Does Existing Debt Covenant Tightness Affect Leverage: Evidence from SFAS 160 during the Financial Crisis

Transcription

1 Does Existing Debt Covenant Tightness Affect Leverage: Evidence from SFAS 160 during the Financial Crisis Moshe Cohen Unaffiliated Sharon Katz* Columbia Business School Sunay Mutlu Kennesaw State University Gil Sadka University of Texas at Dallas October 2016 ABSTRACT: We use an exogenous accounting-based shock to the distance to covenant violation to explore the relation between debt covenant tightness, leverage, and investment. We find that the shock to covenant tightness led to an increase in leverage. The increase was highest for firms that were close to violating the affected covenants and otherwise financially sound. We also examine how the additional debt affected firms financial behavior during the financial crisis and find that it did not result in an increase in investments or cash but rather was associated with lower profitability and lower likelihood to enter default. JEL classification: G01, G30, G31, G33, M21, M41 Keywords: Debt, covenants, financial constraints, leverage, investments, default * Corresponding author: Sharon Katz, Columbia Business School, Uris Hall 3022 Broadway, Room 605A, New York NY sk2280@columbia.edu, Phone: We would like to thank Anna Costello, Fabrizio Ferri, Trevor Harris, Robert Herz, Laurie Hodrick, Colleen Honigsberg, Alon Kalay, Steve Kaplan, Doron Nissim, Daniel Paravisini, Michael Roberts, Amit Seru, Jacob Thomas, Regina Wittenberg-Moerman, and seminar participants in the American Accounting Association Annual Meeting, Columbia Business School Finance Lunch Seminar and the Burton Workshop, Colorado Summer Accounting Conference, Drexel University, Midwest Finance Association Annual Meetings, Pennsylvania State, Tel-Aviv University, Five-Star Conference on Research in Finance at New York University, University of California San Diego, University of Minnesota, University of Texas at Dallas, the Western Finance Association Annual Meeting, and Yale University for valuable comments and suggestions. We would like to thank Timothy Scully, Paul Tylkin and Ayung Tseng for research support. We gratefully acknowledge financial support received from the Program for Financial Studies at Columbia Business School. Any errors are our own.

2 1. Introduction Debt covenants are ubiquitous in public and private financial contracts and private equity (Bradley and Roberts 2003; Kaplan and Stromberg 2003). They serve to address conflicts between debtholders and stockholders (e.g., Tirole 2006; Jensen and Meckling 1976) by pledging control rights to creditors in the event of a covenant violation. Prior empirical work shows that covenant violations occur frequently and that firms alter their corporate financial policies toward the preferences of debtholders following a covenant violation (e.g., Chava and Roberts 2008; Dichev and Skinner 2002; Nini, Smith, and Sufi 2009, 2012; Roberts and Sufi 2009a, 2009b). Absent from the literature is a study of how covenants affect the firm s leverage before violation. Graham and Leary (2011, p. 338) note: Among the open questions is the degree to which contract features such as covenants and renegotiation affect debt-equity choices. For example, Roberts & Sufi (2009b) show that covenant violations trigger control rights changes and thereby influence financing choices ex post. It would be interesting to know whether the fear of losing control rights, or limits on renegotiation, drive firms to non-debt financing ex ante. Research on the costs and benefits of leverage focuses on tax and bankruptcy considerations, information asymmetry, market timing and credit supply. (See Stein (2003) and Graham and Leary (2011) for detailed surveys.) 1 However, there is an open question about whether debt covenants per se explain some of the (unexplained) cross-sectional variation in leverage. Prior studies, such as by Beneish and Press (1993), demonstrate that, on the one hand, 1 Trade-off theories highlight a number of benefits to debt, such as Jensen s (1986) free cash flow, Myers and Majluf s (1984) low private information sensitivity and the pecking order, and a number of costs to debt, such as risk shifting, Myers s (1977) debt overhang, inefficient liquidations, and bankruptcy costs. 1

3 lenders do not exercise control rights over some technical covenant violations but that, on the other hand, some of them do penalize financially sound firms when these firms violate covenants. But barring technical violations, prior studies cannot differentiate between covenant violations and financial distress, which are highly correlated. Thus it is not clear whether a financially sound firm will restrict its leverage to avoid a covenant violation or because it foresees financial difficulties. The challenge in testing this question empirically is to find a shock to the distance to covenant violation (before a technical default) that does not relate to the firm s true financial position. This paper identifies and exploits such an exogenous shock to covenant slack and studies how it impacts firm leverage choices while not changing firm fundamentals. We also study how the resulting increase in leverage impacts firm investment, cash accumulation, earnings, and probability of default. We exploit the Statement of Financial Accounting Standards No. 160 (SFAS 160) accounting change, which aims to harmonize U.S. accounting standards with international ones. The change took effect in December 2008 and reclassified minority interest (the minority stake in an acquired subsidiary) as equity on firms balance sheets. 2 As we explain in Section 2.1., the reclassification exogenously increased the shareholder s equity reported on the balance sheet. Thus firms that had contracts with covenants stipulated in terms of balance sheet equity and minority interest were affected. For example, firms constrained by a maximum debt-to-equity covenant could hold more debt since their balance sheet equity increased. Other affected covenants include the leverage, net worth, tangible net worth, and debt-to-tangible-net-worth 2 Firms with minority interest are those that acquired the majority but not the entirety of other firms, thus creating a minority stake in the acquired subsidiaries that is reflected on their balance sheets. For example, when firm A acquires 80% of firm B, the 20% of firm B not owned by firm A is minority interest, which is recorded on firm A s balance sheet. 2

