Voluntary Disclosure and the Cost of Capital Evidence from Firms

Voluntary Disclosure and the Cost of Capital: Evidence from Firms’ Responses to the Enron Shock Christian Leuz The Graduate School of Business, University of Chicago and Catherine Schrand The Wharton School, University of Pennsylvania Preliminary August 2008 Abstract This paper examines the link between voluntary disclosure and the cost of capital. We exploit an exogenous cost of capital shock that the Enron scandal in fall 2001 created for other U.S. firms and analyze whether and how firms respond to this shock. These tests are opposite to the typical research design that analyzes cost of capital responses to disclosure changes, thus mitigating endogeneity concerns in traditional cross-sectional disclosure studies. We use a unique method to estimate betas that allows for exogenous shocks to the firm’s cost of capital as well as changes in systematic risk due to endogenous disclosure responses to the transparency crisis. Our analysis shows that beta shocks around the Enron scandal are associated with an increase in voluntary disclosures in mandatory annual reports. Firms extend the number of pages in their annual 10-K filings, notably the section containing financial statements and footnotes. The increase in disclosure is particularly pronounced for firms that have positive shocks and large financing needs and growth opportunities. We find evidence suggesting that interim disclosures, such as earnings announcements, conference calls, and 8-K filings are complementary to the increase in disclosure in the 10-K. We also show that firms’ disclosure responses are effective in reducing the impact of the transparency crisis on U.S. firms’ costs of capital. JEL classification: Key Words: G14, G15, G30, K22, M41, M42 Transparency, Disclosure, Cost of capital, Capital market shocks, Crises, Annual reports, Accounting scandals 1. Introduction The link between voluntary disclosure of financial information and firms’ cost of capital is one of the most fundamental relations in academic accounting and finance, and understanding it is of substantial interest to firms, who provide information to capital markets, as well as to financial market regulators. Various theoretical models predict that an increase in a firm’s commitment to disclosure is negatively related to its cost of capital (e.g., Verrecchia, 2001; Lambert et al., 2007). Despite its importance and plausibility, the nature of this link is still an open question. Much of the empirical literature provides cross-sectional evidence that firms with more extensive voluntary disclosures exhibit less information asymmetry and have a lower cost of capital.1 However, there are substantial concerns about whether this relation can be interpreted in a causal way. Firms likely choose disclosures with the effect on their cost of capital in mind, creating an endogeneity problem for which it is difficult to find valid instruments. An alternative identification strategy is to use a natural experiment. This paper takes that route. The events surrounding the collapse of Enron led to substantial concerns about the transparency of financial statements and disclosures of U.S. firms. If disclosure and the cost of capital are related as predicted by economic theory, these concerns about the lack of transparency should increase firms’ costs of capital. This information-related cost of capital shock in turn should trigger firms to re-evaluate their disclosure policies and lead to changes in their disclosure behavior. By relating cost of capital shocks to subsequent disclosure responses, we in essence conduct the reverse experiment to the prior literature. Thus, we exploit the Enron debacle as an exogenous shock for other U.S. firms and provide new and complementary See, e.g., Botosan (1997), Healy et al. (1999), Leuz and Verrecchia (2000), Botosan and Plumlee (2002), Hail (2003), Schrand and Verrecchia (2005). In addition, there are studies on the cross-sectional relation between accruals quality and the cost of capital (e.g., Francis et al., 2004 and 2005). 1 -1- evidence that disclosure and the cost of capital are indeed related as predicted by economic theory. Our analysis focuses on the period when the news about Enron’s losses and accounting irregularities hit the market in Fall 2001. The news quickly grew into widespread concerns about U.S. accounting practices and, in particular, about firms’ disclosures of special purpose vehicles, off-balance sheet financing, and related-party transactions (see Appendix). The Enron-specific events created wider concerns about transparency of U.S. financial statements in general. The events culminated in passage of the Sarbanes-Oxley Act in August 2002. Four features make the news events surrounding Enron’s collapse a powerful setting to address the broad question of the relation between disclosure and cost of capital. First, the shock caused investor concerns about a systematic lack of transparency in financial reporting. Although the shock also likely had an impact on cash flows, specifically for firms in certain industries, for most firms it was primarily an information-related shock. Hence, a disclosure response seems appropriate. Second, the shock was triggered by Enron-specific events and hence is exogenous to other U.S. firms.2 Third, the shock occurred during a relatively short window. Finally, it occurred during the fourth quarter of 2001. Thus, firms had the opportunity to respond in annual financial reporting. We measure shocks to firms’ costs of capital around the Enron events using an econometric technique suggested by Lockwood and Kadiyala (1988) that allows for a quadratic shape to the beta estimate during the event window as the transparency crisis was unfolding. The quadratic form is important because it can accommodate the impact of a firm’s endogenous interim disclosure responses on its cost of capital during the event period. In using this The shock may not be exogenous to other firms in the energy industry. We therefore check that our results are not driven by or sensitive to the inclusion of energy firms. 2 -2- technique, we also attempt to separate shocks to the cost of capital from changes in unsystematic risk or updated assessments of firms’ future cash flows. The main analysis examines firms’ disclosure responses to the information-related cost of capital shock. We analyze responses in firms’ mandatory annual SEC filings (Form 10-K), earnings announcements, and 8-K filings. We also examine the impact of other potentially complementary interim disclosures including conference calls. Our sample comprises 2,172 U.S. firms with December fiscal-year ends and the required financial data from 1999 to 2001. Using this sample, we document that the cost of capital shocks are associated with an increase in the firms’ disclosures in their subsequent annual 10-K filings. Firms extend the number of pages in their 10-K filings, notably the section containing financial statements and footnotes. The increase in disclosure is particularly pronounced for firms that experience positive beta shocks and are likely to be more sensitive to their cost of capital because they have larger financing needs and more growth opportunities. This link between cost of capital shocks and voluntary disclosure responses is robust to a broad set of alternative variable and model specifications. We do not find a significant relation between the beta shocks and changes in the length of firms’ annual earnings announcements. However, there is a positive relation between the beta shocks and 8-K filings. An analysis of interim disclosures after the shock, including 8-K filings and conference calls suggest that firms used such interim disclosures in response to the crisis, and the interim disclosures mitigated the effects of the shock. However, accounting for such interim disclosures does not eliminate the relation between the cost of capital shocks and disclosure in the 10-K. Thus, firms appear to view the mandatory 10-K filing and interim disclosures as complementary activities to reduce the transparency problems during this time period. -3- These results contribute to the substantial prior literature on voluntary disclosures and firms’ cost of capital by explicitly linking information-related cost of capital shocks to firms’ disclosure responses. In doing so, we complement prior studies and increase the confidence in evidence on the link between voluntary disclosure and firms’ cost of capital. Prior studies are generally based on cross-sectional analyses of firms’ disclosure choices and their cost of capital (see survey by Leuz and Wysocki, 2007). A few studies examine changes in firms’ disclosure policies; they typically presume that the changes in disclosure occur as a function of changes in firms’ financing needs or growth opportunities (e.g., Healy et al., 1999; Leuz and Verrecchia, 2000). Some studies exploit firms’ disclosure choices in situations where they access capital markets, essentially assuming that cost of capital benefits drive their choices (e.g., Lang and Lundholm, 2000; Schrand and Verrecchia, 2005). But none of these studies identifies an exogenous event that changes the disclosure/cost of capital tradeoff. The paper is organized as follows. Section 2 summarizes the events in Fall 2001 and delineates our predictions. In section 3, we define our disclosure and cost of capital proxies. Sections 4 and 5 present the results of our analyses on the relation between the cost of capital shocks and firms’ subsequent disclosure responses. Section 6 presents the analysis of cost of capital responses to these disclosure responses. Section 7 concludes the paper. 2. Hypothesis development and research design On October 16, 2001, Enron announced a third quarter loss of $618 million and hinted at problems with its partnerships. The announcement raised concerns about conflicts of interests and about the quality of its financial reporting. On November 29, 2001, Enron announced it would restate its earnings back to 1997 and add $628 million of debt to its 2000 balance sheet. -4- Between these events, the SEC launched an inquiry into Enron’s accounting practices and then initiated a formal investigation that included Enron’s auditors. Subsequently, Enron was removed from the S&P 500, its debt was downgraded to junk and, eventually, it filed for bankruptcy. We provide more details on the sequence of events in Appendix A.1. The news of Enron’s financial problems came as a surprise to the market. Subsequently, the financial reporting system was blamed, at least in part, for allowing Enron to hide its financial condition, in particular its financial obligations, through the use of special purpose vehicles and partnerships. Enron became one of the largest financial scandals in the history of the U.S. and led to what has been described as a “transparency crisis” that quickly spread to other firms in the economy. Appendix A.1 identifies a number of articles in the financial press that suggest that the market viewed the Enron scandal as indicative of widespread corporate transparency and governance problems in the U.S. rather than an Enron-specific event. We argue that this transparency crisis created an exogenous cost of capital shock for U.S. firms. We analyze whether firms change their disclosure behavior in response to this information-related shock. The analysis is a specification in changes, like many studies of the relation between disclosure and the cost of capital. However, most disclosure studies specified in changes examine the impact of a change in disclosure policy on cost of capital. Our analysis assumes that firms were engaging in an optimal equilibrium disclosure strategy prior to the shock, and examines the effects of an exogenous shock on the disclosure decision. There are several theories that link voluntary corporate disclosures and firms’ cost of capital (e.g., Verrecchia, 2001; Leuz and Wysocki, 2007). One strand of literature that provides a direct link to firms’ beta factors is based on the idea of estimation risk. It starts from the premise that important parameters of the distribution of stock returns, like a firm’s beta factor, -5- have to be estimated from a historical time-series of returns, and then analyzes the role of information in the estimation (e.g., Brown, 1979; Barry and Brown, 1984 and 1985; Coles and Loewenstein, 1988). More recently, Jorgensen and Kirschenheiter (2003), Hughes et al. (2007), and Lambert et al. (2007) re-examine the issue of estimation risk using information structures that lend themselves more naturally to the link between firm-specific disclosures and the cost of capital. For instance, Lambert et al. (2007) model estimation risk using an approach that can accommodate firm-specific disclosures that are informative about firms’ expected future cash flows. Based on this information structure, they show that the assessed covariances of a firm’s cash flows with the cash flows of other firms decrease as the quality of the firm’s disclosure increases. This effect unambiguously moves the firm’s cost of capital closer to the risk-free rate, reducing the risk premium to risk-averse investors. Lambert et al. (2007) also show that this information effect is not diversifiable because it is present for all covariance terms with other firms. Only the effect of disclosure on the firm-specific variance is likely to be diversifiable when investors can form portfolios of many stocks. The covariance effects, however, should manifest themselves in firms’ beta factors as well as the market risk premium for the economy.3 These models imply that an exogenous shock to the market’s assessment of the quality of firms’ accounting information and financial disclosures should lead to cost of capital shocks. Assuming that firms were optimizing their disclosure policies with respect to the cost of capital prior to the shock, the models further imply that, ceteris paribus, firms’ optimal disclosure policies change as a result of the shock.4 Moreover, firms that are more sensitive to their cost of We focus on shocks to firms’ beta factors as our analysis is cross-sectional in nature. Shocks to the equity premium apply to all firms and hence simply exacerbate or dampen shocks to the beta factor. Besides, it is difficult to estimate the equity risk premium or changes therein over short time periods. 