The Market Response to Implied Debt Covenant Violations

Transcription

1 The Market Response to Implied Debt Covenant Violations Derrald E. Stice Doctoral Candidate Kenan-Flagler Business School The University of North Carolina at Chapel Hill Campus Box 3490, McColl Building Chapel Hill, NC November, 2010 ABSTRACT Prior research has shown that announcements of debt covenant violations are associated with a significant negative stock price reaction. I predict and find that the market is able to infer likely earnings-based debt covenant violations on an earnings announcement date using contemporaneously available public information and in conjunction with the announced earnings. These implied debt covenant violations are associated with significant negative stock price reactions on the earnings announcement date. This finding provides evidence that market participants use the information in earnings announcements to update their assessments of the probability of a future violation of an earnings-based debt covenant; this probability revision occurs some weeks before the actual disclosure of the violation. For my sample of firms with an earnings-based debt covenant, I predict and find that there is an attenuated market reaction to the subsequent disclosure of the existence of a debt covenant violation. I create a control sample of firms with debt covenants based on non-earnings accounting data. For this control sample, the negative stock price reaction does not occur at the earnings announcement date but is instead delayed until the actual explicit public disclosure of the violation. Overall, consistent with prior studies, the market reaction to news of a debt covenant violation is negative. For a sample of firms with an earnings-based covenant, I demonstrate that this news is impounded in prices at the earnings announcement date, well in advance of the formal public disclosure of the violation. My results suggest that the cost of debt covenant violations in the cross-section is higher than estimated in the previous literature.

2 1. Introduction Past research demonstrates that public announcements of debt covenant violations are associated with negative stock price reactions. These public announcements of debt covenant violations are often associated with the filing of a quarterly (10-Q) or annual (10-K) report with the SEC. Over half of financial debt covenants are based on some variant of accounting earnings. Because almost all publicly-traded firms make a preliminary announcement of earnings weeks before the SEC filing, market participants have advance access to earnings-based data that allow them to inference impending debt covenant violation announcement. In this study I investigate whether the market uses information in realized earnings to price the likelihood of a debt covenant violation (hereafter DCV). I claim that it is possible for market participants to infer the existence of an impending earnings-based debt covenant violation using earnings data reported on the earnings announcement date. I find evidence of a negative price reaction on earnings announcement dates when realized earnings imply a high likelihood of an earnings-based debt covenant violation even when there are no explicit disclosures related to DCVs on those dates. I also document the absence of a significant price response to the actual announcement of a debt covenant violation for firms that previously reported earnings that implied a high likelihood of a debt covenant violation. These results complement the findings of prior studies that report negative stock price reactions to the announcement of DCVs and suggest that the cost of debt covenant violations in the cross-section is higher than previously estimated. Beneish and Press (1993) estimate the average costs of DCVs attributable to increased interest rates and renegotiation fees are between one and two percent of the market value of equity for their sample of firms. In a subsequent study, Beneish and Press (1995) investigate the stock price reaction to a DCV disclosure. They find that announcements of technical default of 1

3 debt covenants are associated with significant -3.52% return in the three-day period surrounding announcements of debt covenant violation. They also find that the SEC filing date of the 10-K or NT 10-K is the first public disclosure of a DCV for over 60% of observations, suggesting that firms tend to wait until the latest possible date under SEC regulations to reveal the existence of an unresolved technical default. 1 Because earnings are announced an average of several weeks before financial statements are officially submitted to the SEC (see Alford, Jones and Zmijewski, 1994), it is possible that new information about possible earnings-based DCVs becomes available to investors well before the firm officially acknowledges them in an SEC filing. Consistent with this possibility Nini, Smith and Sufi (2009) report that returns are significantly negative in the months leading up to and including the SEC filing date for firms that disclose DCVs. I argue that if new information about the likelihood of DCV is revealed in prior announced earnings, then returns will be decreasing in a measure of this change in likelihood on the date of an earnings announcement. Furthermore, to the extent that this measure is positively correlated with actual DCV, I expect that the negative price response to DCV disclosures typically observed on to the SEC filing date will be attenuated. To test these predictions I construct an earnings-announcement implied covenant violation measure using the reported earnings from the announcement, information available at the earnings announcement date, and debt covenant-specific information from Dealscan. Based on reported earnings, I calculate the debt-to-ebitda ratio on the date of the announcement and 1 SEC Regulation S-X states that any breach of covenant which exists at the date of the most recent balance sheet filed and which has not been subsequently cured shall be stated in the notes to the financial statement (SEC (1988)). A more recent reiteration of this directive is reported in Sufi (2007): companies that are, or are reasonably likely to be, in breach of such covenants must disclose material information about that breach and analyze the impact on the company if material (SEC (2003)). 2

