16
The above studies indicate that the relationship between firms’ decisions whether to disclose CCR and COE will likely vary depending on report users’ judgments regarding the
materiality of CCR across industries. Based on these arguments, our next hypothesis examines the role of materiality on the association between disclosing CCR and COE:
H2: The association between disclosing CCR in Form 10-K and COE is stronger for firms
where users judge such disclosures as material than for firms where users judge such disclosures as not material.
IV. RESEARCH DESIGN
Sample and Data
We obtain our sample from the intersection of the SP 500 index firms and the Ceres and CDP databases for the period 2008 to 2014. In order to minimize changes in our sample over
this period we use firms that were included in the SP 500 index on December 31, 2008. We hand-collect data on whether or not firms disclose CCR in Form 10-K from Ceres’ SEC
sustainability disclosure search tool.
14
The tool searches the text of SEC annual filings of SP 500, Russell 3000, and FT Global 500 firms and identifies the relevant issue of the disclosure
e.g., climate change risk, water risk. We choose 2008 as the initial year of our analyses because Ceres’ database provides SEC filings starting in fiscal year 2008. The last year of our analyses is
2014 because that is the last year for which CDP climate change data are publicly available.
15
To proxy for CCR materiality based on users’ judgments, following Khan et al. 2016 we use SASB’s Materiality Map
™
to identify the materiality of CCR on an industry-by-industry basis. To identify whether an issue is judged to be material for companies in a given industry,
SASB gathered input from a panel of over 200 industry experts and SASB staff. The panel scored
14
Available at http:ceres.orgresourcestoolssec-sustainability-disclosure
.
15
Although 2014 is the last year for which CDP climate change data is publicly available, we are able to obtain data for 2015 from Ceres. In robustness tests we examine the sensitivity of our results to extrapolating the firms’
participation in the CDP survey for 2015 using 2014 data.
17
sustainability issues based on three components: evidence of investor interest, evidence of financial impact, and forward-looking impact see Appendix A for further details. We use the
Materiality Map
™
scoring on two sustainability issues directly related to CCR, namely 1 GHG emissions, and 2 environmental and social impacts on assets and operations, to classify each
sample firm as in an industry where users judge CCR as either material or not material to investors.
We collect data on our sample firms’ participation in the CDP climate survey to control for voluntary disclosures of CCR information through channels other than the SEC filings. The
CDP survey elicits voluntary information on, for example, climate change risks and opportunities, carbon emissions in metric tons, emission reduction targets, and managerial
compensation. CDP does not mandate independent assurance on the data. Table 1 provides our sample selection criteria. We start with all SP 500 firms available
in the Ceres and CDP databases from 2008 to 2014. The result is 3,226 firm-year observations 496 unique firms. We lose 227 firm-year observations for which we are unable to calculate our
COE measure. This is because we exclude firms with negative book value of equity or negative earnings forecasts for years one and two, or we are unable to obtain analyst forecasts for these
firms. The sample is further reduced by three observations for unavailable Compustat data, resulting in a final sample of 2,996 firm-year observations 465 unique firms for Hypothesis 1
tests. We further exclude 49 firm-year observations that are missing a 4-digit SIC code needed to match with SASB’s industry codes for user-based materiality classification. Thus, our final
sample for Hypothesis 2 tests consists of 2,947 firm-year observations 458 unique firms. | Insert Table 1 about here |
18
Descriptive Statistics
Figure 2, Panel A provides the percentage of firms that disclosed CCR information and the percentage of firms that participated in the CDP climate survey from 2008 to 2014. In 2008,
less than half of the firms 46.2 percent disclosed CCR information. Notably, that percentage increased almost ten percentage points from 2008 to 2009, and a further five percentage points in
2010. These increases make intuitive sense, as the years coincide with the issuance of the SEC’s 2010 interpretive guidance on climate change disclosures, which became effective in February
2010. There was a steady growth in the percentage of firms disclosing CCR until 2012, but the growth then tapers off. In 2014, almost two-thirds of the firms disclosed CCR information.
Figure 2, Panel A also shows growth in the firms’ participation in the CDP climate survey, from 58 percent in 2008 to 67 percent in 2014.