4 covenant ratios. Note that this reclassification of minority interest into equity does not change an affected firm s real financial position. We acknowledge, however, that since this accounting change occurred in the heart of the financial crisis, our findings may not generalize beyond this period. We test whether firms that received additional covenant slack, due to SFAS 160, without a change to their financial position, increased their leverage. We employ a difference-indifference-in-difference methodology. Our model controls for the appearance of the new accounting standard, for the differing behavior of affected firms (e.g., firms affected by covenants stipulated in terms of balance sheet equity) and for the differing behavior of firms with minority interest. Specifically, we examine how SFAS 160 s reclassification of minority interest differently had an impact on affected firms with minority interest while controlling both for nonaffected firms with minority interest and for affected firms that did not have minority interest. Finally, by using firm fixed effects, the model also controls for how affected firms with minority interest behaved prior to experiencing the shock. Consistent with the conjecture that covenants affect leverage, we find that firms with minority interest and equity-based covenants (henceforth referred to as affected covenants ) increased their leverage, relative to a host of matched control groups, and that the strength of their response increased with the size of their minority interest. The strongest response was for firms that were close to violating their equity-based covenants (henceforth constrained firms ) with minority interest. Our results imply that these firms increased their leverage, relative to the benchmarks, by approximately 2.6%. This shows that debt covenants constrain firms leverage throughout the life of the debt contract. Our findings survive a host of controls and firm and period fixed effects as well as a battery of robustness checks. For instance, excluding all firms 3

5 with frozen-gaap, a specific exclusion of minority interest, or both from the relevant covenants little affected our findings, due to the rarity of such instances in our sample. 3 Our hand-collected sample reveals that the increase in leverage for constrained firms mainly comes from an increase in borrowing under existing short-term and long-term credit facilities (revolvers), consistent with these firms exploiting the increase in covenant slack under existing contracts that the new standard produced. Using revolver information in the Capital IQ database, we further document that only firms with affected covenants, including constrained firms, drew on their credit facilities, while control firms decreased their borrowing. The findings also highlight that our results are driven largely by covenant constraints rather than a period effect. On average, leverage declined for control firms during this period of financial crisis and declined even more for both constrained firms without minority interest and unconstrained firms with minority interest. In contrast, we report an increase in leverage only when debt-covenant slack increased as the result of SFAS 160, adding support for the conjecture that covenants per se affect leverage decisions before violations. As noted above, prior studies, such as by Beneish and Press (1993), demonstrate that lenders do not exercise control rights over some technical covenant violations. Therefore, if a firm is otherwise financially sound, it is not obvious a lender will restrict the firm s behavior. However, barring technical violations, prior studies cannot differentiate between covenant violations and financial distress (which are highly correlated). Hence we also examine how financial distress affects firms response to our exogenous shock. We expect that financially distressed firms measured by the Whited and Wu (2006) index and probability of default were less likely to increase their leverage in response to the relaxation of the debt covenant 3 Under frozen GAAP, the covenants reflect the accounting rules as they were at the time the contract was signed. 4

6 constraint to avoid further deterioration in their financial position and inefficient liquidation. We find that they did not increase leverage when SFAS 160 increased their covenant slack. We also find that treated firms with a credit rating were more likely to increase their leverage. These results confirm that firms consider all costs and benefits of leverage, including the potential surrender of control rights, following a covenant violation. 4 The results suggest that, for financially sound firms, covenants were ex post inefficient, likely since contracts are incomplete (Aghion and Bolton 1992), and due to significant costs associated with technical default and/or renegotiation. These findings complement the results of Beneish and Press (1993), who show that technical default can impose costs also on financially sound firms. We also test the impact of the increased debt on corporate financial behavior. The finance literature has long debated whether financial frictions restrict corporate financial activity. In other words, do firms forgo profitable investments due to their inability to raise capital as a result of these frictions? The difficulty in the literature is that a firm s financial position depends on its investment opportunity set. Thus it is difficult to determine whether financial frictions or poor investment opportunities are driving the lack of investment. This question was a focus of economic and political debate during the financial crisis. 5 The effect of an intervention in market-based lending was at the heart of the debate. Specifically, many commentators saw a link between increased bank lending and increased investments and economic growth and thus concluded that bank lending should be encouraged. Others would have afforded more discretion to the banks and trusted their lending expertise. 4 However, we do not find that firms with more tax benefits to gain from increasing their leverage respond more strongly (employing the methodology of Graham (2000)). 5 Obama: Time for banks to boost lending : [G]iven the difficulty business people are having as lending has declined and given the exceptional assistance banks received... we expect them to explore every responsible way to help get our economy moving again. (Associated Press, Dec. 14, 2009) 5