4 then compare it to the stated covenant ratio at the date of the announcement. 2 I validate the predictive ability of the measure on a large sample of firms, some of which report an actual DCV. The measure has predictive ability up to two quarters in advance of an actual DCV. Next, I examine the 3-day return around earnings announcements for firms meeting data requirements for my sample. After controlling for earnings surprises and a set of control variables employed in the prior literature, I find that announcement date returns are decreasing in the measure of implied covenant violation. Furthermore, I find no evidence of a negative stock price response to a DCV on the SEC filing date among firms for which there is a disclosed debt covenant violation. In the next section I develop my hypotheses. I describe the sample selection procedures and variables used in this study in section 3. Section 4 presents the empirical results. A summary and conclusion is provided in section Background and Hypothesis Development 2.1. Debt Covenants and the Costliness of Debt Covenant Violation Debt covenants are included in debt contracts to reduce lender risk by limiting managers ability to extract rents from debt holders 3 and by giving lenders control of the firm during bad economic states of the firm (see Aghion and Bolton, 1992). Debt holders only suffer from economic losses, and are relatively unaffected by economic gains, so they are concerned about gaining control of the firm as quickly as possible when their investment is at risk. Covenant inclusion is costly to the firm, but the commitment to turning over firm control to the lender 2 Typically this is the debt-to-ebitda ratio from the previous quarterly earnings statement, adjusted to incorporate current period earnings, because the ratio contains several components that may not be available to the market at the earnings announcement date. I discuss the assumptions underlying the estimation of these components measures and possible biases in the next section. 3 Jensen and Meckling (1976) list unwarranted distributions to shareholders, issuance of higher priority debt claims, and investments in negative net present value projects for purposes of empire building and diversification as potential actions that debt covenants attempt to prevent. 3

5 during bad states generates ex ante more favorable borrowing terms for the borrowing firm (Bradley and Roberts, 2004). While inclusion of debt covenants may grant more favorable terms to the borrower, violation is costly. 4 It is known that creditors gain influence over firm decisions in bankruptcy. 5 However, Baird and Rasmussen (2006) claim that lenders exert strong influence over firms after covenant violation, even before bankruptcy. Recent research has provided evidence that covenant violations are associated with a shift of firm control to lenders. 6 For example, Nini, Smith, and Sufi (2009) find that DCV are followed by increases in CEO turnover, increases in corporate restructurings, slowdowns in mergers and acquisitions, decreases in capital expenditures, and reductions in debt use and dividend payouts. Chava and Roberts (2008) also report that capital investment decreases after financial covenant violation. Roberts and Sufi (2009a) show that covenant violations lead to restricted availability of debt financing, and Sufi (2009) shows that DCV lead to a decrease in the availability of a line of credit. These studies provide evidence that firms that violate debt covenants transfer power to lenders, even before payment default. Once a debt covenant is triggered, lenders can choose to accelerate the loan or renegotiate the contract. Renegotiation can be costly - Beneish and Press (1993) estimate that the average costs of DCV attributable to increased interest rates and renegotiation fees are between one and two percent of the market value of equity for their sample of firms. Recently, Roberts and Sufi 4 Smith (1993) provides a review of early work in this area. 5 For example, Gilson (1990) shows that creditors become large shareholders during bankruptcy. 6 Roberts and Sufi (2009b) review recent work pertaining to shifting control rights and renegotiation in their excellent survey paper. 4

6 (2009a) have found additional evidence that covenant violations are associated with increased interest rates. DCV may also result in the costly imposition of additional covenants during the negotiation process. Core and Schrand (1999) use an option pricing framework to model firm value when firms face costs associated with DCV and test the implications of their model on a sample of thrift institutions. Interestingly, Core and Schrand provide evidence that current information about a firm can affect current equity value even if that information is not correlated with future cash flows as long as that information changes the probability of violating a debt covenant in subsequent periods. That is, an increase in the number of covenants can potentially create a larger future news information set that negatively affects firm value, even if that news provides no information about future cash flows. Consistent with DCV being costly, Watts and Zimmerman (1978) posit that managers will choose accounting methods that will decrease the probability of debt covenant violation. Several studies have found evidence consistent with this assertion. Sweeney (1994) finds that firms that are approaching debt covenant default respond with income-increasing accounting changes. DeFond and Jiambalvo (1994) examine a sample of firms that violated debt covenants and find that in the year before and in the year of the covenant violation, total accruals and working capital accruals are significantly positive. Beatty and Weber (2003) find that firms with debt covenants are more likely to adopt income increasing accounting policies than are firms without debt covenants. Beneish and Press (1995) investigate the stock price reaction to a DCV disclosure. Motivated by prior studies demonstrating the costliness of violation, Beneish and Press predict that the announcement of DCV will lead to a negative stock price reaction. Beneish and Press utilize a sample of 87 firms that violate an accounting-based DCV. They use financial statement 5

