating and prior rating. Between -6 and 6. Greater number implies downgrade
(e.g., AAA to AA+ = 2 - 1 = 1).
CAR
Cumulative stock return over the two day period following the rating change less the corresponding return on
the CRSP value-weighted index.
NonZeroSOFT
Dummy variable equal to one if the absolute value of SOFT is greater than zero, and zero otherwise.
RAT_prior
Prior rating.
RepealRegFD
Dummy variable equal to one if observation is later than October 4, 2010, zero otherwise.
IGRADE
Dummy variable equal to one if a rating is revised from investment grade to speculative grade or vice versa,
and zero otherwise.
DAYS
Number of days since last rating change.
DAYS_ln
Log of number of days since last rating change.
25
APPENDIX B: WORD LISTS BY TOPIC
Below are the word lists by topic category used to construct the proxies for qualitative disclosure used in the study.
Debt
Balance sheet, bond, bonds, bondholder, bondholders, cash, cashflow, coverage, debt, debts, default, defaults, financing, funding, debt
issue, debt issues, leverage, liability, liabilities, liquid, liquidity, loan, loans, payable, payables, coupon payment, coupon payments,
stressed, distressed, fixed income, interest, notes, covenant, covenants, solvent, solvency, credit
Earnings
Earnings, income, loss, losses, profit, profits, EPS
Forward-Looking
Anticipate, believe, could, expect, forecast, intend, may, might, plan, project, should, will
Please refer to Bozanic, Roulstone, and Van Buskirk (2013) for the full list of forward-looking terms.
Uncertainty
Ambiguous, cautious, contingent, depends, doubt, precaution, risky, speculation, uncertain, volatile
Please refer to Loughran and McDonald (2011) for the full list of uncertainty-related terms.
26
TABLE 1
Descriptive statistics
Panel A presents summary information on our sample of 3,055 firm-years (927 firms) over the period from 2002 to 2013 for the main
analysis. Panel B presents summary information on our sample of 751 rating changes (462 firms) representing 415 downgrades and 336
upgrades for the price response analysis. All firm characteristics and textual variables are measured the year prior to the measurement of
Moody's ratings and adjustments. All variables are defined in the Appendix.
Variable
Panel A
SOFT
HARD
RAT
RAT_IND_REP
RAT_IND_ADJ
DEBT_wc
EARNINGS_wc
UNCERT_wc
FLS_sc
NT_wc
FOG_ln
mean
min
p25
p50
p75
max
sd
N
0.52
0.26
11.00
10.00
11.00
1.4%
0.5%
1.3%
23.0%
2.3%
3.00
-5.00
-4.00
1.00
2.00
2.00
0.6%
0.1%
0.7%
13.0%
1.3%
2.80
-1.00
0.00
9.00
8.00
8.00
1.1%
0.4%
1.1%
20.0%
2.0%
2.90
0.00
0.00
11.00
10.00
11.00
1.3%
0.5%
1.3%
23.0%
2.3%
3.00
1.00
1.00
14.00
13.00
13.00
1.6%
0.6%
1.4%
26.0%
2.6%
3.00
6.00
4.00
19.00
18.00
18.00
2.5%
1.0%
1.9%
35.0%
3.7%
3.50
1.70
1.00
3.40
3.60
3.30
0.4%
0.2%
0.2%
4.0%
0.4%
0.11
3,055
3,055
3,055
3,055
3,055
3,055
3,055
3,055
3,055
3,055
3,055
Firm Controls
SIZE
AT
LEV
INTCOV
ROA
CFO
TANG
LOSSni
MB
BSEG
GSEG
IO_ln
HHI
RETVOL
FSD
8.50
13,408
0.37