4 The notion that the Enron collapse led to revised estimates of firms’ cost of capital does not imply that markets were inefficient before. The events at Enron rationally changed investors’ assessment of the cost of capital. Furthermore, it is not necessary for our analysis that the response to Enron is rational. An irrational market reaction, which one could argue is even more exogenous, is still likely to trigger a disclosure response. 3 -6- capital because they have greater external financing needs and growth opportunities are expected to increase their disclosures more than firms that are less sensitive. Although not necessarily predicted by the theories, we predict that the response to cost of capital shocks is asymmetric in the sense that only firms with a positive shock are likely to respond by increasing their disclosure. This prediction assumes that the cost of retaining disclosures is minimal. In fact, the litigation environment may make it costly to reduce disclosure from period to period. Hence, we predict no disclosure change in response to a negative cost of capital shock. There are several reasons that the predicted relations may not be borne out by the data. First, it is possible that the Enron crisis did not result in a shock to firms’ cost of capital, despite widespread claims in the press and market participants at the time. Second, the link between accounting information and the cost of capital may not exist in the way predicted by theory. Third, it is conceivable that firms are unable to address the transparency concerns via increased voluntary disclosures. The following stylized timeline is used to illustrate our specific predictions about the impact of the transparency crisis on firms’ cost of capital at various points in time and the timing of firms’ disclosure responses: Pre-event-period Event period ↑ ↑ ↑ ↑ ↑ Exogenous shock occurs Pre-report period Post-report period ↑ Release of mandatory accounting report We assume that, during the pre-event period, firms’ choices of their disclosure policy are optimal with respect to their cost of capital implications. During the event period, an exogenous information-related shock occurs that affects investors’ assessments of the precision of firms’ financial reports. This shock in turn changes the firm’s cost of capital. The event period is more -7- than a month (details to follow) as news about the extent of the problems was unfolding. In addition, during this period, a firm may have responded to the exogenous shock by changing its disclosure activities. Responses could be disclosures, such as conference calls or press releases, or costly financial signals, such as dividends or stock repurchases. Thus, the event period represents a period of adjustment, in which the firm moves from one cost of capital equilibrium to (possibly) another. The pre-report period is the time after the event (or shock) has taken place but before the release of mandatory reports, such as the earnings announcement or the 10-K filing. The cost of capital during this time period reflects the Enron shock as well as actions that the firm has taken during the event period, if any, to mitigate the shock. It can be viewed as the new equilibrium level of a firm’s cost of capital prior to the release of its mandatory accounting reports. The earnings announcement and the annual report are another opportunity for the firm to respond to the event in addition to more immediate disclosure responses during the event period. We consider seven independent window specifications to define the pre-event period, the event period, and the pre-report period (see Appendix A.2). In all cases, the pre-event window starts on May 1, 2001, which for firms with December fiscal-year ends is after the year 2000 annual report season, and runs through August 31, 2001. This period precedes the market closings caused by the terrorist attacks on September 11. The pre-report period ends either on January 15, 2002 or January 28, 2002, prior to the annual earnings announcement season. Choosing the later date increases the length of the pre-report period and hence the power to measure systematic risk during the period, but it is more likely that the estimate includes the effects of early earnings announcements. Similarly, the choice among the different windows involves trading off a longer event period, which allows us to more precisely measure the effects -8- of the Enron events, against a shorter pre-report period, which reduces the power of our estimation procedure during this period. Throughout the paper, we present results for Window 1, which is centered on November 8 and has an event period length of 39 days. The event period ends on January 15, 2002. The Appendix provides evidence to justify this choice. The results are generally robust to alternative windows, in particular those that maximize the length of the period over which the respective betas are estimated. The post-report period is consistent across all analyses and is defined as May 1, 2002 to August 31, 2002. Based on the time line illustrated above, we make the following predictions about firms’ disclosure responses to the exogenous transparency shock associated with Enron. First, we expect firms with larger positive shocks to increase disclosure more in their mandatory reports. It is possible that firms respond to the shock immediately (i.e., during the event period) and provide interim disclosures, and if effective, do not respond further in the mandatory report. Ex ante, however, it seems plausible that interim disclosures may not be sufficient to completely eliminate the transparency issues or may require confirmation in 10-K filings because these annual reports are audited and reviewed by the SEC. We also predict a positive relation between the cost of capital shocks that remain prior to the release of mandatory reports and firms’ disclosures in mandatory reports. This prediction accounts for the possibility that firms respond to the shock immediately during the event period with voluntary interim disclosures (or costly signals), and that the interim disclosures can mitigate the effect of the shock. If interim actions are fully credible and restore the equilibrium cost of capital, we may not observe an association between the remaining cost of capital shock in the pre-report period and firms’ responses in the mandatory reports. -9- Finally, we predict that increases in disclosure in the mandatory report will decrease the net effect of the shock. This prediction is similar to tested predictions in traditional voluntary disclosure studies that examine the impact of a firm’s disclosure activity on its cost of capital, but the Enron setting may offer an advantage in terms of power to detect an effect of disclosure in the 10-K, given the magnitude of the information-related problems in this environment. Although these tests for the effectiveness of the disclosures complete the analysis, we note that it is not necessary that we observe a cost of capital reaction to the changes in disclosure in order to justify managers’ decisions to increase disclosure. We may observe no reaction to the increased disclosure if capital markets did not view the information as credible, or if the disclosures did not reduce concerns over transparency about value-relevant information. Firms may have increased “irrelevant” disclosures if they believed that providing them would mitigate litigation risk (Rogers, 2005). 3. Empirical proxies and descriptive statistics The sample includes all firms on Compustat that have fiscal year ends on December 31st and non-missing values for total assets and earnings announcement dates in each of the three years from 1999 – 2001. We delete REITS, Limited Partnerships, Trusts and Funds, but do not apply further industry filters. Data for 1999 and 2000 are required to create control variables and also benchmarks to measure changes in disclosure. The final sample has 2,172 firm observations for which we have return data to compute the cost of capital shocks, Compustat data to compute the primary control variables, and data to create the disclosure proxies. We use variations of the following cross-sectional model to test our main hypotheses about the impact of a cost of capital shock on disclosure (firm subscripts omitted): -10- Change in disclosure = f(Cost of capital shock) + control variables + ε The dependent variables are proxies for changes in disclosure choices in mandatory reports, including 10-K filings and earnings announcements, associated with the year ended December 31, 2001, and 8-K filings during the event period. The proxies are described in Section 3.1. The cost of capital proxies are measured in levels or changes at various points in time during the window outlined above. These proxies are described in Section 3.2. 3.1 Accounting disclosures (1) We examine changes in disclosure in three mandatory accounting reports: the December 31, 2001 annual report/10-K, the fourth quarter 2001 earnings announcement, and 8-K filings. The proxy for a change in the disclosure behavior in the 2001 10-K is the percent change in the page count from the December 31, 2000 10-K to the December 31, 2001 10-K (% ΔPAGES ). Page count data are from the Global Securities Information (GSI) database, which covers all SEC filings that are available in the EDGAR database. We eliminate all filings with less than 10 pages. Most of these observations are references to a fuller document and not the page count of the 10-K. We eliminate observations that switch from (to) filing a 10-K to (from) a 10-KSB. The page counts are for the body of the 10-K (i.e., Items 1 through 15), which includes the exhibits and financial statement schedules in Item 15. The page counts exclude separate exhibits and appendices to the 10-K filing beyond Item 15. While those exhibits and appendices could include useful information, such as material contracts, these exhibits are not sufficiently standardized (in format and content) such that variation in page counts would provide a meaningful proxy for variation in information content. We also measure page count changes for three sections within the 10-K. For all three variables, we exclude observations that appear to have rearranged the location of information -11- within the 10-K.5 The variable %ΔMDA is the percent change in pages for item 7, Management’s Discussion and Analysis of Results of Operations and Financial Condition (MD&A). The variable %ΔFS is the percent change in the pages of 10-K sections that commonly include financial tables. The sections used to create this proxy are Item 6 (Selected Financial Data), Item 8 (Financial Statements and Supplementary Data), and Item 15 (Exhibits, Financial Statement Schedules, and Reports on Form 8-K). The variable %ΔDISCUSS is the percent change in the pages of 10-K sections that commonly include narrative discussions of the type of information we expect to be informative in response to a transparency crisis. The sections used to create this proxy are Item 1 (Business), Item 2 (Properties), and Item 7 (MD&A). These two proxies ( %ΔFS and %ΔDISCUSS ) address concerns that the page changes for the entire 10-K occur in less important parts and should give us a sense for the type of disclosure responses that firms choose. In combining multiple items, these two proxies mitigate problems associated with small numbers of pages in the individual sections and with rearranging the location of information within the 10-K. For these reasons, we do not present results for changes in pages of other individual sections of the 10-K (e.g., Item 13 Related Parties), even though ex ante they might seem relevant, because the number of pages of these sections is typically too small to create meaningful page proxies. For all of the percent change variables, the benchmark is the page count in the prior year. We create an alternative proxy that uses the average of the 1999 and 2000 page counts. Results are robust to this alternative specification and are not presented. We set the percentage change Observations that report an item in the current year that had been incorporated by reference in the previous year, or vice versa, are eliminated under this criterion. We identified observations that potentially rearranged the 10-K using various data filters. For these observations, we examined the 10-K to verify that disclosures within the 10-K were rearranged across various sections. 