7 disclosure of DCV and news media articles as proxies for the date at which the market learns about DCV. They find that announcements of technical default of debt covenants are associated with a significant negative 3.52% return in the three-day period surrounding announcements of debt covenant violation Debt Covenant Violation Prediction The goal of this study is to explore the ability of the market to identify DCV before they are officially announced. This study focuses on firms with earnings-based debt covenants. For firms containing earnings-based debt covenants in their debt contracts, the earnings announcement date provides the market with a unique opportunity to learn about DCV without any disclosure from the firm. It has not yet been determined what the stock market response is to the release of earnings information that may inform the market about DCV before an actual announcement of a violation is made by the firm. The aim of this study is to investigate whether the market reacts when sufficient information exists to infer DCV. In order to test the market reaction to implied DCV, it is necessary to construct a measure that captures a high likelihood of covenant violation in a timely manner. Two recent studies have attempted to create measures predicting covenant violation. Dyreng (2009) develops a model of covenant violation drawing from the bankruptcy prediction literature beginning with Beaver (1966) and Altman (1968). Dyreng creates a model that assumes that covenant slack is a linear combination of accounting and market variables. Murfin (2009) creates a measure of contract strictness that attempts to capture the ex ante probability of covenant violation. Murfin incorporates four measures in creating strictness : the number of covenants, the tightness of each covenant, the scale of each covenant, and the covariance of covenant ratios. An advantage of both of these studies is that these techniques can be applied to a broad sample of firms. 6

8 The approach that I take is more restrictive. Because I am interested in examining the ability of the market to infer earnings-based DCV, I require each contract in my sample to include an earnings-based covenant. I investigate the prevalence of various covenants in debt contracts and the availability of different covenant components at the earnings announcement date. This investigation (which I will discuss in the sample selection section) leads me to choose to focus my analysis on the debt-to-ebitda covenant. In addition, the approach that I take in creating my measure of implied DCV is fundamentally different than the techniques employed in Dyreng (2009) and Murfin (2009) in one important way. I am interested in the market response to earnings news that arrives at the earnings announcement. I construct my measure by taking the contracted covenant ratio from Dealscan, information contemporaneously available, and the earnings reported at the earnings announcement date to derive an implied measure of slack. A key assumption of my measure is that it be updated at the earnings announcement. Two of Murfin s measure components, the number of covenants and the covenant scale, are static and do not change. The remaining two components, covenant slack and covenant ratio covariance, are updated quarterly using Compustat data. Because I restrict my sample to firms that report earnings before they file financial statements with the SEC, the necessary data are largely unavailable at the earnings announcement date. I discuss the data available at the earnings announcement date in the next section. I also include variables used in Dyreng (2009) as controls. The technique that I employ generates noise from at least two sources. First, the language used to specify a debt covenant varies by contract. Identically named covenants need not be identically calculated. Second, even knowing the calculation used for a specific covenant in a specific contract may not guarantee exact measurement. Variation can occur from non- GAAP accounting data certified by the CFO that is not publicly available (Murfin, 2009, Chava 7

9 and Roberts (2009), Leftwich (1983)). Because of the noise that my technique generates, I define the most extreme group of slack observations by year as the firms most likely to have violated a covenant. I discuss the construction of my measure in more detail in the next section. By constructing my measure in this fashion I am able to determine if the market is able to infer a high likelihood of DCV on the earnings announcement date. In order to test my hypotheses, I must be reasonably assured that the measure that I employ captures likely DCV. The first analyses that I perform validate the measure of implied violation. To validate the measure, I use the sample of financial-statement-disclosed financial covenant violations created by Nini, Smith, and Sufi (2009). Their sample contains all the financial covenant violations disclosed in the financial statements, not just debt-to-ebitda. The sample of implied violators constructed using my measure, may or may not end up in the Nini et al (2009) sample. One reason for inclusion in the earnings announcement implied debt covenant violation sample, and not the Nini et al (2009) sample, is that the firm obtained a waiver or came to some sort of resolution with the lender before the filing date. It could also be that the violation was not material. The Nini et al sample only includes financial covenant violations, but many of these covenants are not earnings-based, such as leverage and net worth covenants. So, a firm may disclose violation of a financial debt covenant, but this violation may not be earnings-based covenant related. This may prevent Implied_Violation from being a powerful predictor of subsequent DCV disclosure in the financial statement. Implied_Violation is a market-observable measure constructed for each firm on each earnings announcement date. If this measure has the ability to predict subsequent DCV disclosure in the financial statements, then the market has the information necessary to update beliefs about the probability of future DCV disclosure. Again, the emphasis on the probability update is important because it not certain that a firm will disclose DCV, even if the firm enters 8