7.60
0.09
0.46
0.48
0.19
0.93
6.40
7.60
18.00
0.30
0.03
0.03
5.30
205
0.01
0.00
-0.26
-0.21
0.02
0.00
0.04
0.00
0.00
0.00
0.02
0.01
0.01
7.60
1,945
0.22
1.90
0.05
0.14
0.26
0.00
0.46
2.00
3.00
18.00
0.13
0.02
0.02
8.40
4,538
0.32
3.70
0.08
0.27
0.48
0.00
0.74
4.00
6.00
18.00
0.23
0.02
0.03
9.40
12,149
0.46
8.50
0.12
0.50
0.69
0.00
1.20
10.00
12.00
19.00
0.40
0.03
0.03
12.00
202,352
1.50
100.00
0.39
6.80
0.94
1.00
4.60
24.00
39.00
22.00
1.00
0.09
0.05
1.40
26,464
0.22
12.00
0.08
0.74
0.25
0.40
0.71
5.90
7.10
3.30
0.24
0.01
0.01
3,055
3,055
3,055
3,055
3,055
3,055
3,055
3,055
2,509
2,509
2,509
2,509
2,509
2,509
2,509
Panel B
Price response analysis
Downgrades
RCHANGE
CAR
IGRADE
DAYS
1.30
-0.015
0.08
1,171
1.00
-0.430
0.00
8
1.00
-0.042
0.00
175
1.00
-0.011
0.00
606
1.00
0.016
0.00
1,607
5.00
0.400
1.00
19,313
0.66
0.100
0.27
1,604
415
415
415
415
Upgrades
RCHANGE
CAR
IGRADE
DAYS
-1.10
0.002
0.10
1,016
-5.00
-0.170
0.00
22
-1.00
-0.017
0.00
376
-1.00
0.003
0.00
750
-1.00
0.019
0.00
1,282
-1.00
0.260
1.00
9,242
0.53
0.038
0.29
1,044
336
336
336
336
27
TABLE 2
Pearson correlation
The table presents pairwise Pearson correlation between key variables. The sample consists of 3,055 firm-years (927 firms) with available financial and adjustments data. All firm characteristics and textual variables are measured the year
prior to the measurement of Moody's ratings and adjustments. All variables are defined in the Appendix. * denotes 5% significance.
SOFT
RAT
DEBT_wc
EARNINGS_wc
UNCERT_wc
FLS_sc
BSEG
GSEG
IO_ln
HHI
RETVOL
MB
FSD
SIZE
LEV
INTCOV
ROA
CFO
TANG
LOSSni
SOFT
1.0000
0.3021*
0.0353
-0.1145*
0.0603*
0.0864*
-0.0573*
0.0644*
0.0382*
0.0628*
0.1312*
0.0366
0.0112
-0.0751*
0.0045
0.0515*
0.0858*
0.0628*
-0.0187
-0.0130
RAT
1.0000
0.3821*
-0.2221*
0.0324
0.0717*
-0.1958*
-0.1374*
-0.1185*
-0.0176
0.5004*
-0.4440*
0.0893*
-0.6143*
0.4839*
-0.4831*
-0.3543*
-0.3426*
0.0635*
0.3963*
DEBT_wcARNINGS_w
UNCERT_w FLS_sc
1.0000
0.1037*
0.0994*
0.0105
-0.0814*
-0.1181*
-0.0938*
-0.0552*
0.1408*
-0.2955*
0.0401*
-0.2028*
0.3889*
-0.2844*
-0.2255*
-0.2831*
0.0382*
0.1634*
1.0000
0.3112*
0.0420*
0.1061*
0.1541*
0.1805*
0.1873*
-0.1962*
0.1443*
-0.1168*
0.0769*
-0.1956*
0.1610*
0.1441*
0.0889*
-0.1865*
-0.1299*
1.0000
0.5165*
-0.0347
0.0098
0.0595*
0.0435*
0.0428*
0.0549*
-0.0250
-0.0566*
-0.0612*
0.0115
-0.0115
0.0003
-0.0358*
0.0169
1.0000
-0.1259*
-0.0719*
-0.0318
-0.0354
-0.0052
-0.0340
0.0239
0.0126
0.0018
-0.0195
-0.0778*
-0.0284
0.0977*
0.0115
BSEG
GSEG
IO_ln
HHI
RETVOL
MB
FSD
SIZE
LEV
INTCOV
ROA
CFO
TANG
1.0000
0.2905*
0.1693*
0.1015*
-0.0502*
-0.0314
-0.0849*
0.1345*
-0.1760*
0.0899*
0.0570*
0.0640*
-0.2159*
-0.1045*
1.0000
0.2534*
0.1896*
0.0309
0.0872*
-0.1379*
0.1195*
-0.2469*
0.1983*
0.1429*
0.1977*
-0.1644*
-0.0629*
1.0000
0.2229*
-0.1369*
0.1678*
-0.1889*
0.1647*
-0.2410*
0.1626*
0.1328*
0.1405*
-0.1657*
-0.1071*
1.0000
0.0287
0.0266
-0.1222*
-0.0121
-0.0796*
0.0472*
0.1022*
0.0015
-0.2954*
0.0184
1.0000
-0.3486*
