Dynamic threshold values in earnings-based covenants

Journal of Accounting and Economics ] (]]]]) ]]]–]]]

Contents lists available at ScienceDirect

Journal of Accounting and Economics journal homepage: www.elsevier.com/locate/jae

Dynamic threshold values in earnings-based covenants$ Ningzhong Li a, Florin P. Vasvari b, Regina Wittenberg-Moerman c a

University of Texas at Dallas, 800 West Campbell Road, Richardson, TX, USA London Business School, Regent’s Park, London NW1 4SA, United Kingdom c Leventhal School of Accounting, University of Southern California, Los Angeles, CA,USA b

a r t i c l e i n f o

abstract

Article history: Received 22 October 2012 Received in revised form 23 June 2015 Accepted 17 July 2015

We examine the role of dynamic covenant threshold values in syndicated loan agreements. We document that 45% of syndicated loans specify dynamic covenant thresholds in earnings-based covenants and that these changing thresholds typically become tighter over the life of a loan. We find that covenants with a tight trend provide an important signaling mechanism that meets the needs of borrowers that experience an inferior financial performance at loan initiation but expect future performance improvements. Specifically, we find that these covenants provide underperforming borrowers with a grace period by requiring less restrictive initial thresholds. At the same time, they allow these borrowers to credibly convey information to lenders about their future prospects via gradually more demanding subsequent thresholds. Our empirical evidence also suggests that while lenders entering into tight threshold trend covenant contracts receive weaker covenant protection over the grace period, they benefit from having stronger control rights in subsequent periods. & 2015 Published by Elsevier B.V.

JEL classifications: G17 G21 G32 M41 Keywords: Syndicated loans Financial covenants Covenant threshold trend Signaling hypothesis Incomplete debt contracting theory

1. Introduction Theoretical research suggests that, in the presence of asymmetric information, borrowers can use debt contractual terms to credibly convey to lenders favorable information about their future prospects (e.g., Chan and Kanatas, 1985; Besanko and Thakor, 1987; Garleanu and Zwiebel, 2009). Manso et al. (2010) and Demiroglu and James (2010) show that strong borrowers signal their “good type” to lenders by committing to performance pricing provisions (PPP thereafter) and the tight slack in financial covenants at loan initiation, respectively. However, borrowers with inferior financial performance but ☆ We appreciate the helpful comments of an anonymous referee, the editor (Wayne Guay), Anne Beatty (discussant), Phil Berger, Lamont Black (discussant), Doug Diamond, Scott Liao (discussant), Haresh Sapra, Doug Skinner, Amir Sufi, Scott Richardson, Michael Roberts, Vikrant Vig, Andrew Winton and Franco Wong (discussant), participants at the 2013 AAA annual meetings, the 2013 MIT Asia Conference in Accounting, the 49th Bank Structure Conference at the Chicago Fed, the 2015 Midwest Finance Association Meetings, the University of Minnesota Empirical Conference and seminar participants at London Business School, Tilburg University, the University of Chicago, the University of Michigan and the University of Southern California. We thank Greg Nini, David Smith and Amir Sufi for covenant violations data and Yiwei Dou for covenant renegotiations data. We also thank Ying Huang, Yun Lou, Yu Xie and Sundipika Wahal for excellent research assistance. We gratefully acknowledge the financial support of the AXA Research Fund, the London Business School RAMD Fund, the University of Chicago Booth School of Business, and the University of Texas at Dallas. Regina Wittenberg-Moerman also gratefully acknowledges the financial support of the Neubauer Family Fellowship. This paper was previously circulated under the title “The Information Content of Threshold Values in Earnings-Based Covenants.” E-mail addresses: [email protected] (N. Li), [email protected] (F.P. Vasvari), [email protected] (R. Wittenberg-Moerman).

http://dx.doi.org/10.1016/j.jacceco.2015.07.004 0165-4101/& 2015 Published by Elsevier B.V.

Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i

2

N. Li et al. / Journal of Accounting and Economics ] (]]]]) ]]]–]]]

promising future prospects often access debt markets. Our study sheds light on how these borrowers can convey their type to lenders by using a previously unexplored feature of financial covenants – the presence of threshold values that become stricter over the life of the loan. We focus on earnings-based covenants, which represent the most common financial covenants in syndicated loans: interest coverage (IC), fixed charge coverage (FCC), debt service coverage (DSC), debt to cash flows (DCF) and minimum EBITDA covenants. We find that around 45% of loan contracts specify, for at least one of these covenants, a grid that designates how the covenant thresholds change over the life of the loan, including the exact date when the new threshold applies and its value. The vast majority of these threshold grids have a tight trend, which sets stricter threshold values over a contract’s duration relative to the threshold at contract initiation. For example, an interest coverage ratio may be set at 1.5 during the first four quarters after a loan’s initiation, 1.75 during the following two quarters and 2.5 thereafter. We expect tight threshold trend covenants to meet the special signaling needs of borrowers underperforming at loan initiation but expecting their future performance to improve. The tight threshold trend feature may endow these borrowers with a period of a temporary reduction in the restrictiveness of covenant thresholds following the loan’s issuance (i.e., a “grace period”), thus granting them some time to enhance their financial performance. At the same time, this feature offers underperforming borrowers the opportunity to credibly convey information to lenders about their future prospects by requiring gradually more demanding subsequent thresholds. We argue that covenants with a tight trend provide an important signaling mechanism for borrowers who experience poor performance at loan initiation. These borrowers cannot rely on constant threshold covenants with a tight slack to convey their future prospects to lenders as their poor initial performance is likely to immediately trigger violations of such covenants. Underperforming borrowers also cannot signal via interest increasing PPP, as these provisions allow borrowers to only commit that their performance will not deteriorate in the future, without offering an opportunity to convey expectations about future performance improvements.1 The use of covenants with a tight trend in loan contracts is also consistent with the predictions of incomplete contracting theory, which suggests that covenants designate a state-contingent allocation of control rights between the borrower and lenders based on pre-specified contractible signals that reflect the borrower’s underlying performance (e.g., Aghion and Bolton, 1992; Aghion et al., 1994; Dewatripont and Tirole, 1994). While the required level of the contractible signal (i.e., the performance threshold) in constant threshold covenants remains the same over the loan’s life, tight trend covenants allow a dynamic ex ante allocation of control rights because their thresholds change over the loan’s life. More specifically, by demanding a less restrictive initial threshold following the loan’s initiation, tight trend covenants can endow borrowers with a grace period during which they retain control rights despite their relatively poor performance; these covenants then shift control rights to lenders if the borrower continues to perform poorly and cannot meet the more demanding thresholds over subsequent periods. Another important advantage of tight trend covenants is that they potentially mitigate credit rationing for borrowers underperforming at loan initiation and reduce their cost of debt financing. In the absence of the tight trend feature, we expect these borrowers to experience difficulties in accessing credit because lenders are unlikely to issue credit to poorly performing firms unless they can commit to future performance improvements. In other words, we expect that tight trend covenants facilitate access to credit and decrease loan pricing for temporarily underperforming borrowers by providing a signaling mechanism that allows these borrowers to separate themselves from weaker borrowers that do not expect performance improvements. To support our predictions, we conduct a series of tests. First, we examine the determinants of the presence of tight trend covenants in loan contracts. We find that borrowers are more likely to commit to these covenants if they report losses, have a lower interest coverage ratio and operating cash flows and violate financial covenants prior to the loan’s issuance. These findings support our expectation that borrowers utilize a tight trend structure when they are experiencing poor performance at loan initiation and need a grace period to enhance it. Second, we examine the strictness of the initial threshold values of tight trend covenants relative to constant thresholds. Controlling for fundamental firm and loan characteristics, we find that the initial thresholds of IC, FCC and DCF covenants with tight trends are less restrictive.2 This result suggests that a tight trend provides underperforming borrowers with a grace period following the loan initiation that allows them to improve performance. We estimate that this period lasts between one year (or 20% of a loan’s maturity), if we consider the first increase in the threshold, and two and a half years (or 50% of a loan’s maturity) if we consider the period until the threshold reaches its final value. In addition, we find that final thresholds in tight trend covenants are significantly more demanding than constant thresholds, potentially compensating lenders for the weaker covenant protection over the grace period. Third, we analyze tight trend borrowers’ realized future performance. We find that, relative to borrowers who do not commit to tight trend covenants, borrowers with tight trend covenants experience a deterioration in profitability, interest

1 We note that the interest decreasing provisions, which decrease the interest rate when a borrower’s performance improves, cannot serve as a signaling mechanism because they do not impose any ex ante costly commitment on the borrower (e.g., Asquith et al., 2005). As it is no more costly for weak borrowers to commit to interest decreasing provisions than it is for stronger borrowers, the former will mimic the strong borrowers and also commit to the interest decreasing performance pricing provisions, which will prevent a separating equilibrium. 2 We exclude from these tests DSC covenants due to their low frequency and EBITDA covenants because their thresholds change with firms’ size and may be negative, which complicates cross sectional comparisons.

Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i

N. Li et al. / Journal of Accounting and Economics ] (]]]]) ]]]–]]]

3

coverage and debt to EBITDA ratios in the first year following the loan’s issuance. The decrease in their performance, coupled with underperformance at loan initiation, further supports our prediction that these borrowers cannot commit to “standard” constant thresholds at loan initiation and need a grace period. However, we also find that this performance deterioration is temporary, as tight trend borrowers show improvements across all three performance measures over the two subsequent years. This performance improvement is consistent with the tight trend feature serving as a signaling mechanism: initially underperforming borrowers commit to tight trend covenants to signal their expectations of improved future performance to lenders. Fourth, we investigate whether a tight trend mitigates credit rationing for underperforming borrowers. We acknowledge that we cannot examine borrowers that were denied credit because they could not commit to tight trend covenants in the loan contract. We therefore provide a number of supportive descriptive analyses. We find that a significant proportion of tight trend borrowers would have likely violated covenants at loan initiation had they committed to constant threshold covenants, suggesting that the tight trend feature potentially allows underperforming borrowers to obtain access to loans. We also find that the majority of borrowers with the weakest performance at loan initiation based on the IC, FCC and DCF ratios commit to tight trend IC, FCC and DCF covenants, respectively. This result further indicates that lenders are more willing to issue credit to poorly performing borrowers if these borrowers provide a credible commitment to improve their performance in subsequent periods. We next show that the tight trend feature is included in the majority of other loans issued by tight trend borrowers within the 12 month period centered on the current loan’s issuance date, suggesting that a tight trend is likely to be the prevailing mechanism that allows these underperforming borrowers to obtain credit. Further, we document that the initial level of covenant protection is priced more favorably for tight trend covenants relative to constant threshold ones, implying that tight trend borrowers would pay a higher interest rate in the absence of the tight trend feature. Last, we address lenders’ incentives to enter into tight trend covenant contracts that potentially provide them with weaker protection over the grace period. Prior research suggests that more restrictive financial covenants endow lenders with stronger bargaining power in ex post renegotiations, allowing them to retain a higher proportion of the efficiency gains that can be realized in renegotiations (e.g., Huberman and Kahn, 1988; Aghion et al., 1994). Our findings of significantly higher final thresholds in tight trend covenants relative to the constant thresholds suggest that tight trend covenants endow lenders with stronger bargaining power in ex-post renegotiations following the grace period. We supplement this evidence by showing that tight trend covenants are more likely to be renegotiated relative to constant threshold covenants. We infer that, relative to contracts with constant threshold covenants, lenders entering into tight trend covenant contracts that provide weaker covenant protection over the grace period are compensated by potential rents in ex-post renegotiations due to their stronger bargaining power and a higher renegotiation probability over the following periods. Our study contributes to the literature along several dimensions. First, we expand the literature on the critical role of covenants in debt contracting (e.g., Christensen et al., 2015). Prior work examines mainly the determinants of specific covenants and the number of covenants in loan contracts (e.g., Bradley and Roberts, 2004; Demerjian, 2011; Christensen and Nikolaev, 2012) or explores the factors that explain the financial covenants’ tightness at loan issuance (e.g., Dichev and Skinner, 2002; Chava and Roberts, 2008; Drucker and Puri, 2009; Demiroglu and James, 2010; Murfin, 2012). More closely related to our study, by focusing on net worth covenants, Beatty et al. (2008) examine the specific terms of covenant structure and its economic determinants. Our contribution lies in the analysis of the role of a previously unexplored but widely used financial covenants feature – a tight trend in threshold values.3 Second, our findings also contribute to the limited empirical research on the informativeness of debt contractual terms (Demiroglu and James, 2010; Manso et al., 2010). We show that a tight trend covenant provides an important signaling mechanism for borrowers experiencing a temporarily inferior performance at loan initiation. Our findings also suggest that by providing an opportunity for underperforming borrowers to signal their type and to separate themselves from weaker borrowers that are not expecting performance improvements, tight trend covenants potentially mitigate credit rationing and reduce the cost of debt capital. Third, we highlight the important role played by earnings-based covenants in the ex-ante allocation of control rights in loan contracts. We demonstrate that a tight threshold trend facilitates a dynamic allocation of control rights between borrowers and lenders, thus corroborating the incomplete contracting theory. When borrowers underperform at loan initiation but are expected to improve their future performance, it is more efficient for lenders to initially relax the lending criteria over a grace period, providing borrowers with some time to enhance their performance, but then to obtain control rights if the borrowers cannot meet the increasingly demanding thresholds. Covenants with constant thresholds, even if set very tightly at loan origination, are not able to support this ex ante dynamic allocation of control rights. Finally, our paper is relevant to the research on financial covenant restrictiveness (e.g., Drucker and Puri, 2009; Murfin, 2012). With the notable exception of Dichev and Skinner (2002) and Beatty et al. (2008), other studies evaluate financial covenant slack based solely on its initial threshold. Given that earnings-based covenants are the most frequently used covenants in loan contracts and that a tight trend feature is often used in these covenants, our study highlights that ignoring