5 -12- variables equal to missing when the percentage change is greater than the 99th percentile or less than the first percentile. Winsorizing rather than truncating the variables yields similar results. [INSERT TABLE 1 HERE.] Table 1, Panel A reports descriptive statistics for the page count variables. The average 10-K in 2001 has 65 pages ( PAGES 2001 ), which is a 17% increase over 2000. This increase is statistically different from zero below the 1% level and it exceeds the average page change from 1999 to 2000, which equals only 3.4%. The difference between the increase from 1999 to 2000 and from 2000 to 2001 is statistically significant below the 1% level. These findings suggest that firms significantly expanded their 10-K filings in 2001 beyond the normal rate. The MD&A section in 2001 has on average 13.8 pages, which is a 38% increase over 2000. The section with narrative discussions (financial tables) has on average 27 (35) pages, which represents an increase of 22% (17%) over year 2000. The MD&A, narrative, and financial sections are longer than in 2000 for over 75% of the sample observations. The proxy for a disclosure change in the fiscal 2001 earnings announcement is the change in the number of words relative to the fiscal 2000 earnings announcement (%ΔWORDS). The number of words in each announcement is determined based on a manual search of earnings announcements for our sample firms’ on Dow Jones Interactive. We use only announcements on Business Wire or PR Newswire, as these two sources furnish unmodified press releases.6 We delete extreme changes in the word count that are in the 1st and 99th percentile. We eliminate firms if the earnings announcement in any of the three years was within 30 days of the start of trading for the firm on CRSP. These observations are likely the IPO announcement. 6 -13- Table 1 reports that the sample firms have an average of 2,077 words in their earnings announcements (WORDS2001), which is a 16.9% increase over the number of words in the earnings announcement around the same time in the previous year. There is more cross-sectional variation in the percent increase in words than in the percent changes in the page count variables, as indicated by the greater coefficient of variation and the inter-quartile range. The third disclosure mechanism we consider is 8-K filings. Firms are required to use the 8-K form to notify investors of any unscheduled material event that is important to shareholders or the SEC. The SEC defines required reportable events (e.g., shutting down a plant or certain executive changes), and it also requires more generally that firms use the 8-K form to update any information in previous SEC filings such as 10-Ks or 10-Qs. Abnormal returns around 8-K filing dates for reportable events suggest that the filings are informative (e.g., Beneish, Hopkins, Jansen, and Martin, 2005).7 The proxy for a disclosure change in 8-K filings is the change in the count of filings during the period Nov 1, 2001 to April 30, 2002 relative to the same period one year earlier (Δ8KCOUNT). The count excludes filings related to the impact on the firm of the September 11, 2001 terrorist attacks. %Δ8KCOUNT represents the percentage change in this variable. Table 1 reports that the sample firms filed 1.475 8-Ks on average during the period. The median number of filings was zero; the 75th percentile is two. Excluding filings related to the September 11th Anecdotal evidence suggests that some firms used the form 8-K to make transparency-related disclosures. For example, on December 5, 2001, American Express filed an 8-K indicating that it was updating its Regulation FD Disclosure, which simply reported: “In view of the situation involving Enron, the company is analyzing its exposure, but preliminarily believes any impact will not be material.” Another 8-K followed on February 6, 2002, that stated: “First, in terms of the financial impact of Enron's bankruptcy, back in early December we filed an 8-K indicating that we preliminarily believed the impact from Enron would not be material. Having now completed our review, this statement is still true. Second, we have also reviewed our business and accounting practices in light of Enron's recent issues. Based on this review, we can see no parallels between our businesses and what we understand the practices were at Enron. Here are a few specifics.” The 8-K continues with a discussion of special purpose entities and off-balance-sheet financing. 7 -14- attacks, the 75th percentile is one. The average percentage increase in 8-K filings (excluding filings related to September 11th) is 7.8%. Table 1, Panel B provides a correlation matrix for the page count variables, word count variables, and proxies for firm characteristics that are commonly associated with disclosure levels. The correlations between page counts in the 10-K for 2001 and word counts in the fiscal 2001 earnings announcement vary from 26% for financial sections of the 10-K to 37% for narrative sections of the 10-K. Both disclosure variables have statistically significant correlations with various firm characteristics and the sign of the correlation is generally plausible, although the magnitude of the correlations is sometimes low. The most significant correlations are between the page count, financial leverage and firm size variables. ROA and lagged stock returns are negatively correlated with page counts. For the count of words in the earnings announcement, the most significant positive correlations are with firm size, analyst following and trading volume. 3.2 Cost of capital shocks To construct measures of systematic risk in the pre-event period, the event period, and the post-event, pre-report period, we employ a regime-switching model suggested by Lockwood and Kadiyala (1988) to estimate event-induced changes in systematic risk. The method, which nests traditional methods, permits systematic risk to change gradually during the event period and exit the period at higher or lower levels than the pre-event period. It provides period-specific beta estimates, which we use to construct cost of capital shocks during the event period and pre-report period. Following Lockwood and Kadiyala (1988) and Cyree and DeGennaro (2002), the model specification is: -15- Rit = ai + bit Rmt + ε it bit = bi1 + bi 2 (T 1 − t )(t − T 2) D1it + bi 3 [(T 2 − T 1) D 2 it + (t − T 1) D1it ] (B1) (B2) where Rit is the daily holding period return for firm i from CRSP on day t (including distributions) and Rmt is the value-weighted return on the market from CRSP for day t. Systematic risk varies as a function of the trading day t (equation B2). T1 and T2 define the event period. They are specified as the number of trading days in the event period relative to day t. For example, our event period is from October 15 through December 5 (inclusive), which contains 39 (CRSP) trading days. Returns observations on November 8 (the center trading day) are assigned a date of t = 0, T1 is set at -19 and T2 is set at 19. D1 and D2 are indicator variables that equal one when the return observation is for a day t that is during the event period or during the pre-report period, respectively.8 From the model parameters, we create separate cost of capital measures for the pre-event, event, and pre-report periods. First, systematic risk for the pre-event period (BETA_PRE) is ˆ equal to the parameter bi1 . This parameter is a constant throughout the pre-event period. Consistent with the assumption that the pre-event period is one during which firms were in 8 The event window includes the third quarter earnings announcement period. We repeat the estimation excluding ˆ ˆ return observations on the earnings announcement date and the previous day. The estimates of b and b , which i1 i3 determine our estimates of the pre-event betas (BETA_PRE) and pre-report betas (BETA_REM), are virtually identical. Since most third quarter earnings announcements occur in mid-to late October, which coincides with the ˆ event period, the exclusion of returns on these days has a bigger impact on b . However, the impact is economically small (at the 5th digit) and the estimates are not significantly different. The results presented in the paper are based on the estimates that include the returns on the third quarter earnings announcement date. i2 -16- equilibrium, this estimate serves as a benchmark for the firm’s beta prior to the transparency shock. During the event period, when D1 = 1 and D2 = 0, systematic risk is determined by the ˆ ˆ ˆ parameter estimates for bi1 , bi 2 , and bi 3 . The quadratic specification allows for the initial cost of ˆ capital shock to be either positive or negative. The parameter bi 2 captures the direction and curvature of the initial shock. The quadratic specification also allows for (but does not require) a recovery of the shock during the event period. The quadratic form is the benefit of the regimeswitching model because it can accommodate the impact of interim disclosure responses on firms’ costs of capital during the event period. Analysis later in the paper confirms that the curvature is, in fact, supported by the data. We compute the event period beta (BETA_EVT) at day t = 0, which represents the peak given the quadratic specification. This beta is computed from (B2) by setting t = 0 and recognizing that our event window is symmetric and hence T2 = -1*T1: ˆ ˆ ˆ ˆ βi 2 = bi1 + bi 2 (T 2) 2 + bi 3T 2 (B3) After the shock and during the pre-report period, systematic risk is: ˆ ˆ ˆ β i 3 = bi1 + bi 3 (T 2 − T 1) (B4) The measure represents the beta that “remains” (BETA_REM) after the market has incorporated the transparency crisis and the firms’ immediate responses to the crisis, but before -17- firms have responded in their earnings announcement or 10-K filings. Like the pre-event beta, the pre-report period beta (BETA_REM) is constrained to be a constant across the period. To provide intuition for the parameters, Figure 1 illustrates time-series patterns of systematic risk for the pre-event period, the event period, and the pre-report period under ˆ ˆ alternative estimates for bi 2 and bi 3 . Panels A through C illustrate the behavior of systematic risk ˆ ˆ ˆ in the cases HIGH positive bi 2 (equals 0.100), LOW positive bi 2 (equals 0.002), and negative bi 2 (equals -0.002), respectively. Within each panel, we present estimates of systematic risk in the ˆ pre-report period for five levels of bi 3 . In all three panels, the pre-event period beta estimates are constant over the pre-event period. The pre-report period betas also are constant throughout the period, but they can be ˆ higher or lower than the pre-event beta. When bi 3 is positive, the pre-report beta is higher than the pre-event beta. Furthermore, the pre-report period beta can be higher than the event period ˆ beta, even when bi 2 is positive. This follows from comparing (B3) and (B4) and requires that ˆ ˆ bi 3T 2 > bi 2 (T 2) 2 . Separately, we estimate the post-report period beta (BETA_POST) after the annual report season from May 1, 2002 to August 31, 2002. We estimate BETA_POST using a standard market model. We use differences in the four beta levels estimates to create proxies for changes in firms’ cost of capital (i.e., shocks) as a result of the Enron events. We define the initial shock (INIT_SHOCK) as the difference between the event period beta in (B3) and the pre-event period beta: INIT_SHOCK = BETA_EVT - BETA_PRE. Recalling that BETA_EVT is the peak of the quadratic, the initial shock metric captures the extent to which a firm’s systematic risk increases -18- during the Enron crisis. We define the remaining shock (REM_SHOCK) as the difference between the pre-report beta and the pre-event beta, i.e., REM_SHOCK = BETA_REM BETA_PRE. The remaining shock captures the extent to which a firm’s systematic risk continues to be elevated after the transparency crisis, and the firm’s potential immediate disclosure responses, but before the 10-K is filed. We define the beta response (BRESPONSE) as the difference between the post-report period beta and the event period beta, i.e., BRESPONSE = BETA_EVT - BETA_POST. A positive value of the beta response variable suggests a greater recovery of the beta relative to the peak level of systematic risk during the event period. Table 2 provides a summary of the cost of capital proxies for Window 1, which is a 39 day event window centered on November 8. The average pre-event beta for the sample firms, which serves as a benchmark and control variable in the analysis, is 0.670. The average and median beta increase from the pre-event period to the pre-report period. Greater than 25% of the sample firms have negative initial shocks, while more than 25% experience an increase in systematic risk greater than approximately 0.5. The remaining shock to systematic risk as of the pre-report period is 0.087 on average (median equals 0.108). The percentiles show that over 25% of the sample firms had lower systematic risk following the event period than during the pre-event period, but over 25% had a 0.6 increase in systematic risk, and the median firm had an increase. [INSERT TABLE 2 HERE.] -19- Panel B reports the average initial shock (INIT_SHOCK) and the average remaining shock (REM_SHOCK) by 2-digit SIC industry code and the proportions of positive observations in each industry. The industries with the largest initial shocks are Airlines (SIC 45), Furniture (SIC 25), Building products (SIC 15), and Auto dealers/gas stations (SIC 55). The airline industry remains among the worst affected after the event period. Approximately 90% of the airlines have a positive remaining shock, and the average remaining shock is close to 1. In our sensitivity analyses, we check that the results are robust to the exclusion of this industry.9 Other industries that have a high proportion (i.e., greater than 80%) of firms with positive remaining shocks are Furniture (SIC 25), Apparel (SIC 23), Oil and gas extraction (SIC 13), and Water transportation (SIC 44). 4. 4.1 Results of mandatory reporting analyses Main results The first analysis relates the percentage changes in the 10-K page count to the cost of capital shocks. The specification of the general model in equation (1) is: %ΔPAGES = α + β S SHOCK + β E BETA _ PRE + δ 1 SIZE + δ 2 ROA + δ 3 DERATIO + δ 4 MB + δ 5 PPE + ε PGS (2) We estimate the model including either the initial shock (INIT_SHOCK) or the remaining shock (REM_SHOCK). We predict a positive association between the initial shock 9 Carter and Simkins (2004) document that the events of September 11 had a significant effect on airlines. The effects were concentrated in passenger carriers and not air-freight firms. They suggest that the airlines were subject to uncertainty about pending government regulation. Doherty et al. (2003) similarly document significant effects of the events of September 11 on the insurance industry. The results also are robust to exclusion of insurance carriers (SIC 63). -20- and the percent change in pages under the assumptions that 1) the 10-K is a mechanism for firms to increase transparency, and 2) increases in transparency reduce the cost of capital, as predicted by theory. This prediction further assumes that a firm responds to the initial shock by expanding its 10-K regardless of a potential recovery from the initial shock in the interim. This issue does not arise for the remaining shock, for which we also predict a positive association with the percent change in pages. The analysis includes the pre-event beta (BETA_PRE) as a control variable.10 In addition, we include control variables that proxy for commonly cited determinants of disclosure such as firm size, performance, leverage and financing needs (Lang and Lundholm, 2000; Leuz and Wysocki, 2007). The proxies for these constructs are the natural log of total assets at December 31, 2000 (LASSETS2000); return of operating income on average assets for the year ended December 31, 2000 (ROA2000); the debt-equity ratio at December 31, 2000, which is the book value of long-term debt scaled by the market value of equity plus the book value of longterm debt and preferred shares (DERATIO2000); the market-book ratio at December 31, 2000 (MB2000);11 and the ratio of property, plant and equipment, net, to total assets (PPE/TA2000). An extended model also includes changes in several of these variables from 2000 to 2001. All models include one-digit SIC industry controls. (INSERT TABLE 3 HERE.) Table 3, Panel A presents results for eight model specifications. The results for models (1) - (4), which include the initial shock variable and four variations of control variables, show a 10 11 Standard diagnostic tests suggest that multicollinearity is not a problem. Observations of a market-to-book ratio less than zero are set to missing. -21- positive and significant relation between the initial shock (INIT_SHOCK) and the percent change in the page count of the 10-K. The estimates are insignificant in the models that include the changes specifications of the control variables, but close to conventional significance levels, considering that we use two-sided p-values. The results for models (5) - (8), which include the remaining shock variable and four variations of control variables, show that the relation between the remaining shock (REM_SHOCK) and page count changes also is significantly positive and slightly stronger than for the initial shock. These results are robust to alternative sets of control variables. In particular, the results are similar and the inferences the same if we use the market value in computing firm size, employ contemporaneous, rather than lagged controls and include only changes in the control variables in the model. The positive association for the remaining shock is consistent with our hypotheses and the notion that firms use mandatory reports to reduce their cost of capital. The significantly positive association for the initial shock suggests that there are complementarities between expanded disclosures in the mandatory reports and increased interim disclosures (or other mechanisms to address the beta shocks). That is, despite the possibility that the initial shock has been mitigated during the event period, firms with a greater initial exogenous shock respond with greater disclosure in their 10-K filings, consistent with the notion that any interim disclosures or actions need to backed up by expanded 10-K filings. While these results already suggest complementarities, we provide more direct evidence on this issue when we analyze immediate disclosure responses during the event period in Section 5. Although we do not make ex ante predictions about the control variables, the results are consistent across all eight models and accord with our intuition. Firms with higher pre-event -22- period betas, lower ROA, deteriorating performance, greater and growing firm size increase their 10-K page counts more. To increase the power of our page count proxy to measure an increase in informative disclosure, we next analyze the determinants of page count increases in specific sections of the 10-K. The variable %ΔMDA is the percent change in pages for MD&A; the variable %ΔFS is the percent change in the pages of 10-K sections that commonly include financial tables; and the variable %ΔDISCUSS is the percent change in the pages of 10-K sections that commonly include narrative discussions of the type of information we expect to be informative in response to a transparency crisis. Table 3, Panel B, presents the results. We tabulate results for either the initial shock (INIT_SHOCK) or the remaining shock (REM_SHOCK), the three disclosure changes, and the levels of the same control variables and industry dummies as in Panel A (model 2).12 We find that the initial shock is positively associated with the page count changes in all three 10-K sections but the relation is statistically significantly only for the page count changes in the financial statement sections of the 10-K. The remaining shock exhibits a similar pattern. The magnitudes of the coefficients on the shocks are similar to those in Panel A, suggesting that our tests simply lack power likely due to the smaller sample size. Next, we estimate the same models as in Table 3 using the percentage change in the word count of the earnings announcement in 2001 relative to 2000 (%ΔWORDS) as our disclosure response variable. In all regressions, the association between %ΔWORDS and the cost of capital shocks is insignificant and the p-values are far from conventional significance levels. For brevity, we do not tabulate these results. We subject the word count proxy to numerous 12 The results are similar but stronger when we use model 1 and weaker but consistent when we include changes in the control variables (model 3). -23- refinements, e.g., we adjust the word counts for the occurrence of unusual corporate events and by industry; we convert the percentage changes into a simple variable indicating increases, decreases and approximately no change in the size of the earnings announcement; we focus on large changes only, drop small earnings announcements (< 500 words), and use log changes. These refinements do not change the result that the association of interest is insignificant. Thus, we do not think that it is simply noise in the earnings announcement proxy that generates this result and the contrast to the findings for 10-K page counts. One potential explanation is that earnings announcements are simply not well suited to address the transparency concerns that arose after the collapse of Enron. Another (but more minor) issue is that firms’ earnings announcements and our beta estimation may overlap, which could result in confounding effects. 4.2 Robustness tests We subject the results in the previous section to a large battery of robustness tests to address a number of potential concerns. One potential omitted variable in the analysis is a fundamental change in the firm’s operations during the period that is correlated with both its beta and its required disclosures. For example, discontinuing operations may be associated with changes in systematic risk and it is likely to be associated with changes in page and word counts given the SEC disclosure requirements associated with discontinued operations. To control for fundamental changes in the firm, we adjust the changes in page count and word count for the existence of events that could change these counts without being an intentional decision to change transparency or a voluntary disclosure. Our reviews of financial statements, as well as disclosure requirements of the exchanges and the SEC, suggest that the existence of the following irregular items (IRREGITEM) in the current year and two previous -24- years affect the length of the financial statements and the earnings announcements: Discontinued operations, new segments, extraordinary items, accounting changes, certain special items (specifically litigation reserves or restructuring charges), and acquisition activity. We create seven indicator variables equal to one if the firm has evidence of these events based on Compustat data for the three years 1999, 2000, and 2001. We regress the changes in the page (word) count on these indicator variables and use the residuals from the following model (firm subscripts omitted) as dependent variable in our Table 3 regressions: %ΔPAGES = d + 2001 y =1999 j =1 ∑ ∑k 7 jy IRREGITEM jy While these adjustments provide a useful sensitivity check, it is a priori not obvious that the adjusted proxies dominate the unadjusted page (or word) changes. The transparency concerns after Enron and the cost of capital shocks may have induced firms to provide some of these irregular items. For instance, “coming clean” on certain transactions is likely to result in extraordinary or special items. Nevertheless, our results are unaffected by these adjustments. That is, the positive association between the cost of capital shocks and the page changes continues to be significant and, more importantly, the coefficients of key interest are not materially altered, indicating that unusual corporate events and fundamental changes in the firm are not responsible for our findings. As noted before, these adjustments do not alter our insignificant findings for the percentage change in the word count of the earnings announcement. Another potential concern