7 To examine this question empirically, we require a shock to financial frictions that does not also relate to the firm s investment opportunity set. SFAS 160 provides such a shock. In contrast to findings in earlier research (Fazzari, Hubbard, and Petersen 1988; Rauh 2006; Blanchard et al. 1994), we find that firms did not use the additional capital to increase investments or hoard cash but rather spent it on their operations. We find that these decisions reduced probability of default, potentially preventing or delaying inefficiencies such as fire sales. As we discuss below, our results show that increased financial slack can have differing effects across firms and time. Our focus is on constrained firms during an unusual period, a financial crisis, when investment opportunities were not abundant. This may help explain the difference between our results and those of prior studies. During growth periods, financially sound firms might have benefitted from more access to capital and this access might have boosted macroeconomic activity. In this paper, we show that the increase in lending caused by an exogenous shock during the heart of the crisis did not have the real effects found in other work, which is consistent with banks not lending more during the crisis due to limited investment opportunities. We contribute primarily to the capital structure literature. We use a natural experiment to show that debt covenants impact leverage decisions not only upon violation of covenants but also before technical violation, due to the potential transfer of control rights. 6 Our shock to covenant slack does not relate to firms financial position. By examining the response to such a shock, we can examine the role of the covenants themselves. We thus add to the determinants of the crosssectional variation in leverage by focusing on the impact of debt covenants on leverage (cf., Bradley, Jarrell, and Kim 1984; Titman and Wessels 1988; Rajan and Zingales 1995; Fama and 6 Compare also Beneish and Press (1993, 1995a, 1995b), Chen and Wei (1993), Sweeney (1994), Dichev and Skinner (2002), Roberts and Sufi (2009b), and Nini, Smith, and Sufi (2009). 6

8 French 1995; Harris and Raviv 1991; Frank and Goyal 2008; Parsons and Titman 2008; Graham 1996; Mackie-Mason 1990). Our findings imply that firms consider the costs of inefficient investments, liquidations, and bankruptcy that could result from covenant violations. Our empirical analysis shows that additional covenant slack results in higher leverage and a lower probability of default, supporting trade-off theories. By examining the impact of a shock to leverage on firms financial behavior, we also contribute to the literature on the economic costs of financial frictions. This literature documents that financial frictions influence corporate financial activity, including investment and dividend policy (Fazzari, Hubbard and Petersen 1988; Rauh 2006; Blanchard et al. 1994), spending on nonprimary segments (Lamont 1997), the accumulation of cash and asset sales, and the use of lines of credit (Campello, Graham, and Harvey 2010; Campello et al. 2011). 7 The challenge faced by this work is the endogeneity in the determination of both the availability of external financing and the reaction to any change in it. The investment opportunity set is jointly determined with the availability of external financing, debt maturity, and covenants (Billet, King, and Mauer 2005). As such, firms may have access to limited external financing precisely because their investment opportunity set is poor. This paper adds to this literature by exploiting a new exogenous shock to firms available capital as well as by examining this question in the heart of a financial crisis. We acknowledge, however, that our findings may not generalize beyond this period. 8 Lastly, since SFAS 160 took effect in December 2008, we contribute to the recent line of work studying the impact of the limited availability of capital to firms during the crisis. Prior 7 See Stein (2003) for a survey. 8 To further address the issue of our specific period, we ran our analysis during different period in 2000 (another recession) and did not find a placebo effect. 7

9 work documents that firms with severely limited external financing draw down their lines of credits as liquidity insurance (Ivashina and Scharfstein 2010). In addition, increases in credit lines resulted in increases in investments primarily for cash-rich firms (Campello, Giambona, Graham, and Harvey 2011). Our findings extend this literature by showing that significantly constrained firms do not increase investment or cash holdings but rather pour capital into their operations to shield themselves from default. These results suggest that banks curtailed lending due to poor investment opportunities and loss prevention. The rest of this paper is organized as follows. Section 2 provides the background for the SFAS rule and explains its incidence on firms. Section 3 discusses the data and sample construction. Section 4 details our methodology. Section 5 discusses the empirical findings and their implications. Finally, Section 6 concludes and discusses future research. 2. SFAS 160: Motivation and Implications In December 2007, the Financial Accounting Standards Board (FASB) issued Statement of Financial Accounting Standards No. 160 (SFAS 160) to modify the treatment of noncontrolling/minority interest in a consolidated entity. Before SFAS 160, the minority interest was reported in either the liabilities or in the mezzanine section between the liabilities and equity. As of December 15, 2008, firms must report the minority interest, now termed noncontrolling interest, in the equity section of the balance sheet. (The exposure draft was issued in 2005). 9 The motivation for the rule was a desire to improve the relevance, comparability, and 9 To be precise, only nonredeemable, noncontrolling minority interest is included as equity. Redeemable noncontrolling minority interest is a liability (since those possessing it have the right, and the firm has the corresponding liability, to convert them) and therefore remains in the liabilities or mezzanine section of the balance sheet. However, since no distinction was made before SFAS 160, we are forced to use the entirety of the minority interest. 8