10 technical default. A firm may obtain a waiver, resolve the violation with the lender, or renegotiate the contract before the filing of the financial statements and avoid disclosure. If Implied_Violation correctly identifies firms that are likely to have violated a debt covenant, then values of Implied_Violation equal to 1 should indicate a higher likelihood of future DCV disclosure in the financial statements. The measure is found to be positively associated with DCV disclosed in the financial statements up to two quarters in advance. I present these findings in the empirical results section before presenting tests of my formal hypotheses Hypothesis Development My first hypothesis is motivated by the logic used in Beneish and Press (1995). If DCV is costly, and the market is able to update the probability of DCV on the earnings announcement date, then there should be a negative price reaction to the announcement of earnings that imply a likely DCV. This hypothesis is expressed formally in the alternative form. H1: There is negative stock price reaction to the announcement of earnings that imply a violation of an earnings-based debt covenant. When earnings imply a high probability of DCV the market reacts negatively and immediately. Unlike Beneish and Press (1995), who use a sample of accounting-based debt covenants 7, it is 7 Beneish and Press conduct their tests using a sample of firms with accounting-based debt covenants. Over 90% of the 130 DCV announcements examined by Beneish and Press (1995) involved violations of one or more of the following accounting-based covenants: tangible net worth, current ratio, or leverage. Only 9 of 130 violated covenants were earnings based. Although some information about these three covenants may be indirectly conveyed through an earnings announcement, it is more direct to focus on firms with earnings-based debt covenants in order to address the question of whether the market uses earnings information to infer earnings-based covenant violation. In addition, I investigate the information released concurrently with earnings at the earnings announcement date; it is not usually possible to infer violations for accounting-based covenants such as the ones utilized in Beneish and Press. 9

11 not necessary to wait for a formal announcement from the firm or a news media article to infer DCV for my sample of earnings-based covenant firms. If the measure of implied violation is too noisy to be of use to market participants then there may be no reaction on the earnings announcement date. In addition, if the market only perceives disclosed violations to be costly then there may be no reaction. Dichev and Skinner (2002) document that covenant violation occurs fairly often, 30% of the loans in their sample, and that the most common lender response to DCV in their sample is to waive the violation. 8 If implied violations identified using my measure are likely to be waived, and the market does not view waived violations as costly, then there may be no negative reaction. The observation of an implied DCV may be associated with a positive stock price reaction. The ex post outcome is not known by the market at the earnings announcement date. At the construction of the implied violation measure, it is not known if the firm will obtain a waiver, renegotiate the loan, or go into default. In addition, there may be benefits to shareholders arising from DCV. The fact that a firm is in a position to be violating a debt covenant is usually an indication that firm performance has been poor. If shareholders believe that poor management is behind this performance, DCV may be an opportunity to make changes in the management and firm structure. Increased lender control, which reduces managerial discretion, may be optimal from the shareholders point of view. There is some evidence in Nini, Smith, and Sufi (2009) to support this positive viewpoint. The authors document an increase in CEO turnover and corporate restructuring and a decrease in capital expenditures and debt usage after covenant violations. They find that while in the months leading up to DCV firms experience significantly negative operating cash flows and 8 Dichev and Skinner (2002) also find that renegotiation is very common. Roberts and Sufi (2009c) use a large sample of private debt contracts and find that 90% of long-term debt contracts are renegotiated before maturity. 10

12 returns; these firms experience significantly positive operating cash flows and returns in the months after. NSS show that purchasing the stock of firms in the month of a DCV earns 5% more than the risk-adjusted benchmark in the 12 months following violation. So, DCV may be the opportunity that triggers the serious overhaul of the firm that leads to subsequent increased performance and will therefore be internalized by the market on the day of DCV disclosure in the financial statements. These potential benefits notwithstanding, I predict that, on average, the market reaction to news of covenant violation is negative. Consistent with the findings of Beneish and Press (1995), I expect the stock market reaction to the reporting of earnings that imply earnings-based DCV to be negative. The second hypothesis addresses the issue of firms violating covenants in more than one period. When the market is able to infer a high likelihood that a firm has violated an earningsbased debt covenant it should react immediately. This initial stock market reaction should incorporate the news of DCV, so subsequent announcements of earnings that imply DCV should be less informative. Prior studies have focused on first-time covenant violations (Beneish and Press (1995) 9, Nini et al (2009)). First time violations should capture most fully the market perceived cost of violation. However, the announcement of earnings that implies DCV for a firm with a prior period DCV may convey additional negative news. The market may perceive this news to be worse than the initial violation. If a firm violates a covenant and managers obtain a waiver or contract renegotiation from the lender with assurances of improved future performance, a subsequent violation may be perceived as evidence that the financial health of the firm is worse than managers and lenders previously thought. A firm may find it difficult to obtain waiver or 9 Beneish and Press (1995) allow their sample firms to disclose violation only once during their five year sample period. 11