0.0588*
-0.3920*
0.2077*
-0.2174*
-0.3516*
-0.0983*
0.1643*
0.4164*
1.0000
-0.0639*
0.0824*
-0.2081*
0.4648*
0.6233*
0.3463*
-0.1784*
-0.2849*
1.0000
-0.1367*
0.0869*
-0.0597*
-0.1065*
-0.0151
0.1745*
0.0482*
1.0000
-0.3404*
0.2702*
0.0880*
0.1766*
0.0101
-0.2261*
1.0000
-0.4186*
-0.0883*
-0.4433*
0.0658*
0.2939*
1.0000
0.4586*
0.7715*
-0.1941*
-0.2564*
1.0000
0.3054*
-0.2449*
-0.4942*
1.0000
-0.0780*
-0.1823*
1.0000
0.0466*
28
TABLE 3
Regression of Moody's rating on qualitative aspects of disclosure
This table displays the results for the ordinary least squares (OLS) estimation where the dependent variable is
RAT, Moody's rating. The sample consists of 3,055 firm-years (927 firms) with available financial and
adjustments data. All firm characteristics and textual variables are measured the year prior to the
measurement of Moody's ratings and adjustments. All variables are defined in the Appendix. Year fixed
effects are included and standard errors are robust and clustered by firm. Robust standard errors are presented
in brackets. ** denotes p<0.01, * denotes p<0.05, + denotes p<0.1.
Dependent variable
DEBT_wc
EARNINGS_wc
FLS_sc
UNCERT_wc
RAT_IND_REP
SIZE
LEV
INTCOV
ROA
CFO
TANG
LOSSni
MB
BSEG
GSEG
IO_ln
HHI
RETVOL
FSD
Constant
Observations
R-squared
(1)
RAT
(2)
RAT
(3)
RAT
81.049**
(15.035)
-219.357**
(34.481)
4.000**
(1.442)
20.385
(26.168)
0.438**
(0.030)
-0.566**
(0.057)
-0.345
(0.356)
-0.040**
(0.008)
2.241*
(0.905)
0.342*
(0.151)
0.053
(0.254)
-0.651**
(0.133)
-0.442**
(0.104)
-0.029**
(0.011)
0.004
(0.010)
-0.028+
(0.017)
-0.011
(0.291)
60.066**
(5.777)
9.193+
(5.251)
7.825**
(0.806)
82.045**
(14.962)
-211.070**
(31.346)
4.598**
(1.231)
77.047**
(15.058)
-233.342**
(34.571)
0.439**
(0.030)
-0.568**
(0.057)
-0.361
(0.354)
-0.041**
(0.008)
2.258*
(0.903)
0.342*
(0.151)
0.041
(0.252)
-0.645**
(0.133)
-0.435**
(0.103)
-0.029**
(0.011)
0.004
(0.010)
-0.029+
(0.016)
-0.016
(0.290)
60.324**
(5.753)
9.098+
(5.260)
7.894**
(0.804)
58.145**
(22.369)
0.440**
(0.030)
-0.562**
(0.057)
-0.319
(0.357)
-0.040**
(0.007)
2.095*
(0.909)
0.336*
(0.150)
0.119
(0.253)
-0.680**
(0.133)
-0.449**
(0.105)
-0.030**
(0.011)
0.005
(0.010)
-0.028+
(0.017)
-0.005
(0.292)
59.716**
(5.761)
8.962+
(5.259)
8.203**
(0.807)
2,509
0.77
2,509
0.77
2,509
0.77
29
TABLE 4
Regression of Moody's soft adjustment on qualitative aspects of disclosure
This table displays the results for the ordinary least squares (OLS) estimation where the dependent variable
is SOFT, Moody's soft adjustment. The sample consists of 3,055 firm-years (927 firms) with available
financial and adjustments data. All firm characteristics and textual variables are measured the year prior to
the measurement of Moody's ratings and adjustments. All variables are defined in the Appendix. Year fixed
effects are included and standard errors are robust and clustered by firm. Robust standard errors are
presented in brackets. ** denotes p<0.01, * denotes p<0.05, + denotes p<0.1.