10 transparency of the financial information that a reporting entity provides in its consolidated financial statements by establishing accounting and reporting standards for the noncontrolling interest in a subsidiary and for the deconsolidation of a subsidiary (SFAS 160). The rule was passed as part of an overall move by the FASB and International Accounting Standards Board (IASB) to harmonize accounting standards. SFAS 160 coincided with SFAS 141R, which changed the accounting for mergers and acquisition. To address the potential confounding effects of SFAS 141R, we included, in unreported results, the change in minority interest post-2008 as an additional control variable. The change in minority interest will capture the effects of any new acquisitions and specifically the effects of fair value measurement of noncontrolling interest on the balance sheet. Our conclusions are not affected by including this control variable in our models. The effect of this change can easily be illustrated using the balance sheet of the AES Corporation. In 2008, the AES Corporation reported a minority interest of $3.418 billion, total stockholders equity of $3.669 billion, and $ billion of total debt, resulting in a debt-toequity ratio of Its minority interest of $3.418 billion (of which $3.358 billion was a noncontrolling interest relevant to the accounting change) was in a mezzanine section of the balance sheet and not included in the equity tally. After the accounting change took effect at the end of 2008, the restated 2008 balance sheet had a total equity of $7.027 billion (the sum of the $3.358 billion minority interest and the $3.669 billion of shareholder s equity), reducing the debt-to-equity ratio to The potential impact of this change on leverage ratios did not go unnoticed in the accounting literature, both before the rule was passed and afterward (e.g., Urbancic 2008; Mulford and Quinn 2008; Leone 2008; Deitrick 2010). In addition, the FASB exposure draft 9

11 noted that some banks had concerns as well. For example: Wells Fargo and ACLI questioned the usefulness of the proposed classification of equity. They expressed concern about the impact that classifying noncontrolling interests in consolidated equity will have on key financial and performance ratios (FASB Exposure Draft, Comment Letter Summary) Was the Accounting Rule Accounted For? Covenants generally do not automatically readjust in response to mandatory accounting rule changes. The decision whether to adjust them impacts the interest rate on the loan (Beatty, Ramesh, and Weber 2002). Historically, covenants used rolling GAAP as opposed to frozen GAAP (Leftwich 1983; Gopalakrishnan and Parkash 1995; Christensen, Lee, and Walker 2009). 11 Under rolling GAAP, the covenants reflect the accounting rules as they are at the time, as opposed to the way they were at the time the contract was signed. A recent study (Christensen and Nikolaev 2013) documents that contracts increasingly include an option to freeze the GAAP. When this option is exercised, mandatory accounting changes do not affect debt covenants. In the context of SFAS 160, exercising this option is costly since, as mentioned, SFAS 160 coincided with SFAS 141R, which changed the accounting for mergers and acquisitions. Freezing the GAAP would therefore require two different valuations for every M&A transaction and impose significant record-keeping costs. 12 We note that, even if some banks had requested to freeze the GAAP, they would have had to renegotiate. We verified our conjecture empirically by hand-collecting and reading the debt contracts. 10 See for a summary of the comment letters associated with the SFAS 160 exposure draft. These concerns were likely related to the power the rules afford the borrowers. 11 Practitioners acknowledge this as well. See, for example, Steven Marks, managing director & head of U.S. REITs, Fitch Ratings, AICPA CPA Letter Daily, October 26, Mandatory accounting changes can also impose additional contracting costs because they increase the costs of the investigation and resolution of unintentional violations (Leftwich 1983; Watts and Zimmerman 1990; Beatty et al. 2002). 10

12 Specifically, for each of our 114 firms that were close to violating the covenants (including both constrained firms and violator firms), 13 we hand-collected the credit agreement contracts and amendments that were in affect at the end of 2007 when SFAS 160 was issued as well as the new credit agreements and amendments post-sfas 160 in the years In particular, we hand-collected 311 contracts for 103 close-to-violation firms and examined whether they employed frozen GAAP, rolling GAAP, or rolling GAAP with option to freeze. We also examined whether the contract terms were modified as a response to or in anticipation of SFAS 160. We examined whether the affected covenant terms (the required benchmark ratios such as net worth and leverage ratio) were altered and whether there were new terms relating to minority interest. Finally, we looked for any evidence of a renegotiation of the contracts. Consistent with Christensen and Nikolaev (2013), we find that the majority of debt covenants (78.6%) have the option to freeze the GAAP and 18.4% have rolling GAAP. Only 3% of our sample had frozen GAAP. Overall, 21.4% of our firms could not exploit the accounting change because their contractual terms either explicitly included/excluded minority interest or specifically employed frozen GAAP. For the remaining firms, SFAS 160 was not accounted for ex ante, and there is no evidence that it engendered renegotiation of contracts ex post. Our findings are consistent with the Frankel et al. (2010) finding that, before SFAS 160, only 46 out of 450 examined credit agreements (10.2%) used a definition of net worth that would not have been affected by the accounting change. In the contracts examined in the post-sfas 160 period, this number was even lower: 30 out of 384 (7.8%). The lack of renegotiations is also consistent with the literature. Campello et al. (2011) find, regarding credit lines, that, for more than one-third of their sample firms, even a violation 13 We did not require the availability of all control variables for the hand-collected sample, as we did for our main sample reported in Table 1. 11