Figure 2, Panel B shows percentages of firms that disclosed CCR, partitioned by user- based SASB materiality. From 2008 to 2014, the number of CCR disclosers for both the
material and not-material groups grew by 20 percentage points. Over the same period, the percentages of CCR disclosers are consistently higher, by an average of 20 percentage points, for
the material-CCR group relative to the not-material-CCR group. | Insert Figure 2 about here |
Panel A of Table 2 shows that, averaged over our sample period, 60 percent of the firms disclosed CCR and 65.2 percent participated voluntarily in the CDP climate survey. Further,
while 40.5 percent both responded to the CDP climate survey and disclosed CCR cell 4, about 15 percent neither responded to the CDP climate survey nor disclosed CCR cell 1. Notably,
almost 25 percent responded to the CDP climate survey but chose to not disclose CCR cell 3.
16
16
The null hypothesis of independence between disclosing CCR information in Form 10-K and participation in the CDP climate survey is rejected Chi-square = 9.634; p 0.01.
19
This is counter-intuitive, since these firms have voluntarily committed scarce resources to respond to the CDP survey and some firms also provided independent assurance on this
information, and yet they chose to not disclose CCR. Panel B of Table 2 shows materiality as judged by users i.e., based on SASB’s panel of experts, partitioned by whether firms disclosed
CCR. Averaged over the seven-year period, the majority of firms in our sample 61.4 percent belong to industries where users judge CCR as not material.
| Insert Table 2 about here | Table 3 shows the sub-samples of firms that participated in the CDP climate survey
Panel A and those that did not participate Panel B, partitioned by users’ materiality judgments and by whether firms disclosed CCR. Notably, both panels show that the majority of firms
disclosed CCR, regardless of whether they participated in the CDP climate survey. | Insert Table 3 about here |
Empirical Models and Variable Definitions
As discussed earlier, using data from disclosing firms to draw inferences about non- disclosing firms without adjusting for the systematic differences between them can give rise to
biased coefficients, and thus, erroneous conclusions. This is likely the case in Griffin, Lont, and Sun 2017, who use the GHG emissions of the disclosing firms to estimate the GHG emissions
of the non-disclosing firms. This method incorrectly treats the non-disclosing firms as if they were identical to the disclosing firms, thus assuming away self-selection. Thus, Griffin et al.’s
puzzling finding—the market penalizes firms that choose to voluntarily disclose their GHG emissions—runs counter to both economic and voluntary disclosure theories and leaves
unanswered the question, “why would firms choose to voluntarily disclose their GHG emissions if they are penalized by the market for the act of disclosing?”
20
Following Matsumura et al. 2014, we correct for self-selection using propensity score matching to compare the COE of the firms that disclose CCR with the COE of non-disclosing
firms H1. Further, we examine the COE differences between disclosers and non-disclosers after partitioning by user-based CCR materiality judgments based on SASB’s panel of experts H2.
Implied COE is the internal rate of return that equates the current stock price to the present value of all expected future cash flows to equity. This rate is an ex-ante estimate of the COE, given
market expectations about future growth. Specifically, the value of the firm at time t is expressed as:
1 where P
t
is the market value of common equity on the date of the earnings forecast at time t from the daily CRSP files, FCFE
t+i
is free cash flow to equity at time t + i, and r
e
is the implied COE. We rely on prior accounting and finance research e.g., Hail and Leuz 2009; Hann,
Ogneva, and Ozbas 2013 to estimate the implied COE. COE, our measure of implied COE, is a composite COE constructed using the median of four measures: Easton’s 2004 price earnings
growth PEG model, Gebhardt et al. 2001 GLS, Claus and Thomas 2001 CT, and the price-earnings ratio.
17
The four models differ in the assumptions made to forecast expected
future cash flows. We follow Hann et al. 2013 in operationalizing these models.
Following prior research, we use median analyst forecasts as our proxy for FCFE. Analyst forecasts for year 1 correspond to the fiscal year that ends after the forecast date. That is,
if the first-year analyst forecast year 1 in IBES is for the previous year because the earnings
17
This is consistent with prior research that aggregates various measures to calculate a composite COE measure see, e.g., Hail and Leuz 2009. Aggregating across measures reduces the idiosyncratic errors that may be present
in any single measure.
21
for the previous year have not yet been announced, we do not use that forecast. Instead, we use the second-year forecast, which is the forecast for the current fiscal year, as the year 1 forecast.
In residual earnings models such as CT and GLS that require an estimate of book value, we use the book value at the end of the prior year beginning of current year. Since the first forecast is
only for part of the year, we discount only for the proportionate number of days remaining through the year end.