is that beta estimates for infrequently traded stocks are downward biased (Scholes and Williams, 1977). As a result, the estimated beta shocks may in part reflect differences in liquidity, which matters because liquidity and corporate disclosure are -25- known to be related (e.g., Welker, 1995; Healy et al., 1999; Leuz and Verrecchia, 2000). To address this potential issue, we examine the relation between %ΔPAGES and the two beta shocks for a subsample of stocks with a share price above $5, which are less likely to have nonsynchronous returns. The $5 restriction reduces the sample size to 1,610 observations (for model 2), but the results are very similar and the inferences are the same. The results in models (1) – (6) are also robust to controlling for the log of average daily volume as a proxy for the downward bias in the beta estimates.13 A third potential correlated omitted variable is past performance. The tabulated results are for a model that includes ROA2000 and the change in ROA from 2000 to 2001. But operating performance generally lags stock returns. Moreover, losses may trigger additional language in firms’ earnings announcements and annual reports (Li, 2007). To address these issues, we expand models (3) and (6) by including the stock return over the past year (ending on August 31, 2001) and an indicator variable for loss firms (in 2001 and, alternatively, 2000). The results are robust to the inclusion of these variables and the inferences remain the same. Finally, in an effort to rule out spurious correlations as an explanation for the results, we examine whether the initial shock and remaining shock variables have any explanatory power for the percent change in page counts from 1999 to 2000. They do not and are close to zero, as it should be. Another way to gauge the issue is to analyze the correlation between the residuals from the switching regime model and volume. This analysis (untabulated) reveals that the correlation are on average small, approximately 5-6%, with approximately 20-30% of the correlations being significant and positive and 5-15% of the correlations being significant and negative. The correlations are similar across the pre-event, event, and pre-report periods. 13 -26- 4.3 Cross-sectional analysis of the 10-K responses In this section, we present a cross-sectional analysis of the 10-K responses. We identify two potential sources of cross-sectional variation in the expected benefits to expanding disclosures in the 10-K, which should lead to cross-sectional variation in the response. Exploiting this cross-sectional heterogeneity in the expected benefits of expanded 10-K disclosures improves identification. The first prediction is that firms respond to positive cost of capital shocks but not to negative shocks. This prediction assumes that the cost of retaining existing disclosures in the 10K is low, which creates a non-linearity in the net benefits of disclosure changes. The second prediction is that firms with a greater sensitivity to cost of capital shocks will exhibit a greater response. Firms with higher financing needs and larger growth opportunities are more likely to require ready access to equity markets and thus are more likely to be sensitive to shocks. In particular, such firms will be sensitive to positive shocks. Thus, we expect the strongest reaction to the shock as measured by %ΔPAGES for firms that have positive shocks and high equity market external financing needs. We estimate our main page change model separately for firms with positive shocks and a high sensitivity to cost of capital shocks vs. the (residual) sample of firms that have either a negative shock or a low cost of capital sensitivity. We use two proxies to identify firms that are more likely to be sensitive to equity market cost of capital shocks because of an expectation of using equity markets to access capital. Firms with large financing needs are identified by an above-median investment cash flow scaled by total assets during fiscal 2001. Firms with greater growth opportunities are identified by an above median market-to-book ratio at the end of fiscal 2000. We consider alternative partitioning variables for firms’ financing needs and growth -27- opportunities and obtain similar results. There are 1,055 observations with positive shocks and 1,018 observations with negative shocks. Approximately 1/3 of the sample observations have both positive shocks and high sensitivity to financing needs. We expect the results in terms of page count response to be strongest for this group. (INSERT TABLE 4 HERE.) Table 4 presents the cross-sectional page change results. Columns (1) through (4) present results using above-median investment cash flow to identify high-sensitivity firms; Columns (5) through (8) present results using above-median growth opportunities to identify high-sensitivity firms. Separate models estimate the impact of the initial shock and the remaining shock on %ΔPAGES. All models include industry fixed effects. Using either proxy for a high cost of capital sensitivity, the reaction to the initial beta shock (INIT_SHOCK) is significant at the 5% level within the positive shock/high sensitivity sub-sample, while it is not significantly different from zero for the residual firms. The coefficient estimates on INIT_SHOCK in models (1) and (2) and the coefficient estimates in models (5) and (6) are significantly different at less than the 10% level.14 With respect to the remaining shock (REM_SHOCK), the coefficient estimate on REM_SHOCK for the positive shock/high sensitivity sub-sample is directionally greater than that on the residual sample, but we cannot reject the null hypothesis that the two coefficients are equal. 14 The results are similar if we simply introduce an indicator for positive shocks and an interaction term. For this specification, the effect of negative initial shocks is again negative but far from conventional significance levels, while the effect of positive shocks is positive and significant as expected. -28- 5. Disclosure during the event-period The results in Tables 3 and 4 indicate that firms that have a greater initial exogenous shock respond with greater disclosure in their 10-K filings, regardless of subsequent changes to systematic risk that occur during the event period, including possibly endogenous changes due to interim disclosures. These results suggest that firms view mandatory reports as complements to other more immediate disclosure responses to the transparency shock. In this section, we expand the regression model to include proxies for four disclosure mechanisms that firms might have used as a more immediate disclosure response to the transparency shock: 8-K filings, conference call activity, disclosure in the earnings announcement, and a market-based measure of the firm’s response to the transparency shock ˆ derived from the regime switching model ( bi 2 ). Broadly speaking, the analysis generates two key results. First, the association between the beta shocks and the changes in 10-K pages are robust to the inclusion of the immediate disclosure responses in the model. Second, the immediate disclosure proxies are positively correlated with the 10-K page responses, which suggests that the interim disclosure responses and the 10-K responses are complementary. (INSERT TABLE 5 HERE.) Models (1) and (2) include a truncated version of Δ8KCOUNT to measure 8-K filing activity. Δ8KCOUNT equals the change in the count of 8-K filings for changes of -1, 0, and +1. For changes less than or equal to -2 (greater than or equal to +2), Δ8KCOUNT = -2 (+2). The coefficient estimates on INIT_SHOCK and REM_SHOCK remain significant. The change in the number of 8-K filings is positively correlated with the percentage change in 10-K pages. -29- Models (3) and (4) include a binary variable (NEWCALL) which measures the initiation of conference calls in the period after November 1, 2001 until January 15, 2002. We obtain data on calls from First Call. From November 1, 2001 to January 15, 2002, 229 firms that had not hosted a call during the same period in 2000 hosted a call. Of these, 183 firms also had not hosted a call in the same period of 1999. The firms that initiate new conference calls also increase page counts in the 10-K. The coefficients on the initial shock and the remaining shock variables remain positive and significant. Models (5) and (6) include the percent change in words in the annual earnings announcement (%ΔWORDS) as a determinant of the percent change in pages in the 10-K. These models provide no evidence that firms’ choices regarding additional informative disclosure in earnings announcements, as measured by word count changes, are related to their decisions about additional informative disclosure in the 10-K as measured by page count changes. ˆ Model (7) in Table 5 includes the bi 2 parameter from the regime switching model as a proxy for a firm’s immediate responses to the transparency crisis and the resulting beta shock. ˆ As illustrated in Figure 1, a larger bi 2 implies a larger curvature in the event period beta. One explanation for this curvature is that a firm has responded to the initial shock, for example, through interim disclosure or other actions that enhance a firm’s credibility, such as changes in ˆ dividend policies. In this sense, bi 2 is an indirect measure of a firm’s responses to the initial shock in the regressions using the remaining shock. It should capture the effects of all actions the firm takes during the event period, regardless of whether we can identify or measure them. ˆ The downside of this measure, however, is that bi 2 may also capture changes in the cost of capital unrelated to a firm’s direct actions, including recoveries in the cost of capital due to -30- information transfers from other firms. However, as long as these other changes are not systematically biased, they should introduce only noise into our analysis. Table 5, Column 7 shows a positive and significant association between the percent ˆ change in 10-K pages and bi 2 , which indicates that there are complementarities between a firm’s immediate responses to the shock and the disclosure responses in the 10-K filing. The results in Table 5, taken together, emphasize that using the regime-shifting model that allows for the quadratic estimate of systematic risk during the event period is an important innovation. We intentionally define the event period to be long enough to capture the series of individual announcements that we predict affect systematic risk. However, having an extended event period increases the likelihood of confounding events during the event period, in particular, due to firms’ responses to the initial shock. Ignoring this possibility by using a model with a fixed estimate of systematic risk is contradictory to the fundamental prediction of the paper, which is that firms do respond to the shock. Table 6 provides a more detailed analysis of firms’ interim 8-K filing responses to the transparency crisis. Table 6 is the analog to the analysis of 10-K responses in Table 3, but for 8K filings. To capture the unexpected 8-K activity for the period, we use the truncated version of the Δ8KCOUNT variable: Δ8KCOUNT equals the change in the count for changes of -1, 0, and +1 and equals -2 (+2) for changes less than or equal to -2 (greater than or equal to +2). We also estimate the model of the percent change in the number of filings (%Δ8KCOUNT). [INSERT TABLE 6 HERE.] -31- Table 6 indicates a significant and positive association between changes in 8-K filing activity and the initial cost of capital shock. The relation between the remaining shock and the 8K filings is negative. Given the timing of the 8-K filings as interim reports, this result suggests that the 8-K filings helped mitigate the cost of capital effects during the event window. 6. Analysis of beta responses to 10-K filings Table 7 presents an analysis of the change in systematic risk between the event period and the post-filing period (May 1, 2002 – August 31, 2002). The dependent variable in the analysis is BRESPONSE. The model includes the same control variables as in our other models. The explanatory variable of interest is %ΔPAGES. Our prediction is that more informative disclosure, as measured by the percent increase in 10-K pages, will lead to a lower post-filing beta and thus a more positive beta response. Thus, we expect a positive coefficient estimate on %ΔPAGES. [INSERT TABLE 7 HERE.] The results are consistent with the positive association. The relation is significant in models (1) and (2) that include controls in levels only. When control variables specified in changes are included in the regression, the t-statistic for the coefficient estimate drops to 1.50. 