13 did not lead to any renegotiation. In our case, it is even less likely that the accounting change would trigger a renegotiation that would neutralize the positive shock to borrowers since the change benefits borrowers. Thus, while some sophisticated lenders may have noticed the changes, they would have had to renegotiate to the detriment of their borrowers. In a renegotiation, the borrower s position is likely to have improved (weakly) given this favorable accounting change. Thus, whether renegotiations took place or not, the borrowers exposed to treatment, on average, improved their standing. Furthermore, if some renegotiations had occurred or if some firms chose not to respond to the shock out of the fear of hurting their borrowing relationships, the magnitude of the average empirical response is weakened. Thus the findings in this paper can be seen as weaker than the full response by firms to a relaxation of the slack in their debt covenants. 14 The impact of other mandatory accounting changes on firm value and behavior, under the assumption that the debt covenants did not adjust to incorporate the changes, has been documented before. For example, Lys (1984) documents a negative stock price reaction related to SFAS 19 (full cost accounting for oil and gas exploration), and Imhoff and Thomas (1988) find that SFAS 13, which modified lease accounting, impacted the capital structure. Espahbodi, Espahbodi, and Tehranian (1995) document a positive stock price reaction to SFAS We further examine the issuance of relationship loans around SFAS 160. We find that firms with Affected covenants and with minority interest generated on average 65% of their new loans from lenders they already work with (relationship loans) both before and after SFAS 160. In contrast, for the non-affected firms, the frequency of relationship loans significantly decreased from 56% to 52% after SFAS 160, consistent with the overall decline in leverage in the post period for the control firms. It also would be interesting to examine whether the response differed for firms that had longer relationships with their lenders. However, given the structure of our data where a unit of observation is a firm quarter and where we identify constrained firms based on the most restrictive relevant covenants they have categorizing them by the nature of their relationship is not straightforward. In other words, there may be multiple contracts (on average 3) and multiple lenders per borrower, and therefore we would be faced with decisions regarding how to characterize the group of lenders (and lead arrangers), some with relevant covenants, some with binding relevant covenants, and some with other covenants. 12

14 (recognition of deferred tax assets). More recently, Frankel et al. (2010) document abnormal returns surrounding the release of SFAS 160 and further find these returns to be increasing with the level of minority interest. Frankel et al. (2008) and Demerjian (2011) likewise provide evidence that uncertainty driven by accounting estimates hampers debt contracting. However, these studies do not examine the effect on firms that were constrained by affected covenants. In addition, SFAS 160 is unique, compared with the accounting changes examined in prior studies. It does not change the measurement of minority interest, only its location on the balance sheet. Thus the standard does not add uncertainty due to measurement issues. Our findings could however be attenuated by the impact of SFAS 141(R) on measurement. Finally, anecdotal evidence suggests some parties did not notice the change. For example, despite the discussions surrounding the rule, it was only in February 2011 that Capital IQ (a leading provider of financial data and analytics) sent out a letter to its members cautioning that calculations should be reviewed and that the Compustat variables relating to noncontrolling minority interest and equity were updated to account for this change. In addition, we discussed the issue of the option to freeze GAAP in general and the implications of SFAS 160 on lending in particular with several corporate banking lending executives. They confirmed that mandatory changes in accounting are only dealt with once a contract is renewed and are not a reason for renegotiation. In summary, parties wrote their contracts based on the accounting rules as they were at the time. Therefore the implementation of SFAS 160 caused a real change in the value of the existing contracts and firms bargaining power following the accounting change. Furthermore, we do not use variables after 2008, so this shock is not affected by changes in the investment opportunity set or changes in the firms financial condition. In this sense, the shock is exogenous 13