13 renegotiation concessions from a lender more than once. Chen and Wei (1993) model the lender decision to waive a covenant violation or call the debt. They predict, and find empirically, that lenders are more likely to grant a waiver to firms with lower estimated probabilities of bankruptcy. Violations that occur after a waiver has been obtained in a previous period, may increase the likelihood of bankruptcy. These subsequent violations may be associated with negative stock price reactions more severe than the initial reactions. Perhaps a violation after an initial violation indicates a severe deterioration of the economic state of the firm and the market responds accordingly. All else equal, investors prefer firms to not violate debt covenants. An implied DCV is bad news; however, an implied DCV in a period after a prior implied DCV conveys less violation surprise. If deterioration of the underlying economics of the firm persists after the initial violation, I expect this deterioration to be captured in interim stock price movements and not on the day that the subsequent violation is inferred. Since the market has already reacted to the initial news of implied DCV, I predict that subsequent implied DCV will be less informative to the market. This leads to the following hypothesis stated in the alternative form: H2: The negative stock market reaction to an implied debt covenant violation will be attenuated for firms that have previously reported earnings that implied violation. This hypothesis predicts that less news will be conveyed to the market through an implied DCV if that firm has experienced a violation in a previous period. The market may be learning something about the future costs that a covenant-violating firm faces, but the market will only be reacting to the incremental costs incurred by an additional violation. I predict that these costs, on average, are lower than the costs impounded at the initial violation. 12

14 While the market may consider repeat violations to be less costly than initial violations, news from firms that indicates recovery from a previous violation may be deemed positive. Nini, Smith, and Sufi (2009) document positive abnormal returns in the months (and years) after announcement of DCV. Lender intervention may lead to increases in firm efficiency, a reduction in negative NPV projects, and decreases in value-reducing manager behavior. Thus, I predict that the market will perceive news of a covenant violation reversal positively. Violation reversals are characterized by implied or disclosed violations in one period being followed by a lack of implied or disclosed violations in the next. I express this hypothesis in the alternative form as follows: H3: There is a positive stock market reaction when a previous period implied debt covenant violation reverses. If the market perceives earnings indicating DCV reversal to be positive then a positive stock price reaction should be observed. On the other hand, if the market is uniformed by earnings suggesting DCV recovery, then no reaction will be observed. The primary objective of this study is to extend the findings of previous papers documenting a negative stock price reaction to the announcement of DCV. I identify a group of firms for which it is possible to infer DCV without any disclosure from the firm. For these firms it is unnecessary to wait for an official announcement of DCV. Beneish and Press (1995) find that announcements of technical default of debt covenants are associated with significant stock price declines. In their sample, the SEC filing date of the 10-K or NT 10-K represents the disclosure date for over 60% of observations. Their results suggest that the majority of DCV announcements are coincidental with the balance sheet filing date - an empirical fact that I verify 13

15 in the sample employed in this study. 10 The evidence suggests that firms avoid making disclosures about technical default until they are required to do so. Beneish and Press employ a sample of firms that violated an accounting-based debt covenant. The sample of Beneish and Press contained less than 7% of violated covenants relating to earnings-based debt covenants. Beneish and Press predict and find that there is a negative stock market reaction on the date in which a covenant violation is disclosed. For my sample of earnings-based covenant firms, I predict that it is unnecessary to wait for the financial statements to be filed, the market will react on the day that earnings imply a high likelihood of violation. For these firms, the announcement effect of a violation will be smaller than for firms for which it is more difficult to infer DCV on the earnings announcement date. This leads to my final hypothesis, stated in the alternative form. H4: The negative stock price reaction to an actual announcement of a debt covenant violation will be attenuated for firms for which it was possible to infer debt covenant violation at the earnings announcement date. 3. Sample and Variable Definition 3.1. Covenant Choice Motivation My empirical strategy is to investigate whether the market identifies and reacts to earnings announcements that imply a debt covenant violation. I focus my analysis on one particular earnings-based debt covenant, debt-to-ebitda. I choose to focus my analysis on debt-to-ebitda because it is the most common debt covenant in Dealscan, a dataset of private 10 I investigate a random sample of over 100 firm quarters for which a violation is disclosed in the financial statements. A news media article disclosing violation precedes the financial statement filing in approximately 3% of my hand collected sample. This low financial statement preemption rate mitigates concerns that focusing on the SEC filing date is too restrictive. In addition, robustness tests conducted at the end of this study on firms with accounting-based covenant provide evidence that this assumption may be reasonable. 14

16 debt agreements created and offered by the Thomson Reuters Loan Pricing Corporation (TRLPC). Debt-to-EBITDA is an included covenant in almost half of all loan agreements with financial covenants. It may be possible that firms disclose information other than earnings on the earnings announcement date that can be used by market participants to infer DCV for other financial covenants, not only earnings-based covenants. Prior research has shown an increase in concurrent disclosures (Francis, Schipper, and Vincent, 2002). These additional disclosures may allow violations for other covenants to be inferred. To investigate this possibility, I hand collect 50 random earnings announcements from my sample of firms and check for the availability of covenant component data for several widely used debt covenants. 11 To establish that the earnings announcement date reported numbers were reliable, I then compared the reported covenant components to the numbers reported in the firms subsequently filed financial statements. 12 These results are presented from Table 1. Panel A of Table 1 indicates that 30% of the earnings announcements in my random sample provided no covenant component information in addition to earnings. The most common covenant component disclosed was tax expense, and it appeared in almost two-thirds of the earnings announcements. Interest expense was the second most commonly disclosed component, appearing in half the announcements. EBITDA was explicitly disclosed in 12% of announcements. The mostly commonly reported non-earnings covenant information was current assets and liabilities. These components appeared in 38% of the sampled earnings 11 These covenants are listed in the notes for Table I also checked each earnings announcement for news of a debt covenant violation. For firms that would eventually disclose a covenant violation in their financial statements, I never found disclosure of a violation in the earnings announcement. 15