Prior studies retain only one earnings forecast per year e.g., Hail and Leuz 2009; Hann et al. 2013. Unlike these other studies, we retain all earnings forecasts made during the year for
each firm to calculate the COE numbers used in our composite measure. Prior research shows that analyst forecasts tend to exhibit an upward bias earlier in the fiscal year, but are then revised
downwards over the year, and finally exhibit a downward bias at earnings announcement Richardson, Teoh, and Wysocki 2004. Such biases in analyst forecasts can lead to systematic
biases in COE calculations Easton and Sommers 2007. Using all available forecasts reduces this bias as well as any errors that may arise in the COE measure from errors in the retained
forecast. We take the median of all the COE numbers for each measure for each firm-year. We then take the median across the four measures to calculate our composite COE measure for each
fiscal year for each firm. In robustness analyses we also aggregate the four measures using their means, rather than medians, to calculate the composite COE.
Propensity Score Matching
We use propensity score matching Rosenbaum 2005 to compare the COE of the firms that disclose CCR with the COE of the non-disclosing firms. We use the probit model in
Equation 1 to calculate the propensity scores:
22
DISC_10K =
+
1
BETA +
2
BM +
3
SIZE +
4
FIPI +
5
ROA +
6
EXCH +
7
STRNG +
8
CNCRN +
9
CDP +
1 where DISC_10K is an indicator variable that is coded 1 if the firm discloses CCR information in
Form10-K in year t, and 0 otherwise. All independent variables, discussed below, are measured contemporaneously.
We match the disclosers with the non-disclosers on the Fama-French three factors: market beta BETA, book-to-market ratio BM, and firm size SIZE. BETA is the correlation
between firm-specific returns and market returns. We use monthly returns for the firm and the CRSP value-weighted index for the market returns. We calculate betas using returns for the five
years prior to and including fiscal year t, but require a minimum of ten months of data. For firm- years with fewer than ten months of data, we substitute the mean beta for the firm as the beta for
that fiscal year.
18
Following Francis, Nanda, and Olsson 2008, we predict a positive association between BETA and DISC_10K. We control for firm growth by including the firm’s book-to-
market ratio BM, measured as the book value of common equity divided by the market value of common equity at the end of the fiscal year. Because larger firms are more likely to provide
more environmental disclosures Stanny 2013; Matsumura et al. 2014, we include the log of firms’ total assets as our proxy for SIZE.
International product market interactions affect environmental disclosures Matsumura et al. 2014; Khanna, Palepu, and Srinivasan 2004; Stanny and Ely 2008, and EU firms with higher
proportions of international sales are more likely to provide CCR disclosures. Therefore, to control for international product market interactions, we include annual pre-tax foreign income
as a proportion of total pre-tax income FIPI and expect a positive coefficient for this variable.
18
We make this substitution for fewer than 15 firm-years less than 0.5 percent of our sample.
23
Consistent with prior research that documents a positive association between firm performance and disclosures e.g., Miller 2002, we expect a positive coefficient on our measure of firm
performance, ROA, measured as income before extraordinary items divided by total assets. Firms choose the exchange on which to list their securities and this choice is a function of
both firm-level characteristics and the exchange’s listing requirements, including disclosure requirements Corwin and Harris 2001. Therefore, we also match firms on the stock exchange
EXCH on which they trade. In general, larger and older firms are more likely to list on the NYSE, but since the vast majority of our sample firms 80 percent are listed on the NYSE the
remaining firms trade on NASDAQ and are likely to be among the largest global firms, we do not predict a sign on EXCH.
Empirical evidence indicates that firms that are more environmentally proactive are more likely to disclose environmental information Matsumura et al. 2014. Thus, similar to
Matsumura et al. 2014, we control for the firms’ environmentally proactive performance ratings, measured as STRNG, and for their environmentally damaging actions ratings, measured
as CNCRN, to proxy for the firms’ environmental performance. We collect environmental performance ratings data using the KLD database. Consistent with prior research Cho et al.
2012; Matsumura et al. 2014 we do not aggregate STRNG and CNCRN because KLD’s proactive dimensions are distinct from the damaging dimensions. Similar to Matsumura et al.
2014, we expect a positive coefficient for STRNG, and do not predict a sign for CNCRN. If the KLD score is missing for an observation, we set it equal to zero.
To address the possibility that firms may be providing CCR information through channels other than Form 10-K, we include an indicator variable, CDP. If, according to CDP, the firm
participated in the CDP climate survey and the response is publicly available in that year i.e.,
24
CDP response status is AQ, AQL, or AQSA, then we code the firm as 1, and 0 otherwise.
19
To test H2, we classify each sample firm as in an industry where users i.e., SASB’s panel of experts judge CCR as either material coded = 1 or not material coded = 0.
V. RESULTS