7. Conclusions In this paper, we analyze the relation between voluntary disclosures and the cost of capital using an exogenous cost of capital shock that the Enron scandal in Fall 2001 created for -32- other U.S. firms. By linking firms’ subsequent disclosure responses to the cost of capital shocks, we mitigate endogeneity concerns in cross-sectional studies of voluntary disclosure choices and cost of capital. Using a unique method to estimate event-period betas and cost of capital shocks, we show that the events surrounding the Enron scandal increase the average beta for U.S. firms. We document that beta shocks are associated with an increase in voluntary disclosures in mandatory reports. Firms extend the number of pages in their annual 10-K filings, notably the section containing financial statements and footnotes. The increase in disclosure is particularly pronounced for firms that experience positive beta shocks and have large financing needs and growth opportunities. We also find evidence suggesting that interim disclosures, such as earnings announcements, conference calls, and 8-K filings are complementary to the increase in disclosure in the 10-K. Moreover, there is evidence that firms’ disclosure responses are effective in reducing the impact of the transparency crisis on U.S. firms’ costs of capital. These results complement prior studies and increase our confidence in evidence on the link between voluntary disclosure and firms’ cost of capital. -33- Appendix Section A.1 describes Enron-related events between October 16, 2001 and January 26, 2002. From these, we identify seven potential sets of pre-event – event – pre-report windows that are outlined in Section A.2. The final section reports the results of ex post analysis of the beta estimates over the various windows to assess their validity. A.1 Event dates related to Enron We identify potentially significant events from published timelines of the Enron scandal (e.g., Platts, 2002; Washington post, 2002), and from independent searches of news reports. Potentially important event dates in the context of our study are those that changed investors’ prior beliefs about the likelihood or magnitude of private information about the sample firm. We identify three types of event dates. First, we identify six Enron-specific events that are also indicators that financial reporting or corporate governance problems may be widespread. Oct. 16: Enron first announces its huge losses and hints of conflict-of-interest problems with its partnerships. Oct.22: Enron announces that the SEC has launched an inquiry. Oct 29: The SEC moves its inquiry to DC signaling a higher likelihood of securities violations. Oct 31: The SEC initiates a formal investigation. Nov 8: Enron restates earnings back to 1997. Nov 29: The SEC’s investigation is expanded to include Arthur Andersen LLP. These events are specific to Enron, but they are not necessarily the most significant for Enron based on an analysis of its stock returns. This set does not include events such as a credit rating downgrade of Enron’s debt or a takeover offer of Enron which are significant for Enron but unlikely (in our view) to have broad implications for information problems at other firms. Second, we identify events that directly indicate that the information problems at Enron are expected to be systemic. Nov 19: Representative John Dingell requests that the Public Oversight Board (POB) review Andersen’s audits of Enron -34- Nov 29: SEC Chairman Harvey Pitt calls pro forma earnings "unstructured and undisciplined" and suggests that the SEC is focusing attention on misleading financial reporting. Dec 4: The Big-five firms pledge to work together to address financial reporting problems. Jan. 7: It is reported that the Big Five have petitioned the SEC to improve disclosure regulation. Jan 9: The Senate Banking Committee, headed by Sen. Paul Sarbanes, announces its plans to hold a hearing Feb. 12 to examine accounting and investor protections. The US Justice Department's fraud section also announces that it will form a special task force to examine the collapse of Enron. Third, we identify financial press discussions that suggest that the Enron scandal was not an isolated problem but rather an indicator of corporate transparency and corporate governance problems. Nov 5: “What Enron’s Financial Reports Did – and Didn’t – Reveal --- Auditor Could Face Scrutiny on Clarity of Financial Reports” (Wall Street Journal, p. C1) Nov 8: In a DJ Newswire column, the author speculates that Enron’s partnership accounting problems, which resulted in significant earnings restatements, could push the FASB to prioritize its projects on SPEs from having standards of ownership including provisions for minimum ownership by unaffiliated outsiders to control-based standards. The article also questions Arthur Andersen’s responsibilities given the restatements. Nov 26: Business Week cover story is “CONFUSED ABOUT EARNINGS? You're not alone. Here's what companies should do--and what investors need to know.” Jan 26: “Trying not to be the next Enron, companies scrutinize practices” appears on p.1 of the New York Times Saturday business section. These articles are not events per se. However, it is possible (or even likely) that the articles affected public opinion about the systemic nature of the transparency problem. Existing research has shown that articles in the financial press do affect investor opinions (e.g., Foster, 1979; Foster, 1987; Pruitt et al. 1999; Huberman and Regev, 1999; Chang and Suk, 1998). It is also possible that the articles reflect public opinion rather than create it, which is another reason to search for and consider these dates. We ended the search for such articles in January 2001 because that is the start of the earnings announcement season. -35- A.2 Identification of event windows Figure A1 shows the value-weighted and equal-weighted index from August 1, 2001 through February 28, 2002 and provides some context for the window selection. The figure illustrates the high volatility period associated with the September 11th terrorist attacks. The arrows point to the start dates and the end dates that we consider for the event period. Figure A1: Index returns We identify seven combinations of windows to estimate systematic risk for the pre-event period (BETA_PRE), the event period (BETA_EVT), and the pre-report period (BETA_REM). In the timelines below, T1 and T2 define the endpoints of the event period. The center date of the event period is noted for each window. In all scenarios, the end of the pre-event period is August 31, 2001. There is a discontinuity between the end of the pre-event period and the beginning of the event period ranging from 44 to 62 days. The discontinuity allows us to avoid -36- the week-long trading halt and subsequent high volatility period associated with September 11. In addition, the discontinuity is advantageous because the pre-event period is not contaminated when we do not properly identify the start of the event period. The discontinuity does not affect the estimation procedure. Alternative Windows for the Beta Estimation −−−−−−−β1−−−−−− Window 1: 05/01/01–8/31/01 Window 2: 05/01/01–8/31/01 Window 3: 05/01/01–8/31/01 Window 4: 05/01/01–8/31/01 Window 5: 05/01/01–8/31/01 Window 6: 05/01/01–8/31/01 Window 7: 05/01/01–8/31/01 T1 −−−−−−−−−β2−−−−−−−−−− Nov 8 10/15/01–12/05/01 Nov 12 10/19/01–12/05/01 Nov 19 11/02/01–12/05/01 Nov 27 10/15/01-01/10/02 Nov 29 10/19/01–01/10/02 Nov 20 10/15/01-12/31/01 Nov 23 10/19/01–12/31/01 T2 −−−−−−−−β3−−−−−−−− 12/06/01 – 01/15/02 12/06/01 – 01/15/02 12/06/01 – 01/15/02 1/11/02-01/28/02 1/11/02-01/28/02 1/01/02-01/28/02 1/01/02-01/28/02 We consider three start dates for the event window: October 15, October19, and November 2, 2001. Monday October 15 is day t-1 relative to Enron’s announcement that it lost $618 million and which provided the first indication of possible conflict-of-interest questions related to the Fastow-run partnerships and the “quality” of Enron’s prior-period earnings. Friday October 19 is trading day t-1 relative to Enron’s announcement of an SEC investigation of the partnerships and of suspicions by analysts that Enron will release additional bad news (October 22). November 2 is the start date that makes Monday November 19 the center of the event window. It is first reported on November 19 that Representative John Dingell has requested that -37- the Public Oversight Board (POB) review Andersen’s audits of Enron (and Waste Management). The public oversight Board declines to review specific cases but indicates that it will consider a systemic review.15 We consider two end dates for the event window. The end date in windows 1-3 is December 5, which is day t+1 relative to the pledge by the big-five firms to work together to address financial reporting problems, especially related to SPEs and market risks. The event periods in these windows include all of the Enron-specific events and all but three of the systemic and financial press events. The pre-report period begins December 6 and ends on January 15, 2002, which is the first percentile of fourth quarter earnings announcement dates. In windows 4 and 5, the event period extends to January 10, 2002, which is day t+1 relative to the last systemic event and it includes all but the last financial press event. Because of the longer event period, we are forced to extend the length of the pre-report period to January 28 in order to have a sufficiently long pre-report period to estimate the model. January 28 represents the first quartile of earnings announcement dates. In windows 6 and 7, we arbitrarily set the end of the event window at December 31, 2001. This specification provides more equally sized windows for the event-period and the pre-report period. A.3 Analysis of beta estimates This section reports analyses that guided our decision to focus on window 1.16 In choosing among the windows, the primary trade-off is that a longer event window, which is more An analysis of cumulative abnormal returns (CARs) on the individual event dates (not tabulated) suggests that November 19 is a consistently important event. Explanations for the rally on the 19th in the financial press were preThanksgiving volume, and that it was a continuation of a rally due to lower oil prices and the fact that no bad news had been released. 16 Similarly, Lockwood and Kadiyala (1988) estimate the model for every combination of event windows with a start date within 30 days prior to the one day event (T1) and an end date within 30 days after the event (T2). The window (T1, T2 combination) that maximizes the log likelihood function is chosen as the best firm-specific model. 15 -38- inclusive of important events, mechanically shortens the pre-report window because the report date (either the earnings announcement date or the 10-K date) is fixed. Several analyses favor the shorter event periods (windows 1-3). The means (medians) of the adjusted R2s from the model estimation are higher for these windows than for the other windows (although the differences are not significant). Windows 1-3 also produce the smallest number of negative β1 estimates (regardless of the start date). Finally, the residuals for windows 1, 2 and 3 are mean zero during the pre-event period and the event period. They are significantly positive during the pre-report period (at approximately the 10% level). In windows 4-7, the residuals during the event (pre-report) period are positive (negative) and significant. None of the analyses indicate a difference between windows 1-3; thus, we report results for the window that allows the earliest possible start date (window 1). -39- References Barry, C., and S. Brown, 1984, Differential Information and the Small Firm Effect, Journal of Financial Economics 13, 283-294. Barry, C., and S. Brown, 1985, Differential Information and Security Market Equilibrium, Journal of Financial and Quantitative Analysis 20, 407-422. Beneish, Messod D., Patrick E. Hopkins, Ivo Ph. Jansen, and Roger D. Martin, 2005, Do auditor resignations reduce uncertainty about the quality of firms’ financial reporting?, Journal of Accounting and Public Policy 24 (5), 357-390. Botosan, C., 1997, Disclosure Level and the Cost of Equity Capital, The Accounting Review 72, 323-349. Botosan, C., and M. Plumlee, 2002, A Re-Examination of Disclosure Level and the Expected Cost of Equity Capital, Journal of Accounting Research 40, 21-40. Brown, S., 1979, The Effect of Estimation Risk on Capital Market Equilibrium, Journal of Financial and Quantitative Analysis 15, 215-220. Carter, D. and Simkins, B., 2004, The market's reaction to the unexpected, catastrophic events: The case of airline stock returns and the September 11th attacks, Quarterly Review of Economics and Finance, 44 (4): 539-558. Chang, Saeyoung and David Y. Suk, 1998, Stock prices and the