17 announcements (always together). Debt and equity appeared in almost a third of announcements, but not always in the same announcement. Covenant components pertaining to capital expenditures, tangible assets, and cash holdings appeared in almost no earnings announcements. Panel B of Table 1 compares the disclosed components to the actual components appearing in the financial statements. Different components are reported in the financial statements for 21% of the sample. For the firms that reported the same components in the earnings announcement as in the financial statements, there was an aggregation or scale change in 53% of the observations. For example, interest income and expense was commonly netted in the earnings announcement. Additionally, numbers in the earnings announcement were occasionally reported on a per share basis or in very large units (millions) in a way that made exact calculation of covenant ratios difficult. Overall, Table 1 provides an indication that while other information is reported in the earnings announcement, it is not common. 13 Most importantly, it would be inappropriate to assume that other covenant information is available at the earnings announcement Implied Violation Variable Definition To test the hypothesis of whether the market is able to use earnings information disclosed on the earnings announcement date to predict subsequent disclosure of DCV it is necessary to construct a measure of implied violation. I construct an earnings-announcement implied covenant violation measure using the reported earnings from the announcement, information available at the earnings announcement date, and debt covenant specific information from 13 Table 1 also provides an indication that other covenant components are available at the earnings announcement. In future versions I hope to collect other covenant component information available at the earnings announcement to strengthen the results of the current version. 16

18 Dealscan. The measure is constructed at the earnings announcement date, which I require to take place before the SEC filing date. In my sample, the SEC filing date is an average of 19 days after earnings are reported. In order to ensure that the measure created can be used by market participants at the earnings announcement, it is critical that all components be readily available to market participants. There are two components needed to construct a measure of implied violation, debt-to-ebitda and the initial covenant ratio. The debt-to-ebitda ratio contains several components that may not be available to the market at the earnings announcement date. While some firm managers may make debt, interest, taxes, depreciation, amortization, or EBITDA available at the earnings announcement, the majority of firms do not. To ensure no look-ahead bias, I assume that only earnings are reported at the earnings announcement date in constructing my measure of implied DCV. I use prior period values for all components of debt-to-ebitda except earnings. Thus, on the earnings announcement date, reported earnings can be added to the implied violation measure to create a new value of implied violation. The second component needed to calculate my measure of implied violation is the contract-specific covenant value from the debt agreement. This contracted covenant value remains constant for each firm until a new debt issue is available. If a different covenant threshold is stated in the new debt agreement, the value changes to reflect the new stated contract value. Taken together, the two components create a firm-specific measure of implied violation that is updated quarterly and is constructed as follows: IDCV t = Cov_Ratio t (LTDt -1 + CurrLTD t-1 / NI t + Interest t-1 + Tax t-1 + DepAmor t-1 ) (1) 17

19 where IDCV t is implied debt covenant violation for quarter t constructed on the earnings announcement date for quarter t. Cov_Ratio is the maximum allowable value of debt-to- EBITDA for firm i in quarter t before a technical violation occurs. LTD t-1 is long-term debt, and CurrLTD t-1 is the current portion of long-term debt for the previous quarter. Interest t-1, Tax t-1, and DepAmor t-1 are the interest, tax, depreciation, and amortization components of EBITDA and are all taken as the values from the previous quarter. This methodology allows me to compute an implied debt covenant ratio for each firm at the earnings announcement date. 14 While IDCV uses reported earnings, contemporaneously available data, and debt agreement covenant values, it is not clear that the measure will have the ability to predict subsequent DCV disclosed in the financial statements. This inability may occur for several reasons. First, the use of quarter old accounting data may limit the ability of the measure to predict subsequent disclosure. Second, the measure may be too coarse since many debt contracts use transformed values of GAAP, not actual GAAP values (Murfin, 2009, Chava and Roberts (2009), Leftwich (1983)). These concerns will bias against the ability of IDCV to predict subsequent DCV disclosure in a statistical power sense. To mitigate some of the effects of the noise in IDCV, I rank IDCV and create an indicator variable that is equal to 1 if IDCV is in the most extreme quintile in a given year. This new variable, Implied_Violation, takes a value of 1 for the observations that most likely represent a covenant violation. My measure is susceptible to both Type 1 and Type 2 errors. I may be incorrectly labeling non-violating firms as violating and vice versa. I have no reason to believe that this misspecification will be systematically related to abnormal returns. 14 Negative values of EBITDA can produce large positive values of implied slack. Negative values of EBITDA occur in 4% of firm-quarter observations. Inferences do not change when these observations are removed. 18