Dependent variable
DEBT_wc
EARNINGS_wc
FLS_sc
UNCERT_wc
RAT_IND_ADJ
SIZE
LEV
INTCOV
ROA
CFO
TANG
LOSSni
MB
BSEG
GSEG
IO_ln
HHI
RETVOL
FSD
Constant
Observations
R-squared
(1)
SOFT
(2)
SOFT
(3)
SOFT
43.365**
(12.989)
-214.973**
(29.818)
3.142*
(1.240)
29.653
(23.783)
-0.318**
(0.032)
-0.314**
(0.050)
0.048
(0.283)
-0.013*
(0.007)
1.039
(0.883)
0.057
(0.118)
-0.126
(0.218)
0.137
(0.117)
-0.255**
(0.096)
-0.025**
(0.009)
0.014
(0.008)
-0.030*
(0.013)
0.173
(0.228)
40.301**
(5.449)
6.945
(4.980)
4.058**
(1.005)
44.859**
(12.854)
-203.008**
(27.163)
4.013**
(1.070)
40.164**
(12.968)
-225.925**
(29.658)
-0.318**
(0.032)
-0.318**
(0.050)
0.030
(0.282)
-0.014*
(0.007)
1.057
(0.880)
0.057
(0.118)
-0.143
(0.217)
0.147
(0.117)
-0.246**
(0.093)
-0.025**
(0.009)
0.014
(0.008)
-0.030*
(0.013)
0.165
(0.227)
40.723**
(5.431)
6.803
(5.000)
4.176**
(1.016)
59.350**
(20.505)
-0.316**
(0.032)
-0.310**
(0.050)
0.069
(0.285)
-0.013*
(0.006)
0.928
(0.888)
0.052
(0.118)
-0.075
(0.216)
0.115
(0.117)
-0.260**
(0.097)
-0.026**
(0.009)
0.014+
(0.008)
-0.030*
(0.013)
0.178
(0.228)
39.989**
(5.433)
6.762
(4.962)
4.338**
(0.985)
2,509
0.18
2,509
0.18
2,509
0.18
30
TABLE 5
Regression of Moody's hard adjustment on qualitative aspects of disclosure
This table displays the results for the ordinary least squares (OLS) estimation where the dependent variable
is HARD, Moody's hard adjustment. The sample consists of 3,055 firm-years (927 firms) with available
financial and adjustments data. All firm characteristics and textual variables are measured the year prior to
the measurement of Moody's ratings and adjustments. All variables are defined in the Appendix. Year fixed
effects are included and standard errors are robust and clustered by firm. Robust standard errors are
presented in brackets. ** denotes p<0.01, * denotes p<0.05, + denotes p<0.1.