15 in our study in that it is not related to either the firms investment opportunities or changes in their financial position. 3. Data 3.1 Sample Construction We extract loan and covenant information from the Loan Pricing Corporation (LPC) DealScan database. We only use deals that could be matched to Compustat. We examine all loans in effect during our sample period, Following Chava and Roberts (2008), the presence of a loan in a given quarter is based on the facility start date according to LPC DealScan database before the observation quarter with a facility end date (latest maturity date) beyond the observation quarter. Post-origination amendments, which impact the specifications of the covenant threshold or the maturity of the contracts were gathered from LPC and linked via the loan identifier. We examine the impact of SFAS 160 on loan covenants. We focus on the equity-based covenants that were affected by SFAS 160, since they are stipulated in terms of firm equity, and where the directional effect is clear. The accounting change mechanically caused the equity section on the balance sheet to increase by the amount of nonredeemable noncontrolling interest. Therefore both the debt-to-equity ratio and the leverage ratio (which is defined as total debt divided by total debt plus total equity) decline since equity in the denominator increased, 16 while 15 We experimented by including a sample period of , as well as by conducting the analysis on an annual instead of quarterly basis, and found similar results. 16 The leverage ratio is defined by DealScan in its glossary as debt divided by capitalization, where capitalization is equal to total debt plus total equity (or total debt plus net worth). While many firms define the term leverage ratio differently, DealScan classifies and codes only total debt to total capitalization as leverage ratio and, for example, will code the debt-to-ebitda ratio as max debt to EBITDA, regardless of the term the company and creditors use in the credit agreement. We verify the definition of leverage ratio on DealScan in 40 randomly selected credit agreements. For examples, see the five-year credit agreement for Amerada Hess Corp, dated May 12, 2006 ( the AARON Rents Inc. credit agreement, dated May 23, 2008 ( and the Ametek Inc. credit agreement, dated September 17,

16 both the net-worth (which is generally defined as equity) and the tangible-net-worth covenant ratios increased as the equity increased. As a result, debt to tangible net worth decreases. We focus on these five covenants. In contrast, covenants based on EBITDA and cash flows (e.g., debt-to-ebitda) might be affected, but such an effect, if any, is not clear. 17 Furthermore, EBITDA and cash flow calculations are not standardized, as private lenders make adjustments to GAAP and customize these financial statement variables in their credit agreements. 18 We generally treat the covenant data in the same way as Chava and Roberts (2008). 19 When there are overlapping deals, the relevant covenant is the tightest one. When the covenants adjust dynamically over the life of the loan, we linearly interpolate the covenant thresholds. We match these data to nonfinancial firms in the Compustat database using the link file created by Chava and Roberts (2008) and compare the covenant requirement as it is in effect at the time. Our unit of observation is a firm-quarter (as opposed to a deal or facility), as the covenant match is intended to identify a group of firms that receive treatment; when more than one covenant type exists, we track the most restrictive one. As shown by Chava and Roberts (2008), the merged sample resembles the Compustat universe. 20 Table 1, Panel A, describes the ( 17 If EBITDA is calculated as operating income plus depreciation and amortization, it will not be affected by SFAS 160. In contrast, if EBITDA is calculated as net income, plus income tax expense, plus net interest expense, plus depreciation and amortization, it might be affected, as the definition of net income has changed due to SFAS 160. If the calculations employ net income before income attributed to noncontrolling interest, the ratio can either increase or decrease the EBITDA value since either income or loss may be attributable to noncontrolling interests in any given year. Thus, since we cannot estimate the impact of SFAS 160 in such cases, we exclude these ratios from our treatment groups, but these firms are included in our control groups. 18 Cash flow, for instance, is defined in more than a dozen ways (Leftwitch 1983; Dechow and Skinner 2002; Chava and Roberts 2008). 19 See their appendix for full details. 20 There is noise in this comparison since the Dealscan database contains aggregate information on the loans and does not adjust for any special definitions the contract may have. As noted by Li (2010), Dichev and Skinner (2002), and Leftwich (1983), there may be some variance in the manner in which debt and net worth are defined in the covenants. This adds noise to our analysis. We still include these covenants since many of the discussions on the adverse effects of the accounting change were centered around these covenants and there is no systematic noise introduced. We also use the net-worth and tangible-net-worth covenants used by Chava and Roberts (2008) and Dichev and Skinner (2002). 15

17 prevalence of these covenants in our matched sample. As an additional robustness test (untabulated), we excluded all firms with frozen-gaap, specific exclusion of minority interest, or both from the relevant covenants, with little change to our findings, due to the rarity of such instances in our sample. We compare the firms corresponding accounting variables to the requirements in the aforementioned covenants and use two measures. We define firms with affected covenants as firms that have Affected covenants (i.e., covenants that are affected by SFAS 160). A firm is Constrained if it is within 30% of the covenant threshold but not in violation of a debt covenant. For example, if the covenant specifies a net worth requirement of $100 million, a firm with a net worth below $130 million and above $100 million is constrained. Table 1, Panel B, describes the prevalence of these thresholds. Our measure of strictness of covenants resembles the one used by Drucker and Puri (2009), who scale the slack by total assets, and satisfies the desirable properties for measuring strictness of covenants suggested by Murfin (2012). It directly achieves the slack from covenant and scale requirements and partially satisfies the number of covenants requirement by focusing on the strictest covenant. As robustness tests, we try different covenant slacks (e.g., within 10% of covenants threshold) and, as an alternative definition of covenants strictness, the distance between the firm s ratio and covenant threshold deflated by the standard deviation of the ratio over the last 12 quarters. (For similar implementation, see Demiroglu and James (2010), Ertan et al. (2013) and Dyreng (2009)). Specifically, we classify firms as constrained if the current slack is below 50% of the standard deviation in the ratio. All results remain qualitatively similar. We further accessed the data on the usage of revolving lines of credit (revolver) through Capital IQ database. Lastly, we accessed the CRSP database for the daily stock file to calculate 16