20 3.3. Data and Sample Selection The private debt contracts in this sample represent a large source of corporate finance. Sufi (2007) reports that 90% of the 500 largest nonfinancial firms in COMPUSTAT obtained a loan through this channel between 1994 and The market for these loans grew to over $1 trillion by the end of his sample period and to over $1.5 trillion by 2005 (Bharath, Sunder, and Sunder (2008)). I use all the debt issues to public firms that have a debt-to EBITDA covenant. Dealscan provides a unique package identification number for each debt issue as well as a company identification number and the stock ticker. I conduct my tests at the firm-deal level. 15 I match these tickers with the COMPUSTAT and CRSP databases to create a dataset that includes all the loan information from Dealscan and all the financial statement information from COMPUSTAT and returns data from CRSP. I require each debt issue observation to have all the required COMPUSTAT and CRSP data. I truncate earnings at the 1 st and 99 th percentiles. I include all the firm-quarters during the range for which I have loan data that have non-missing COMPUSTAT and CRSP data for each of the firms with at least one private debt issue containing a debt-to-ebitda covenant. After constructing a measure of implied violation, I validate the measure using a sample of known violators. This validation allows me to use the measure to observe the market reaction to the announcement of earnings to imply an earnings-based DCV. The sample of known violators is obtained from Greg Nini, David Smith, and Amir Sufi who construct a sample of disclosed covenant violations in Nini, Smith, and Sufi (2009). In addition, the authors provide 15 The deal-level analysis decision is consistent with prior research and motivated in two ways. First, syndicated loan contracts are drafted at the deal level. All covenants and lenders are listed together on this contract, regardless of the number of facilities (loans or lines of credit), so this is the relevant unit of observation. Second, analysis conducted at the facility level biases standard errors downward because the same deal is treated as multiple independent observations. (Sufi (2007), Murfin (2009)) 19

21 the SEC filing date for their sample. I collected all the available financial statement filing dates from the SEC website using Perl and verify the dates provided by Nini et al (2009). The authors make their data available. 16 Nini, Smith, and Sufi (2009) construct an indicator variable each quarter that is set to 1 if the firm discloses DCV in the financial statements. These data cover the range 1997 to 2007, beginning when firms were required to electronically file financial statement with the SEC. Observations within the financial industry are excluded. The authors create the universe of DCV that are disclosed in the financial statements during their sample period. I require sample observations to have all necessary Dealscan, COMPUSTAT, CRSP, and disclosed violation data. In addition, because my empirical design relies crucially on the ability of the market to infer DCV from reported earnings before firm disclosure of violation in the financial statements, I require the SEC filing date to occur after the earnings announcement date for all observations. The final sample consists of 1,354 debt issues in Dealscan from 1997 to These 1,354 debt issues involve 716 unique firms. Descriptive statistics describing the sample selection process can be found in Table Descriptive Statistics Table 3 provides descriptive statistics for the sample data. Panel A of Table 3 reports that sample firms have an average of 2.77 financial covenants per debt issue. The minimum allowable number of financial debt covenants is one, because all debt agreements must have at least a debt-to EBITDA covenant in order to be included in the sample. The maximum number of financial covenants in this sample is seven. The average loan size in the sample is $434M, 16 I gratefully acknowledge the authors (Greg Nini, David Smith, and Amir Sufi) for making these data public. The data can be found on Amir Sufi s website. Please refer to the appendix in Nini et al. (2009) for more information about how the sample was collected and how it can be interpreted. 20

22 and the average interest-spread is 211 basis points. The average contracted covenant value for debt-to-ebitda is Panel B of Table 1 report provides descriptive statistics for the variables used in the DCV prediction model as well as the abnormal returns specifications. Covenant violations (VIOL) are reported in approximately 5% of the quarterly financial statements for the sample firms. This compares to the 7% of firm-quarter observations with a covenant violation found by Nini, et al (2009). Implied violations (Implied_Violation) occur in 20% of firm quarters. This rate is not surprising since the variable is defined as the most extreme quintile of implied covenant slack. The average quarterly earnings for the sample are just over $10M, and the sample firms experience losses in 18% of firm quarters, this is slightly less than the 25% documented by Hayn (1995). Average assets are $1,990M, and the average return on assets is approximately 1%. The firms contained in the sample are clearly large. The average current ratio and interest coverage ratios are 1.92 and respectively. 17 I estimate abnormal returns as the three-day average market model residuals around the event date, here the earnings announcement and the financial statement filing date, following the technique of Collins, Li, and Xie (2009). Firms experience an average earnings announcement day abnormal return of 0.5%. The average financial statement filing date abnormal return is -0.1%. On average, the financial statements are filed days after the earnings announcement. Panel C provides a correlation matrix. In the univariate, disclosed violations are negatively associated with earnings announcement abnormal returns, interest coverage, earnings surprise, return on assets, and the natural log of assets and positively associated with losses. In general, implied earnings-based debt covenant violations are correlated in the same direction as 17 Sample firms used in Dyreng (2009) have a current ratio of 2.02 and an interest coverage ratio of