secondary dissemination of information: The WSJ's insider trading spotlight column, Financial Review 33 (3): 115-128. Clarkson, P., J. Guedes, and R. Thompson, 1996, On the diversification, observability and measurement of estimation risk, Journal of Financial and Quantitative Analysis 31, 69-84. Coles, J., 1988, Equilibrium pricing and portfolio composition in the presence of uncertain parameters and estimation risk, Journal of Financial Economics 22, 279-303. TColes, J., U. Loewenstein, J. Suay, 1995, On equilibrium pricing under parameter uncertainty, The Journal of Financial and Quantitative Analysis 30, 347-374. Core, J., 2001, A Review of the Empirical Disclosure Literature: Discussion, Journal of Accounting and Economics 31, 441-456. Cyree Ken B. and Ramon P. DeGennaro, 2002, A generalized method for detecting abnormal returns and changes in systematic risk, Review of Quantitative Finance and Accounting 19 (4): 399-416. Diamond, D., and R. Verrecchia, 1991, Disclosure, Liquidity, and the Cost of Capital, Journal of Finance 46, 1325-1359. -40- Doherty, Neil A., Joan Lamm-Tennant, and Laura T. Starks, 2003, Insuring September 11th: Market recovery and transparency, Journal of Risk and Uncertainty 26 (2-3): 179-199. Francis, J., LaFond, R., Olsson, P., and Schipper, K., 2004. Costs of Equity and Earnings Attributes. The Accounting Review 79, 967-1010. Francis, J., LaFond, R., Olsson, P., and Schipper, K., 2005. The Market Pricing of Accruals Quality. Journal of Accounting and Economics 39, 295-327. Francis, J., Nanda, D., and Olsson, P., 2005. Voluntary Disclosure, Information Quality, and Costs of Capital, Duke University Working Paper. Gur Huberman and Tomer Regev, 1999, Speculating on a cure for cancer: A Non-Event that made stock prices soar, Working paper, Columbia Business School. Hail, L., 2002, The Impact of Voluntary Corporate Disclosures on the Ex Ante Cost of Capital for Swiss Firms, European Accounting Review 11, 741-773. Healy, P., A. Hutton, and K. Palepu, 1999, Stock Performance and Intermediation Changes Surrounding Sustained Increases in Disclosure, Contemporary Accounting Research 16, 485-520. Healy, P., and K. Palepu, 2001, Information Asymmetry, Corporate Disclosure, and the Capital Markets: A Review of the Empirical Disclosure Literature, Journal of Accounting and Economics 31, 405-440. Hughes, J., J. Liu, and J. Liu, 2005, Information, Diversification, and Cost of Capital, UCLA Working Paper. Jorgenson, B. and M. Kirschenheiter, Disclosure, Betas and Information Quality, Working paper, Columbia University. Lambert, R., C. Leuz, and R. Verrecchia, 2006, Accounting Information, Disclosure, and the Cost of Capital. Wharton School Working Paper. Lang, M., and R. Lundholm, 1993, Cross-Sectional Determinants of Analyst Ratings of Corporate Disclosures, Journal of Accounting Research 31, 246-271. Lang, M., and R. Lundholm, 2000, Voluntary Disclosure and Equity Offerings: Reducing Information Asymmetry Or Hyping the Stock? Contemporary Accounting Research 17, 623. Larcker, D. and T. Rusticus, 2005, On the Use of Instrumental Variables in Accounting Research, University of Pennsylvania Working Paper. Leuz, C., and R. Verrecchia, 2000, The Economic Consequences of Increased Disclosure, Journal of Accounting Research 38, 91-124. -41- Lockwood, Larry J. and K. Rao Kadiyala, 1988, Risk measurement for event-dependent security returns, Journal of Business and Economic Statistics 6 (1): 43-49. Nikolaev, V., and L. van Lent, 2005, The Endogeneity Bias in the Relation Between Cost-ofDebt Capital and Corporate Disclosure Policy, European Accounting Review 14, 677-724. Pruitt, Stephen W., Bonnie F. Van Ness and Robert A. Van Ness, 2000, Clientele trading in response to published information: Evidence from the 'Dartboard' Column, Journal of Financial Research 23 (1):1-13. Rogers, Jonathan L., 2005, Disclosure quality and management trading incentives, Ph.D. Dissertation, University of Pennsylvania. Verrecchia, R., 2001, Essays on disclosure, Journal of Accounting and Economics 32, 97-180. Welker, M., 1995, Disclosure Policy, Information Asymmetry, and Liquidity in Equity Markets, Contemporary Accounting Research 11, 801-827. -42- Table 1 Descriptive statistics for measures of changes in the mandated reports Panel A reports descriptive statistics for the page count of the 10-K filings and the word count of the earnings announcements. PAGES2001 (WORDS2001) are the levels of the number of pages (words) in the 2001 10-K (fourth quarter 2001 earnings announcement). %ΔPAGES (%ΔWORDS) are the percent changes in these variables over the prior year (2000). Page count variables are presented for the entire 10-K, the MD&A section (Item 7); the sum of Items 6 (Selected Financial Data), 8 (Financial Statements and Supplementary Data), and 15 (Exhibits, Financial Statement Schedules, and Reports on Form 8-K), which commonly include financial tables (FS); and the sum of Items 1 (Business), 2 (Properties), and 7 (MD&A), which commonly include narrative discussions about the firm (DISCUSS). The table reports the number of 8-K filings (in total and excluding filings related to the September 11th terrorist attacks) between November 1, 2001 and April 30, 2002 and the percent change in the number of filings over the same period of the previous fiscal year (%Δ8ΚCOUNT). All change variables are truncated at the 5th and 95th percentile of the respective distribution. Panel A: N 10-K Page counts PAGES2001 MDA2001 DISCUSS2001 FS2001 %ΔPAGES %ΔMDA %ΔDISCUSS %ΔFS Word counts WORDS2001 %ΔWORDS 8-K counts Number of 8Ks (without Sept 11) %Δ8KCOUNT (without Sept 11) 2,172 1,279 1,336 1,336 2,172 1,279 1,326 1,314 2,079 2,039 2,172 2,172 2,172 2,172 Mean 64.692 13.755 27.425 34.626 17.0% 38.2% 22.1% 16.6% 2,077 16.9% 1.475 0.963 11.3% 7.8% Std dev 31.963 9.936 15.278 18.025 36.2% 43.7% 31.6% 54.1% 1,304 38.0% 2.688 1.899 60.7% 56.0% p25 44.000 8.000 17.000 25.500 0.0% 11.1% 3.8% 0.0% 1,235 -5.4% 0 0 0 0 Median 62.000 11.000 25.000 32.000 8.9% 30.8% 17.6% 7.7% 1,756 9.1% 0 0 0 0 p75 80.000 18.000 34.000 41.000 23.5% 55.6% 34.6% 22.2% 2,577 30.7% 2 1 0 0 Panel B: Pairwise Pearson correlations between page count variables, word count variables, and firm characteristics commonly associated with disclosure levels. Variable definitions: LASSETS is log of total assets; LSIZE is the natural log of the market value of common equity plus the book value of preferred stock and long-term debt; ROA is return on assets; LAGRET is the cumulative return from September 1, 2000 to August 31, 2001; DERATIO is long-term debt scaled by total assets; MB is the market book ratio (set = 0 for book value < 0); PPE/TA is the book value of net PPE scaled by total assets; ANLST is the number of analysts following the firm; and VOL is average daily trading volume. * indicates significance at 10% level. DISCUSS2001 WORDS2001 PAGES2001 LASSETS DERATIO MDA2001 PPE/TA LAGRET MDA2001 DISCUSS2001 FS2001 WORDS2001 LASSETS LSIZE ROA LAGRET DERATIO MB PPE/TA ANLST VOL 0.75* 0.84* 0.87* 0.27* 0.23* 0.21* -0.09* -0.08* 0.13* 0.07* 0.06* 0.10* 0.09* 0.82* 0.51* 0.34* 0.37* 0.29* -0.02 -0.01 0.14* 0.04 -0.07* 0.17* 0.07* 0.51* 0.37* 0.34* 0.32* -0.10* -0.05 0.10* 0.10* -0.06* 0.23* 0.12* 0.26* 0.37* 0.31* 0.04 -0.00 0.18* 0.09* 0.08* 0.10* 0.06* 0.54* 0.51* 0.10* 0.04* 0.17* 0.10* -0.01 0.27* 0.23* 0.90* 0.39* 0.21* 0.33* 0.10* 0.04* 0.43* 0.35* 0.35* 0.18* 0.12* 0.27* 0.10* 0.56* 0.42* 0.44* 0.16* -0.02 0.18* 0.11* 0.06* 0.14* 0.07* 0.02 -0.09* -0.08* 0.06* 0.16* -0.16* -0.04* -0.03 0.14* 0.13* -0.01 0.03 ANLST 0.48* LSIZE FS2001 ROA MB Table 2 Summary of model parameter estimates and measures of systematic risk Panel A reports descriptive statistics of parameter estimates and systematic risk from estimation of equations (B1) ˆ and (B2). During the pre-event period (BETA_PRE), b measures systematic risk. During the event period, i1 ˆ ˆ ˆ ˆ systematic risk is β i 2 = bi1 + bi 2 (T 1 − t )(t − T 2) + bi 3 (t − T 1) . We compute the event period beta (BETA_EVT) at ˆ ˆ ˆ day t = 0. During the pre-report period, systematic risk (BETA_REM) is β = b + b (T 2 − T 1) . During the posti3 i1 i3 report period, systematic risk (BETA_POST) is estimated using a market model from May 1, 2002 to August 31, 2002. The initial shock (INIT_SHOCK) is BETA_EVT - BETA_PRE. The remaining shock (REM_SHOCK) is BETA_REM - BETA_PRE. The beta response (BRESPONSE) is BETA_POST - BETA_EVT. Panel A: N Model parameters ˆ bi1 ˆ b Mean Std dev 25th % ile Median 75th % ile 2,172 2,172 2,172 0.66990 -0.00022 0.00190 0.69165 0.00273 0.02672 0.19435 -0.00162 -0.00911 0.52719 -0.00012 0.00273 0.94633 0.00116 0.01457 i2 ˆ bi 3 Estimates of systematic risk BETA_PRE BETA_EVT BETA_REM BETA_POST Estimates of shocks INIT_SHOCK REM_SHOCK BRESPONSE 2,172 2,144 2,085 -0.04494 0.08678 -0.07635 0.94692 0.89798 0.85812 -0.52135 -0.33435 -0.52995 0.00849 0.10787 -0.11141 0.46342 0.55186 0.34131 2,172 2,172 2,144 2,085 0.66990 0.62496 0.74800 0.69348 0.69165 0.97246 1.00325 0.47815 0.19435 0.03433 0.09887 0.32904 0.52719 0.56407 0.67850 0.69195 0.94633 1.14465 1.27465 0.98320 Panel B: Panel B reports the average initial and remaining shock by industry and the proportions of positive observations in each industry. Data are presented for industries with more than 5 sample observations. SIC 45 25 15 55 24 72 23 57 29 63 82 34 31 58 37 49 65 27 64 20 62 40 30 48 60 35 61 79 67 80 26 32 38 33 39 13 50 17 36 73 70 47 42 51 10 28 59 87 44 78 16 22 Transportation by Air Furniture and Fixtures Bldg Cnstr-Gen Contr, Op Bldr Auto Dealers, Gas Stations Lumber and Wood Pds, Ex Furn Personal Services Apparel & Other Finished Pds Home Furniture & Equip Store Pete Refining & Related Inds Insurance Carriers Educational Services Fabr Metal, Ex Machy, Trans Eq Leather and Leather Products Eating and Drinking Places Transportation Equipment Electric, Gas, Sanitary Serv Real Estate Printing, Publishing & Allied Ins Agents, Brokers & Service Food and Kindred Products Security & Commodity Brokers Railroad Transportation Rubber & Misc Plastics Prods Communications Depository Institutions Indl,Comml Machy, Computer Eq Nondepository Credit Instn Amusements, Recreation Holding, Other Invest Offices Health Services Paper and Allied Products Stone, Clay, Glass, Concrete Pd Meas Instr; PhotoGds; Watches Primary Metal Industries Misc Manufacturing Industries Oil and Gas Extraction Durable Goods-Wholesale Construction-Special Trade Electr, Oth Elec Eq, Ex Cmp Business Services Hotels, Other Lodging Places Transportation Services Motor Freight Trans, Warehouse Nondurable Goods-Wholesale Metal Mining Chemicals & Allied Products Miscellaneous Retail Engr, Acc,Resh, Mgmt, Rel Svcs Water Transportation Motion Pictures Heavy Constr; not Bldg Cntractrs Textile Mill Products N 13 6 11 9 7 7 7 6 9 98 7 30 9 29 38 70 20 27 10 37 23 8 22 67 281 105 21 21 32 39 21 15 101 29 15 75 52 7 127 254 19 6 22 26 10 173 22 58 10 6 8 9 INIT_SHOCK 1.1047 0.6402 0.5914 0.5488 0.4201 0.3260 0.3180 0.2949 0.2720 0.2342 0.2185 0.2109 0.1957 0.1716 0.1642 0.1343 0.1324 0.1168 0.1073 0.0844 0.0801 0.0663 0.0622 0.0587 0.0434 -0.0021 -0.0072 -0.0345 -0.0431 -0.0440 -0.0549 -0.0823 -0.0833 -0.0883 -0.0933 -0.1393 -0.1519 -0.1540 -0.1625 -0.1895 -0.2199 -0.2330 -0.2510 -0.2648 -0.2854 -0.2893 -0.3160 -0.4143 -0.4219 -0.5024 -0.5086 -0.6135 % POS 0.8462 0.8333 0.8182 0.7778 0.7143 0.5714 0.5714 0.6667 0.4444 0.6735 0.5714 0.6333 0.6667 0.5172 0.6579 0.5714 0.5500 0.5556 0.5000 0.5405 0.6957 0.6250 0.6364 0.5970 0.5552 0.5905 0.4762 0.4286 0.5625 0.4359 0.6191 0.5333 0.4158 0.4138 0.4667 0.4400 0.4231 0.5714 0.4252 0.4567 0.3684 0.3333 0.4091 0.4231 0.5000 0.4046 0.3636 0.3448 0.4000 0.5000 0.2500 0.5556 N REM_SHOCK 13 0.9161 6 0.5301 11 0.4098 9 0.2675 7 0.1426 7 0.7008 7 0.3383 6 -0.3993 9 -0.1121 98 0.0989 7 0.1449 30 0.3002 9 0.4491 29 0.1459 38 0.1416 69 0.0395 20 0.1214 27 0.2916 10 0.0091 37 0.0584 23 0.3442 8 0.2662 22 -0.1174 66 0.3625 281 0.1352 101 0.0079 20 0.5894 20 0.2292 32 0.0838 39 -0.2189 21 0.1528 15 0.2790 100 -0.1050 27 0.3667 15 -0.4003 75 0.4286 52 0.2265 7 -0.3520 121 0.0626 246 0.0416 19 0.0049 6 0.0114 22 0.1540 26 0.2188 10 0.3835 172 -0.2476 21 -0.0484 57 -0.3202 10 0.3883 6 0.7753 8 0.4448 9 0.0308 % POS 0.8462 0.8333 0.9091 0.7778 0.7143 1.0000 0.8571 0.3333 0.7778 0.5000 0.7143 0.6667 0.7778 0.7241 0.5526 0.4493 0.4500 0.7407 0.4000 0.4595 0.7391 0.7500 0.5455 0.6667 0.6192 0.6238 0.7500 0.7000 0.5625 0.3590 0.6191 0.7333 0.4900 0.7037 0.4667 0.8133 0.5192 0.2857 0.5868 0.5732 0.5263 0.5000 0.6818 0.6923 0.7000 0.3837 0.7143 0.3509 0.8000 1.0000 0.5000 0.5556 Figure 1 Illustrations of the time-series patterns in daily systematic risk estimates for the pre-event period, the event period, and the pre-report period for alternative values ˆ ˆ ˆ of b . Panels A through C illustrate levels of systematic risk in