Dependent variable
DEBT_wc
EARNINGS_wc
FLS_sc
UNCERT_wc
RAT_IND_REP
SIZE
LEV
INTCOV
ROA
CFO
TANG
LOSSni
MB
BSEG
GSEG
IO_ln
HHI
RETVOL
FSD
Constant
Observations
R-squared
(1)
HARD
(2)
HARD
(3)
HARD
7.956
(8.624)
16.113
(18.245)
1.013
(0.772)
-22.310
(13.923)
-0.130**
(0.017)
-0.146**
(0.031)
-1.043**
(0.200)
-0.005
(0.004)
-0.906+
(0.471)
-0.051
(0.053)
0.123
(0.143)
-0.017
(0.070)
-0.068
(0.054)
0.002
(0.006)
-0.007
(0.006)
-0.001
(0.007)
-0.077
(0.157)
-5.931*
(2.872)
-0.039
(3.145)
3.742**
(0.462)
6.866
(8.528)
7.042
(17.002)
0.360
(0.638)
6.942
(8.583)
12.570
(17.977)
-0.130**
(0.017)
-0.144**
(0.031)
-1.026**
(0.201)
-0.005
(0.004)
-0.924+
(0.474)
-0.051
(0.052)
0.136
(0.143)
-0.023
(0.071)
-0.076
(0.054)
0.002
(0.006)
-0.007
(0.006)
-0.001
(0.007)
-0.072
(0.158)
-6.212*
(2.855)
0.066
(3.156)
3.667**
(0.460)
-12.745
(11.350)
-0.129**
(0.017)
-0.146**
(0.031)
-1.036**
(0.200)
-0.005
(0.004)
-0.943*
(0.475)
-0.053
(0.053)
0.140
(0.143)
-0.024
(0.070)
-0.070
(0.054)
0.002
(0.006)
-0.007
(0.006)
-0.001
(0.007)
-0.076
(0.157)
-6.019*
(2.884)
-0.098
(3.159)
3.837**
(0.465)
2,509
0.17
2,509
0.17
2,509
0.17
31
TABLE 6
Impact of FOG on regression of Moody's soft adjustment on qualitative aspects of disclosure
This table displays the results for the ordinary least squares (OLS) estimation where the dependent variable
is SOFT, Moody's soft adjustment. The sample consists of 3,055 firm-years (927 firms) with available
financial and adjustments data. All firm characteristics and textual variables are measured the year prior to
the measurement of Moody's ratings and adjustments. All variables are defined in the Appendix. Year fixed
effects are included and standard errors are robust and clustered by firm. Robust standard errors are
presented in brackets. ** denotes p<0.01, * denotes p<0.05, + denotes p<0.1.
Dependent variable
FOG_high
DEBT_wc
EARNINGS_wc
FLS_sc
UNCERT_wc
DEBT_wc*FOG_high
EARNINGS_wc*FOG_high
FLS_sc*FOG_high
UNCERT_wc*FOG_high
RAT_IND_ADJ
SIZE
LEV
INTCOV
ROA
CFO
TANG
LOSSni
BSEG
GSEG
IO_ln
HHI
RETVOL
MB
FSD
Constant
Observations
R-squared
(1)
SOFT
(2)
SOFT
(3)
SOFT
-0.300
(0.640)
36.536*
(15.590)
-139.374**
(43.560)
3.636*
(1.643)
-8.255
(31.456)
15.125
(19.648)
-117.873*
(49.957)
-0.947
(2.249)
67.944
(41.769)
-0.316**
(0.032)
-0.313**
(0.049)
0.034
(0.281)
-0.013*
(0.007)
0.969
(0.878)
0.051
(0.113)
-0.117
(0.216)
0.130
(0.117)
-0.025**
(0.009)
0.015+
(0.008)
-0.031*
(0.013)
0.167
(0.228)
40.487**
(5.459)
-0.247*
(0.096)
7.532
(4.944)
4.231**
(1.057)
0.077
(0.586)
36.795*
(15.422)
-145.049**
(39.128)
3.429*
(1.351)
-0.436
(0.561)
30.894*
(15.530)
-152.607**
(43.248)
-0.317**
(0.032)
-0.317**
(0.050)
0.022
(0.281)
-0.013*
(0.007)
1.011
(0.875)
0.053
(0.115)
-0.150
(0.215)
0.141
(0.117)
-0.025**
(0.009)
0.014+
(0.008)
-0.030*
(0.013)
0.167
(0.226)
40.713**
(5.423)
-0.239*
(0.093)
7.164
(4.969)
4.198**
(1.078)
54.192
(37.598)
-0.314**
(0.032)
-0.309**
(0.050)
0.055
(0.282)
-0.012+
(0.006)
0.866
(0.883)
0.047
(0.112)
-0.066
(0.213)
0.108
(0.118)
-0.026**
(0.009)
0.015+
(0.008)
-0.030*
(0.013)
0.175
(0.228)
40.161**
(5.433)
-0.251**
(0.097)
7.334
(4.923)
4.598**
(1.009)
2,509
0.19
2,509
0.19
2,509
0.18
29.015
(25.921)
19.937
(19.875)
-114.231*
(49.833)
17.962
(19.770)
-102.698*
(47.694)
0.704
(1.996)
32
TABLE 7
Impact of NT on regression of Moody's soft adjustment on qualitative aspects of disclosure
This table displays the results for the ordinary least squares (OLS) estimation where the dependent variable
is SOFT, Moody's soft adjustment. The sample consists of 3,055 firm-years (927 firms) with available
financial and adjustments data. All firm characteristics and textual variables are measured the year prior to
the measurement of Moody's ratings and adjustments. All variables are defined in the Appendix. Year fixed
effects are included and standard errors are robust and clustered by firm. Robust standard errors are
presented in brackets. ** denotes p<0.01, * denotes p<0.05, + denotes p<0.1.