18 expected default frequency (EDF) following Bharath and Shumway (2006). 3.2 Data Description Table 1 summarizes our Compustat-DealScan merged data. We require a firm to have at least one observation before and after SFAS 160 and all main regression dependent and control variables and conduct our main analyses on 49,476 firm-quarters. Panel A describes the sample of firm-quarters with available data. Of our sample, 12.2% of the firm-quarter observations have at least one covenant on DealScan. The number of observations with at least one covenant is 2,119, 2,010, and 1,907 in 2007, 2008, and 2009, respectively. Out of the sample of covenants, the most prevalent is the leverage ratio, which constitutes between 4.7% and 5.4% of the sample. In contrast, the debt-to-equity ratio is much less popular and constitutes less than 0.31% of our sample. The second most prevalent is the net worth covenant, constituting between 4.1% and 5.5% of our sample. The tangible net worth covenant constitutes between 2.6% and 4.1% of our sample. Finally, the debt-to-tangible-net-worth covenant constitutes between 1.5% and 1.9% of our sample. Table 1, Panel B, shows that our sample is well populated, meaning that we have a considerable number of firms with both Affected covenants and minority interest. As mentioned above, we sort our sample into the two groups of covenant constraints: Firms with Affected covenants and Constrained firms that are close to violating covenants (excluding violators). Table 2 reports the summary statistics of our sample firms. Panel A reports average changes for firms with Affected covenants around SFAS 160. Panel B reports average changes for Constrained firms around SFAS 160. We also separate our sample based on minority interest, as firms with minority interest may be distinct. Minority interest arises from acquisitions, and firms that acquire may differ from those that do not because acquirers tend to be larger and more 17

19 mature. Table 2 shows that the sample of firms with Affected covenants and minority interest is not significantly different than the corresponding sample without minority interest, in terms of total debt. However, Constrained firms with minority interest have higher total debt than those without it. The main difference associated with minority interest is size. Specifically, firms with minority interest are larger on average than firms without it. Although there are other notable differences (such as financial distress), the firms are generally similar in terms of our other variables of interest, such as profitability and net worth. 4. Empirical Method The majority of our specifications are triple differences, that is, a difference-indifference-in-difference approach. These regressions will generally take the following form: y it * POST i * MIB 4 t 1 * X * POST * X * POST 2 * Z it, it * MIB 3 * POST Where αi is firm fixed effects, δt is year-quarter fixed effects, X is the relevant constraint measure (firms with Affected covenants or Constrained firms), POST is the period after 2008, and MIB is the measure of minority interest (either a dummy equal to one for having minority interest at all or a continuous measure for the amount of minority interest in the last quarter of 2008). Zit represents the matrix of control variables in the analyses, following Roberts and Sufi (2009a) and Chava and Roberts (2008). 21 Given the firm fixed effects, the coefficients measure the change in the outcome variable engendered by an increase in the corresponding explanatory 21 The control variables include lagged natural logarithm of total assets, the lagged-tangible-assets-to-total-assets ratio, the lagged market-to-book ratio, and a lagged has SP rating indicator as control variables. In addition, covenant controls include 10 variables: the lagged-net-worth-to-assets ratio, the lagged-cash-to-assets ratio, the lagged and current EBITDA-to-lagged-assets ratio, the lagged and current cash-flow-to-lagged-assets ratio, the lagged and current net-income-to-lagged-assets ratio, and the lagged and current interest-expense-to-lagged-assets ratio. 18

20 variables. Throughout the analyses, we cluster the standard errors by firm to flexibly control for serial correlation. All variables are defined in the appendix. We sort firms into two treatment groups based on the intensity of treatment: (1) the broadest affected group consisting of all firms with minority interest and covenants stated in terms of equity (firms with Affected covenants), and (2) firms in (1) that were close to violating their relevant covenants, excluding those in violation (Constrained firms). Our tested conjecture is that all firms balance the benefits of debt with the potential costs of covenant violations, leading them to choose a buffer of equity (covenant slack). 22 Therefore we expect the firms in Group (1) to have weakly increased their debt (relative to a properly constructed control group/counterfactual). The effect for this group is weaker since the covenant buffer constraint may not have been binding, as these firms were not close to tripping any covenants. Firms in Group (2) were more likely to be constrained by the financial covenant as they were, by construction, close to tripping a covenant, and thus we expect this group to have responded more strongly to the accounting-induced slack by increasing leverage. The focus of our analysis is on β4, the triple interaction term. The model controls for general time trends using the period fixed effects, δt. It controls for the different behavior of our treatment firms (firms Affected or Constrained by our specific covenants) post-2008, β2. The model also controls for different behavior of firms with minority interest after 2008, β3. Thus our triple interaction term captures the specific effect of SFAS 160 (post-2008) on firms that are Affected or Constrained by the specific covenants, which were relaxed by having minority 22 Firms may avoid violations for a variety of reasons. For example, renegotiation may be costly, and it may be difficult to prove to creditors that the firm is still solvent despite the violation. More broadly, creditors may have different tastes (e.g., for risk), and they may impose such preferences on the firm when given control rights. 19