23 disclosed violations. Notable exceptions, in the univariate, are earnings surprise and the natural log of assets. This may capture some of the fundamental differences between earnings-based debt covenant firms and other covenant firms. Most importantly however, disclosed violation and implied violation are positively correlated, providing some indication that the measure of implied violation is a reasonable proxy for actual violation. 4. Empirical Results 4.1. Implied Violation Measure Validation Before testing my hypotheses, I attempt to validate my measure of implied DCV. To estimate the predictive power of Implied_Violation I estimate the following logistic regression: VIOL t+δ = α 0 + α 1 Implied_Violation t + α k CONTROLS t + γ t (2) where VIOL is an indicator variable that = 1 if a firm i disclosed a debt covenant violation in its financial statements for quarter t and = 0 otherwise and δ takes on the values of 0 to 4. Implied_Violation is an indicator variable that = 1 for the most extreme quintile by year of implied covenant slack, constructed on the earnings announcement date for quarter t and = 0 otherwise. If Implied_Violation correctly identifies firms that are likely to have violated a debt covenant, then values of Implied_Violation equal to 1 should indicate a higher likelihood of future DCV disclosure in the financial statements, and I should find a positive coefficient on α 1. Focusing on the likelihood of DCV is important. Regardless of whether the firm is in actual technical violation, since a violating firm may obtain a waiver or renegotiate before filing financial statements it does not necessarily ever need to disclose violation in its financial statements. 22

24 Table 4 presents the results of estimating Eq. (2). The coefficient on Implied_Violation is positive and significant, indicating that covenant violations implied at the earnings announcement date increase the probability of firms disclosing a covenant violation in the subsequently filed financial statements. This finding provides evidence that the market has the ability to use information that is available at the earnings announcement date to deduce an earnings-based covenant violation. It is not necessary to wait for an announcement from the firm disclosing covenant violation. The firm-specific characteristics included in the specification attempt to control for the other predictors of DCV disclosure. Many of these variables have significant explanatory power. Larger firms and firms with a higher market-to-book ratio are less likely to disclose DCV in the financial statements. Perhaps larger and more established firms have more ability to negotiate with lenders to obtain a waiver or a renegotiated contract. The probability of DCV disclosure decreases with current and last period ROA, although the last period s ROA is not statistically significant. I also include several of the variables included by Dyreng (2009) that control for other common debt covenants, such as current ratio and interest coverage covenants. 18 Nini et al (2009) reports that the 10-K is often a catch-all report in which firms report information that is not reported in the shorter quarterly reports. I include quarter fixed effects in addition to year fixed effects to control for differential quarterly reporting, and I cluster robust standard errors at the firm level. Table 4 presents results indicating that the Implied_Violation measure is also able to predict disclosure of DCV in the financial statement up to two quarters ahead. Specifications 2 18 Dyreng (2009) also controls for leverage. This variable has very little explanatory power for my sample of firms with a debt-to-ebitda covenant, so I omit it. I include leverage as a control variable for the additional sample of current ratio covenant firms. These firms have a higher incidence of leverage covenants in their debt contracts. 23

25 through 5 present logistic regression results for VIOL t+1 through VIOL t+4. Implied_Violation has a positive and significant coefficient in the t+1 and t+2 specifications, but while positive in the t+3 and t+4 specifications, the coefficients are not significant. The magnitude of the coefficients monotonically decreases over time, consistent with the predictive power decreasing over time. The results provide evidence that Implied_Violation has the ability to predict future (19 days later) DCV disclosure on the earnings announcement date. 19 The distinction between implied DCV and DCV disclosure in the financial statements is important. At the earnings announcement date, the market observes earnings and infers a likelihood of earnings-based covenant violation. Overall, the results of Table 4 are consistent with Implied_Violation being positively associated with disclosed DCV. The coefficient on the indicator variable for whether the firm experienced a loss in the quarter is significant and negative across most specifications. Jiang (2008) investigates the effect of beating earnings benchmarks on a firm s cost of debt. One of the main takeaways of that study is that the benefits of beating earnings benchmarks (zero earnings, last year s earnings, and analysts forecasts) are different in the debt market than in the equity market. Jiang finds that the loss benchmark is the most important in the debt market. This finding may partially explain the strong effect of the loss variable in this specification Market Reaction to Implied Debt Covenant Violations The first hypothesis predicts that implied covenant violations will be associated with negative stock price reactions. To test this hypothesis I estimate the following OLS regression: EA_CAR t = β 0 + β 1 Implied_Violation t + β k CONTROLS t + ε t (3) 19 The measure of implied violation correctly identified 155 of these violations. Random assignment of firms into implied violation status would have yield 95 violators. 24