Dependent variable
NT_high
DEBT_wc
EARNINGS_wc
FLS_sc
UNCERT_wc
DEBT_wc*NT_high
EARNINGS_wc*NT_high
FLS_sc*NT_high
UNCERT_wc*NT_high
RAT_IND_ADJ
SIZE
LEV
INTCOV
ROA
CFO
TANG
LOSSni
BSEG
GSEG
IO_ln
HHI
RETVOL
MB
FSD
Constant
Observations
R-squared
(1)
SOFT
(2)
SOFT
(3)
SOFT
0.414
(0.582)
27.555+
(14.873)
-166.938**
(40.692)
4.834**
(1.606)
31.656
(32.013)
38.556+
(20.112)
-80.526
(49.934)
-3.999+
(2.139)
13.487
(43.304)
-0.316**
(0.031)
-0.339**
(0.051)
0.024
(0.288)
-0.012+
(0.007)
1.058
(0.873)
0.044
(0.118)
-0.120
(0.216)
0.123
(0.118)
-0.025**
(0.009)
0.015+
(0.008)
-0.029*
(0.013)
0.158
(0.222)
39.529**
(5.345)
-0.244**
(0.094)
7.083
(4.933)
3.946**
(1.012)
0.566
(0.547)
29.062*
(14.719)
-153.086**
(36.223)
5.720**
(1.429)
-0.111
(0.520)
22.559
(14.887)
-176.258**
(40.876)
-0.316**
(0.031)
-0.340**
(0.050)
0.001
(0.287)
-0.012+
(0.007)
1.066
(0.871)
0.044
(0.118)
-0.142
(0.215)
0.136
(0.117)
-0.025**
(0.009)
0.014+
(0.008)
-0.030*
(0.013)
0.154
(0.222)
40.094**
(5.348)
-0.230*
(0.091)
6.844
(4.951)
4.002**
(1.030)
-24.733
(37.137)
-0.316**
(0.032)
-0.340**
(0.052)
0.018
(0.289)
-0.011+
(0.006)
0.962
(0.882)
0.038
(0.119)
-0.070
(0.214)
0.108
(0.118)
-0.026**
(0.009)
0.014+
(0.008)
-0.029*
(0.013)
0.158
(0.224)
39.305**
(5.328)
-0.252**
(0.095)
6.792
(4.930)
4.607**
(0.987)
2,509
0.19
2,509
0.19
2,509
0.19
77.635**
(28.620)
42.166*
(20.166)
-75.327
(49.537)
40.713*
(20.094)
-87.711+
(45.711)
-3.712*
(1.816)
33
TABLE 8
Regression of stock price response on rating change
This table displays the results for the ordinary least squares (OLS) estimation where the dependent variable is CAR, the cumulative abnormal return. The sample consists of 751 rating changes (462 firms)
representing 415 downgrades and 336 upgrades during the period from 2002 to 2012. All variables are defined in the Appendix. Standard errors are robust and clustered by firm. Robust standard errors are
presented in brackets. ** denotes p<0.01, * denotes p<0.05, + denotes p<0.1.