21 interest. In other words, any alternative consideration that can drive our results must affect only firms that have minority interest after 2008 and are affected (or constrained) by debt covenants, which are affected by minority interest. Our dependent variables include current debt, long-term debt, total debt, long-term investments, cash holdings, earnings, and financial distress (EDF). Note that we employ the minority interest at the end of 2008 when measuring MIB. Therefore our measure of minority interest is not affected by acquisitions and divestitures following the adoption of SFAS 160. It is also unaffected by SFAS 141R, which changed the measurement of minority interest for acquisitions post As we detail below, robustness exercises include a one-to-one propensity-score matching conducted at firm level at the end of 2008 to mechanically construct a tailored control group of firms without minority interest but that are most likely to carry minority interest. The matched sample was based on financial controls as well as covenant controls that are used in our tests. Since we employ a firm-fixed effect model, we use leverage as the dependent variable. The firmfixed effects are akin to using a changes specification. Furthermore, firms can change leverage by either borrowing or lowering their equity (e.g., paying dividends). However, for completeness and robustness, we also employed a changes-on-changes specification without firm-fixed effects. Our results hold when using the change in debt levels, following Lemmon, Roberts and Zender (2008), as the dependent variable. Our identification argument is therefore based on a large series of specifications and falsification tests and several empirical methodologies all corroborating the same empirical findings. 20

22 5. Empirical Findings 5.1 The Impact on Debt We begin the empirical analyses by estimating our main model on the full sample of 49,476 firm-quarter observations, reported on Table 3. Panel A reports results when we employ the level of minority interest at the end of 2008, while Panel B employs an indicator variable for firms with minority interest at the end of We report triple difference interactions for both firms with Affected covenants and Constrained firms. Constrained firms are likely to benefit from the treatment but are not already saddled by a violation of a covenant. In columns 1 3 of both panels in Table 3, we test whether constrained firms with minority interest increased their short-term debt, long-term debt, and total debt after the adoption of SFAS 160 in The results indicate that Constrained firms increased their leverage when they received an exogenous increase in their covenant slack. Specifically, the findings indicate that Constrained firms increased their total debt by increasing their long-term debt. The size of the coefficient on the triple interaction term in column 3 of Panel A (approximately 0.5) implies that the increase in debt is approximately half the size as minority interest. The coefficient on the triple interaction term in column 3 of Panel B implies that Constrained firms with minority interest increased their debt by approximately 2.6% following the passage of SFAS 160. In columns 4 6 of both panels in Table 3, we test whether firms with Affected covenants and minority interest increased their leverage after the adoption of SFAS 160 in Our findings are consistent with our expectation that the response would be stronger when the covenant constraint was more likely to be binding. Specifically, for the group of firms with Affected covenants with minority interest, the coefficient on the triple interaction term is 0.26, which is lower than size of the triple interaction with Constrained firms. 21

23 This increase in debt is after taking out fixed effects and controls and relative to the control groups. The positive effect thus also results from firms deciding to lower their leverage by less than our counterfactual and is not purely driven by new lending. It is therefore consistent with our interpretation that firms can now lower the amount of equity they keep on their balance sheets and still avoid tripping covenants. The findings also highlight that our results are driven largely by covenant constraints rather than a period effect. Generally, our findings show that leverage is declining during our sample period, post crisis (the coefficient on POST is consistently negative), where firms were largely deleveraging at a varied pace. On average, leverage declined for all firms during the period. For constrained firms (without minority interest) and unconstrained firms with minority interest, which had higher leverage (see for example descriptive statistics in Table 2), leverage was declining more. The higher reduction in leverage for constrained firms without minority interest is consistent with their being affected more by the financial crisis, as documented by Campello, Graham and Harvey (2010). Our findings show that the lower deleveraging or the increase in leverage is apparent only in cases where debt-covenant slack increased as the result of SFAS 160, adding support for the conjecture that debt covenants per se affect leverage decisions before violations Additional Tests Propensity Score Matched Sample One alternative explanation for our results in Table 3 is the systematic differences between firms with and without minority interest. Although the triple difference design with firm and time fixed effects reduces this concern by controlling for firm and time-invariant characteristics, we further address it by estimating our main analysis on a propensity-score matched (PSM) sample. 22