the case of HIGH positive b (equals 0.100), LOW positive b (equals 0.002), and negative i2 i2 i2 ˆ ˆ bi 2 (equals -0.002), respectively. Within each panel, we present estimates of systematic risk in the pre-report period for five levels of bi 3 . ˆ Panel A: HIGH Positive bi 2 ˆ Panel B: LOW Positive bi 2 ˆ Panel C: Negative bi 2 2 40 1.5 35 2.5 30 1 2 25 1.5 0.5 20 b3 = -0.046 b3 = -0.015 b3 = -0.006 15 b3 = 0.002 b3 = 0.020 1 0 20 01 20 050 01 1 20 050 01 9 20 051 01 7 20 052 01 5 20 060 01 5 20 061 01 3 20 062 01 1 20 062 01 9 20 071 01 0 20 071 01 8 20 072 01 6 20 080 01 3 20 081 01 3 20 082 01 1 20 082 01 9 20 090 01 7 20 092 01 1 20 100 01 1 20 100 01 9 20 101 01 7 20 102 01 5 20 110 01 2 20 111 01 2 20 112 01 0 20 112 01 9 20 120 01 7 20 121 01 7 20 122 02 6 20 010 02 4 01 14 0.5 10 -0.5 0 2E +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 2E 7 +0 7 5 -1 0 200105012001051720010605200106212001071020010726200108132001082920010921200110092001102520011112200111292001121720020104 200105092001052520010613200106292001071820010803200108212001090720011001200110172001110220011120200112072001122620020114 -0.5 -1.5 -5 -1 Table 3 Relation between percent changes in 10-K page counts and cost of capital shocks Panel A reports regression results for eight models of percent changes in page counts (%ΔPAGES) on combinations of initial shock (INIT_SHOCK), remaining shock (REM_SHOCK), pre-event beta (BETA_PRE), and control variables. Variable definitions: LASSETS is log of total assets; ROA is return on assets; DERATIO is long-term debt scaled by total assets; MB is the market book ratio (set = 0 for book value < 0); PPE/TA is the book value of net PPE scaled by total assets. The control variables are measured as of fiscal year 2000. The changes are measured from 2000 to 2001, except in those cases where the variable construction uses market values of equity. The regression models include industry fixed effects as noted. Robust t-statistics are in parentheses. *** {**} (*) indicates significance at the 1%, 5%, 10% level in a two-sided test. Panel A: Constant INIT_SHOCK REM_SHOCK BETA_PRE LASSETS2000 ROA2000 DERATIO2000 MB2000 PPE/TA2000 ΔASSETS ΔROA ΔDERATIO ΔMB Model specification (4) (5) 0.1395*** 0.1135*** (3.83) (4.96) 0.0155 (1.58) 0.0203** (2.21) 0.0556*** 0.0451*** 0.0466*** 0.0524*** 0.0504*** (4.62) (3.32) (3.07) (4.03) (3.88) 0.0044 0.0095** 0.0116** 0.0053 (1.17) (2.01) (2.35) (1.41) 0.0093 -0.0011 -0.0745* 0.0103 (0.38) (0.04) (1.81) (0.42) -0.0296 -0.0148 -0.0077 -0.0348 (1.00) (0.47) (0.22) (1.17) -0.0003 -0.0006 (0.73) (0.67) 0.0383 0.0399 (0.87) (0.85) 0.0696** 0.0553* (2.12) (1.75) -0.2717*** -0.1978** (2.68) (2.17) -0.0292 0.0496 (0.33) (0.56) 0.0134 0.0149 (1.33) (1.48) (1) 0.1152*** (5.03) 0.0234** (2.57) (2) 0.1059** (2.46) 0.0207** (2.21) (3) 0.0759* (1.74) 0.0133 (1.31) NO 2,172 0.0136 YES 2,073 0.0209 YES 1,944 0.0248 YES 1,985 0.0211 YES 2,174 0.0126 (6) 0.0377 (0.79) 0.0194** (2.01) 0.0399*** (2.75) 0.0105** (2.19) -0.0033 (0.12) -0.0154 (0.49) -0.0002 (0.46) 0.0347 (0.78) (7) 0.0821* (1.84) 0.0160 (1.55) 0.0439*** (2.77) 0.0115** (2.31) -0.0674 (1.60) -0.0006 (0.02) -0.0008 (0.84) 0.0326 (0.70) 0.0681** (2.08) -0.2683*** (2.66) 0.0186 (0.20) 0.0140 (1.29) (8) 0.0973*** (2.98) 0.0201** (2.02) 0.0493*** (3.59) 0.0547* (1.74) -0.2006** (2.23) 0.0868 (0.94) 0.0154 (1.44) YES 1,986 0.0233   Includes industry fixed effects Observations R-squared NO 2,072 0.0218 YES 1,945 0.0261 Table 3 (continued) Relation between percent changes in 10-K page counts and cost of capital shocks Panel B reports regression results for models of percent changes in page counts for three sections of the 10-K on combinations of initial shock (INIT_SHOCK), remaining shock (REM_SHOCK), pre-event beta (BETA_PRE), and control variables (described in Panel A). The three sections of the 10-K are: MD&A (%ΔMDA); the sections that are generally narrative (%ΔDISCUSS); and the sections that include financial tables (%ΔFS). The regression models include industry fixed effects. Robust t-statistics are in parentheses. *** {**} (*) indicates significance at the 1%, 5%, 10% level in a two-sided test. Panel B: Dependent variable %ΔFS %ΔMDA -0.0248 0.2941*** (0.32) (4.30) 0.0255** (2.05) 0.0022 (0.17) 0.0526** 0.0131 (2.05) (0.60) 0.0096 0.0225*** (0.89) (2.66) 0.0020 -0.0354 (0.05) (0.72) -0.0176 -0.1444*** (0.25) (2.87) -0.0010*** 0.0001 (2.59) (0.11) 0.0949 -0.0412 (1.14) (0.71) YES 1,260 0.0142 YES 1,223 0.0517 Constant INIT_SHOCK REM_SHOCK BETA_PRE LASSETS2000 ROA2000 DERATIO2000 MB2000 PPE/TA2000 Includes industry fixed effects Observations R-squared %ΔMDA 0.3015*** (4.39) 0.0140 (1.15) 0.0176 (0.84) 0.0210** (2.51) -0.0139 (0.28) -0.1405*** (2.87) -0.0002 (0.39) -0.0372 (0.64) YES 1,228 0.0523 %ΔDISCUSS 0.1431*** (2.83) 0.0156 (1.41) 0.0056 (0.39) 0.0184*** (2.72) 0.0582 (1.61) -0.0934*** (2.60) -0.0005 (1.60) -0.0844* (1.87) YES 1,273 0.0446 %ΔDISCUSS 0.1843*** (3.45) 0.0110 (1.16) 0.0039 (0.24) 0.0183*** (2.70) 0.0475 (1.34) -0.0917** (2.56) -0.0004 (1.41) -0.0904* (1.96) YES 1,269 0.0409 %ΔFS -0.0264 (0.34) 0.0438 (1.58) 0.0571* (1.73) 0.0080 (0.66) 0.0038 (0.09) -0.0062 (0.10) -0.0009** (2.44) 0.0750 (0.90) YES 1,255 0.0175 Table 4 Cross-sectional analysis of determinants of the page changes in the 10-K filings Results of regressions on subsamples of firms with positive shocks and high sensitivity to the cost of capital and the residual subsamples. The dependent variable is the percent change in pages (%ΔPAGES). The first four columns present results partitioning the sample based on a positive initial or remaining shock and high financing needs (POS&HIGH=1). Firms are defined as having high financing needs if the firm’s cash flow from investing activities is above the sample median. The last four columns present results partitioning the sample based on a positive initial or remaining shock and high growth opportunities (POS&HIGH=1). Firms are defined as having high growth opportunities if he firm’s market to book ratio at December 31, 2000 is above the median for the sample. The regression models include industry fixed effects. Robust t-statistics are in parentheses. *** {**} (*) indicates significance at the 1%, 5%, 10% level. (1) Proxy for sensitivity: POS&HIGH =1 Constant INIT_SHOCK REM_SHOCK BETA_PRE LASSETS2000 ROA2000 DERATIO2000 MB2000 PPE/TA2000 Includes industry fixed effects Observations R-squared 0.0555* (1.96) -0.0071 (0.73) -0.1096 (0.80) 0.0702 (0.75) 0.0007 (0.67) 0.0590 (0.61) YES 458 0.0269 0.0362** (2.21) 0.0192*** (3.03) -0.0080 (0.27) -0.0139 (0.34) -0.0004 (1.32) 0.0128 (0.25) YES 1,289 0.0230 0.1112 (1.15) 0.0539** (2.17) =0 0.0748 (1.34) 0.0054 (0.46) =1 -0.0396 (0.57) 0.0281 (0.77) 0.0554** (2.10) -0.0002 (0.01) -0.0120 (0.12) 0.2101* (1.80) 0.0006 (0.51) 0.0153 (0.16) YES 512 0.0351 (2) (3) Financing needs POS&HIGH =0 0.1300** (2.43) 0.0121 (1.03) 0.0334* (1.79) 0.0157** (2.55) -0.0211 (0.67) -0.0329 (0.83) -0.0004 (1.35) -0.0237 (0.50) YES 1,234 0.0226 =1 0.1655 (1.45) 0.0611** (2.50) 0.0193 (0.81) 0.0084 (0.93) -0.0013 (0.02) -0.0041 (0.06) -0.0001 (0.21) -0.0127 (0.17) YES 548 0.0220 (4) (5) POS&HIGH =0 0.0790 (1.63) 0.0118 (1.04) 0.0488*** (2.93) 0.0105* (1.87) 0.0084 (0.29) -0.0281 (0.78) -0.0021 (1.44) 0.0573 (1.03) YES 1,525 0.0289 =1 0.0049 (0.06) 0.0297 (1.01) 0.0145 (0.68) 0.0098 (1.06) 0.0000 (0.00) 0.1069 (1.37) -0.0000 (0.00) -0.0470 (0.59) YES 588 0.0290 (6) (7) Growth opportunities POS&HIGH =0 0.1265** (2.54) 0.0184 (1.52) 0.0506*** (2.66) 0.0112** (2.01) 0.0112 (0.34) -0.0632* (1.76) -0.0024* (1.74) 0.0689 (1.22) YES 1,484 0.0288 (8) Table 5 Models of the association between the page changes in the 10-K, cost of capital shocks, and interim voluntary disclosures Results from regressions of the percent change in pages (%ΔPAGES) on the cost of capital shocks and proxies for interim voluntary disclosures. Proxies for interim voluntary disclosures are: Δ8KCOUNT, which equals the change in the number of 8-K filings for changes of -1, 0, and +1 and which equals -2 (+2) for changes less than or equal to -2 (greater than or equal to +2); a binary variable indicating the initiation of conference calls in the period after the shock (NEWCALL); the percent change in words in the fourth quarter 2001 earnings announcement (%ΔWORDS); and a market-based measure of interim disclosure, ˆ the parameter b from the switching model. All models include BETA_PRE, the control variables, and industry fixed effects. See Table 3 for definitions of the i2 control variables. Robust t-statistics are in parentheses. *** {**} (*) indicates significance at the 1%, 5%, 10% level in two-sided tests. Constant INIT_SHOCK REM_SHOCK 0.0451*** (3.32) LASSETS2000 0.0099** (2.08) ROA2000 -0.0026 (0.10) DERATIO2000 -0.0156 (0.50) MB2000 -0.0003 (0.75) PPE/TA2000 0.0386 (0.87) Industry fixed effects YES Proxies for immediate responses: 0.0114 Δ8KCOUNT (1.60) NEW CON CALL %ΔWORDS ˆ b i2 (1) 0.0448 (0.93) 0.0204** (2.17) (2) 0.0325 (0.68) 0.0197** (2.04) 0.0402*** (2.77) 0.0110** (2.28) -0.0052 (0.19) -0.0159 (0.50) -0.0002 (0.49) 0.0350 (0.79) YES 0.0148** (2.09) (3) 0.1258** (2.39) 0.0202* (1.80) 0.0499*** (3.21) 0.0042 (0.71) 0.0051 (0.14) -0.0290 (0.75) -0.0007 (0.82) 0.0368 (0.69) YES (4) 0.1290** (2.39) 0.0261** (2.16) 0.0488*** (2.88) 0.0040 (0.65) -0.0033 (0.08) -0.0199 (0.50) -0.0006 (0.60) 0.0340 (0.64) YES (5) 0.0439 (0.87) 0.0218** (2.19) 0.0475*** (3.32) 0.0104** (2.05) -0.0045 (0.14) -0.0108 (0.33) -0.0003 (0.80) 0.0380 (0.80) YES (6) 0.0891* (1.94) 0.0242** (2.35) 0.0429*** (2.81) 0.0113** (2.22) -0.0101 (0.30) -0.0107 (0.32) -0.0002 (0.56) 0.0338 (0.71) YES (7) 0.0478 (1.00) 0.0278*** (2.78) 0.0519*** (3.49) 0.0081* (1.65) -0.0012 (0.04) -0.0066 (0.21) -0.0003 (0.72) 0.0365 (0.82) YES BETA_PRE 0.0591* (1.83) 0.0551* (1.68) 0.0211 (1.07) 0.0211 (1.06) 8.3627** (2.23) Observations R-squared 2,073 0.0213 2,072 0.0224 1,570 0.0143 1,563 0.0160 1,946 0.0220 1,942 0.0238 2,072 0.0260 Table 6 Relation between percent changes in 8-K filings and cost of capital shocks Regression results for four models of change in 8-K activity on combinations of initial shock (INIT_SHOCK), remaining shock (REM_SHOCK), pre-event beta (BETA_PRE), and control variables. Δ8KCOUNT equals the change in the number of 8-K filings from November 1, 2001 to April 30, 2002 relative to the same period in the previous fiscal year for changes of -1, 0, and +1. For changes less than or equal to -2 (greater than or equal to +2), Δ8KCOUNT = -2 (+2). %Δ8KCOUNT is the percent change in the number of filings. Variable definitions: LASSETS is log of total assets; ROA is return on assets; DERATIO is long-term debt scaled by total assets; MB is the market book ratio (set = 0 for book value < 0); PPE/TA is the book value of net PPE scaled by total assets. The control variables are measured as of fiscal year 2000. The regression models include industry fixed effects. Robust t-statistics are in parentheses. *** {**} (*) indicates significance at the 1%, 5%, 10% level in a two-sided test. Δ8KCOUNT (1) Constant INIT_SHOCK REM_SHOCK BETA_PRE LASSETS2000 ROA2000 DERATIO2000 MB2000 PPE/TA2000 Includes industry fixed effects Observations R-squared -0.0010 (0.04) -0.0346*** (3.59) 0.1278** (2.02) 0.0701 (1.29) 0.0005 (0.50) -0.0296 (0.41) YES 2,121 0.0223 0.3517*** (4.15) 0.0287* (1.90) (2) 0.1791*** (3.81) -0.0226 (1.52) -0.0347 (1.55) -0.0271*** (3.09) 0.1012* (1.90) 0.0129 (0.26) (3) 0.2347** (2.44) 0.0282** (1.99) 0.0257 (1.28) -0.0109 (1.34) 0.0082 (0.16) 0.0393 (0.76) 0.0008 (0.84) 0.0576 (0.79) YES 2,121 0.0128 %Δ8KCOUNT (4) 0.1019** (2.55) -0.0198 (1.28) -0.0090 (0.53) -0.0024 (0.30) 0.0195 (0.42) -0.0172 (0.33) YES 2,223 0.0113 YES 2,223 0.0012 Table 7 Analysis of beta responses to 10-K filings Analysis of the determinants of beta responses to 10-K filings. from the model regresses the beta response (BRESPONSE = BETA_EVT - BETA_POST) on the percent change in pages in the 10-K (%ΔPAGES) and control variables. Control variable definitions: LSIZE is log of firm size (market-based), DERATIO is long-term debt scaled by total assets; and MB is the market-to-book ratio (set = 0 for book value < 0). The control variables are measured as of fiscal year 2001. The changes are measured from 2000 to 2001. The regression models include industry fixed effects. Robust t-statistics are in parentheses. *** {**} (*) indicates significance at the 1%, 5%, 10% level in two-sided tests. (1) Constant %ΔPAGES    LSIZE2001    DERATIO2001    MB2001    ΔLSIZE    ΔDERATIO    ΔMB    Includes industry fixed effects Observations R-squared -0.3011*** (2.65) 0.1047** (1.96) 0.0151 (1.63) (2) -0.4351*** (5.01) 0.1068* (1.96) 0.0163 (1.59) -0.0519 (0.58) 0.0000 (0.00) (3) -0.2376** (2.03) 0.0812 (1.50) 0.0190* (1.85) -0.0769 (0.86) -0.0031 (0.48) -0.0568 (0.91) 0.2324 (1.10) 0.0444 (0.70) YES 1,931 0.0174 YES 2,116 0.0164 YES 2,007 0.0162