RCHANGE
Dependent variable
RCHANGE
(1)
all
CAR
(2)
all
CAR
(3)
all
CAR
(4)
all
CAR
-0.009**
[0.003]
-0.009**
[0.003]
-0.020+
[0.011]
-0.016
[0.015]
0.006
[0.004]
-0.030
[0.020]
-0.006
[0.008]
0.003
[0.003]
-0.001
[0.001]
-0.016
[0.026]
-0.021
[0.020]
-0.009**
[0.003]
-0.012+
[0.006]
-0.006
[0.008]
0.003
[0.003]
-0.001
[0.001]
-0.006
[0.026]
-0.020+
[0.011]
-0.005
[0.008]
0.004
[0.003]
-0.009**
[0.003]
-0.012+
[0.006]
-0.005
[0.008]
0.004
[0.003]
-0.024
[0.032]
-0.017
[0.015]
0.005
[0.004]
-0.001
[0.001]
-0.011
[0.036]
751
0.03
751
0.03
751
0.03
751
0.03
415
0.03
415
0.03
NonZeroSOFT
IGRADE
DAYS_ln
RAT_prior
Constant
Observations
R-squared
(5)
(6)
(7)
(8)
downgrade downgrade downgrade downgrade
CAR
CAR
CAR
CAR
34
(9)
upgrade
CAR
(10)
upgrade
CAR
(11)
upgrade
CAR
(12)
upgrade
CAR
-0.005
[0.009]
-0.005
[0.009]
0.006
[0.005]
-0.003
[0.003]
0.014
[0.020]
0.006
[0.005]
-0.003
[0.003]
0.000
[0.001]
0.013
[0.025]
-0.005
[0.009]
0.000
[0.004]
0.006
[0.005]
-0.003
[0.003]
-0.009
[0.032]
-0.019+
[0.011]
-0.021*
[0.011]
-0.016
[0.015]
0.005
[0.004]
-0.001
[0.001]
0.007
[0.038]
0.014
[0.021]
-0.005
[0.009]
0.000
[0.004]
0.006
[0.005]
-0.003
[0.003]
0.000
[0.001]
0.012
[0.026]
415
0.04
415
0.04
336
0.02
336
0.02
336
0.02
336
0.02
-0.019+
[0.011]
-0.021*
[0.011]
-0.015
[0.015]
0.006
[0.004]
TABLE 9
Regression of stock price response on rating change before and after Repeal of Reg FD
This table displays the results for the ordinary least squares (OLS) estimation where the dependent variable is CAR, the
cumulative abnormal return. The sample consists of 751 rating changes (462 firms) representing 415 downgrades and 336
upgrades during the period from 2002 to 2012. All variables are defined in the Appendix. Standard errors are robust and
clustered by firm. Robust standard errors are presented in brackets. ** denotes p<0.01, * denotes p<0.05, + denotes p<0.1.
RCHANGE
Dependent variable
RCHANGE
RepealRegFD
NonZeroSOFT
NonZeroSOFT*RepealRegFD
IGRADE
DAYS_ln
(1)
all
CAR
(2)
all
CAR
-0.008**
[0.003]
-0.016+
[0.009]
-0.022*
[0.009]
0.031**
[0.011]
-0.006
[0.008]
0.004
[0.003]
-0.018
[0.011]
-0.013
[0.015]
-0.030*
[0.014]
0.035+
[0.018]
-0.018
[0.015]
0.005
[0.004]
-0.014
[0.021]
-0.009**
[0.003]
-0.015
[0.009]
-0.022*
[0.009]
0.031**
[0.011]
-0.007
[0.008]
0.003
[0.003]
-0.001
[0.001]
0.001
[0.027]
751
0.04
751
0.04
RAT_prior
Constant
Observations
R-squared
(3)
(4)
downgrade downgrade
CAR
CAR
(5)
upgrade
CAR
(6)
upgrade
CAR
-0.006
[0.009]
-0.013+
[0.007]
-0.009
[0.006]
0.022**
[0.009]
0.006
[0.005]
-0.003
[0.003]
-0.004
[0.034]
-0.017
[0.011]
-0.012
[0.016]
-0.030*
[0.014]
0.035+
[0.018]
-0.020
[0.015]
0.004
[0.004]
-0.001
[0.001]
0.012
[0.039]
0.018
[0.020]
-0.006
[0.009]
-0.013+
[0.007]
-0.009
[0.006]
0.022**
[0.009]
0.007
[0.005]
-0.003
[0.003]
0.000
[0.001]
0.017
[0.025]
415
0.04
415
0.04
336
0.04
336
0.04
35