3 Fang (2011) also looks at covenants with dynamic thresholds, but, in contrast to our study that utilizes data hand-collected from loan contracts, relies on incomplete covenant data from DealScan. This study also investigates a different question: it analyzes the substitution between a tight trend feature with investment restrictions and loan maturity.

Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i

N. Li et al. / Journal of Accounting and Economics ] (]]]]) ]]]–]]]

4

threshold changes is likely to introduce significant measurement error into the covenant slack estimation.4 Because of the strong link between covenant slack and the occurrence of covenant violations, the accurate estimation of the strictness of financial covenants is particularly relevant to the rapidly expanding research on covenant violations (Chava and Roberts, 2008; Roberts and Sufi, 2009a; Nini et al., 2009, 2012; Freudenberg et al., 2012). The remainder of the paper is organized as follows. Section 2 presents our hypotheses. Section 3 describes the sample and data. Section 4 reports our results and Section 5 concludes. 2. Hypotheses development The theoretical literature suggests that in the presence of asymmetric information, borrowers can credibly inform lenders about their prospects by committing to debt terms that signal these prospects (e.g., Chan and Kanatas, 1985; Besanko and Thakor, 1987; Garleanu and Zwiebel, 2009). Garleanu and Zwiebel (2009) model the design of debt covenants and show that managers with fewer wealth transfer activities at their disposal and who are better informed than lenders signal their “good type” by committing to constant threshold covenants with a tight slack. Demiroglu and James (2010) find that borrowers who commit to tight slack covenants experience significant improvements in their performance over the subsequent three years. Manso et al. (2010) show that interest increasing PPP also allow strong borrowers to signal their type and to separate themselves from weaker borrowers. They find that borrowers whose loans have PPP are more (less) likely to have their credit rating upgraded (downgraded) over the first year following the loan’s issuance, relative to borrowers with loans without these provisions.5 While the prior literature shows that strong borrowers utilize covenants with a tight slack at loan initiation and interest increasing PPP to signal their type to lenders, it remains unclear how borrowers with an inferior financial performance at loan initiation but promising future prospects can credibly convey their type to lenders. We hypothesize that borrowers underperforming at loan initiation but who expect their performance to improve in the future signal to lenders their better future prospects by committing to covenants with a tight threshold trend. This covenant structure may provide borrowers underperforming at loan initiation and/or shortly after it with a grace period – a period of temporarily less demanding covenant thresholds following the loan issuance. These less demanding thresholds allow borrowers some time to enhance their financial performance. At the same time, the tight trend provides underperforming borrowers with an opportunity to credibly convey information to lenders about their future prospects by requiring gradually more demanding thresholds after the grace period ends. The tight trend covenants offer several important benefits to borrowers. First, they provide a signaling mechanism for borrowers underperforming at loan initiation. These borrowers cannot rely on constant threshold covenants with a tight slack to signal their prospects to lenders because they are likely to immediately violate these thresholds due to their inferior performance. Underperforming borrowers actually need a period of temporarily less demanding covenant restrictions during which they can enhance their performance. Similarly, these borrowers cannot rely on interest increasing PPP to signal their better expected performance. Interest increasing PPP impose a higher interest spread if the borrower’s performance decreases following the loan issuance, thus allowing borrowers to only commit to their performance not deteriorating in the future. In contrast, the tight trend allows borrowers to commit to an improvement in their performance in subsequent periods because it imposes thresholds that become more demanding over time. Moreover, because the contractual description of the tight trend structure designates the specific dates by which the performance improvement is expected as well as the extent of the improvement that has to be achieved over each time period, borrowers can utilize a tight trend to signal when and by how much they expect their performance to improve. Second, tight trend covenants can facilitate an ex-ante dynamic allocation of control rights. Within the incomplete contracting theory framework, because contracting parties cannot anticipate or describe every possible future state of the world, state-contingent control rights allocation via financial covenants is a key contractual mechanism that addresses future opportunism by contracting parties (e.g., Hart and Moore, 1988; Aghion and Bolton, 1992; Berlin and Mester, 1992; Aghion et al., 1994; Dewatripont and Tirole, 1994). The theory shows that state-contingent control rights allocation is based on a contractible signal of a borrower’s performance: borrowers retain control rights if their performance exceeds a prespecified level of the signal, while lenders are endowed with control rights otherwise. Compared to constant threshold covenants in which the level of the contractible signal (i.e., the performance threshold) remains the same over the loan’s entire duration, tight trend covenants allow a dynamic ex ante allocation of control rights because their level of a contractible signal changes over the loan’s life. Specifically, a tight trend could relax covenant thresholds following the loan’s initiation, endowing borrowers with a grace period over which they retain control rights despite their relatively inferior performance. The trend then shifts control rights to lenders if the borrower cannot meet the gradually more demanding thresholds over the following periods. This changing nature of the contractible signal makes tight threshold trend covenants a particularly suitable mechanism for state-contingent control rights allocation for borrowers temporarily underperforming at loan initiation. 4 Another source of measurement error in the estimation of covenant slack is lenders’ substantial and frequent adjustments to GAAP numbers when defining covenant thresholds (Leftwich, 1983; Dichev and Skinner, 2002; Beatty et al., 2008; Li, 2015). 5 While the authors utilize the interest rate provisions more generally, they mainly refer to the interest increasing performance pricing provisions that charge a higher interest rate as the borrower’s performance deteriorates (which the authors denote as risk-compensating provisions).

Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i

N. Li et al. / Journal of Accounting and Economics ] (]]]]) ]]]–]]]

5

Lastly, tight threshold trend covenants potentially mitigate credit rationing for underperforming borrowers. Absent a tight trend feature, we expect these borrowers to experience difficulties in accessing credit because lenders are unlikely to provide financing unless borrowers performing poorly at loan initiation can credibly commit to improve their performance in future periods. In other words, tight trend covenants may facilitate access to credit for underperforming borrowers because it provides them with a signaling mechanism to separate themselves from weaker borrowers that do not expect performance improvements. Further, as these borrowers credibly convey their type to lenders and commit to enhance their performance, we expect tight trend covenants to also allow them to obtain lower interest rate spreads relative to the spreads they would have obtained in the absence of the tight trend feature. Although tight trend covenants are expected to bring benefits to borrowers, it is important to understand why lenders agree to covenants that provide them with weaker protection over the grace period. The new information arrival can create an opportunity for an efficiency gain (i.e., a Pareto improvement) that can be realized by modifying the original contract. This efficiency gain is split between the contracting parties according to their bargaining power during the renegotiation process (Hart, 2001; Roberts and Sufi, 2009b; Roberts, 2015). Financial covenants significantly influence the bargaining position of contracting parties in ex post renegotiations, with more restrictive covenants endowing lenders with stronger bargaining power (e.g., Huberman and Kahn, 1988; Aghion, Dewatripont, and Rey 1994).6 If the final thresholds in tight trend covenants exceed constant thresholds, they will potentially endow lenders with stronger bargaining power in ex-post renegotiations following the grace period relative to their respective bargaining power when covenants have constant thresholds. This will likely allow lenders to retain a higher proportion of any efficiency gains created in loan renegotiations. 7 In other words, we suggest that, relative to constant threshold covenants, lenders that agree to tight trend covenants make an ex ante trade-off between a weaker covenant protection over the grace period and the benefits from stronger bargaining power in subsequent periods, when a borrower’s performance is expected to improve. Thus, lenders’ incentives to accept tight trend covenants are consistent with the inter-temporal nature of loan financing: lenders are willing to subsidize borrowers when they are in trouble, while hoping to earn rents when they perform better (e.g., Rajan, 1992; Petersen and Rajan, 1994, 1995; Bolton et al., 2013).8 In addition, covenants with a tight trend provide lenders with the opportunity to issue long-term loans to underperforming borrowers, because lenders can obtain control rights and effectively shorten the loan’s maturity if the borrower’s performance does not sufficiently improve over time. Absent dynamic thresholds, lenders would probably only offer short-term financing to these borrowers, as short maturities allow lenders to re-evaluate borrowers and to condition the continuation of financing on the improvement in their performance (e.g., Diamond, 1991, 1993). Long-term loans are likely to reduce lenders’ re-contracting costs and effort, potentially generating sizable cost savings (e.g., Asquith et al., 2005). These loans also allow lenders to establish stronger relationships with their borrowers. By virtue of this established relationship, incumbent lenders obtain an informational advantage and are typically able to use it to prevent other lenders from taking over the borrowers (e.g., Dell’Ariccia et al., 1999).9 Although our predictions focus on borrowers experiencing an inferior financial performance at loan initiation, we acknowledge that not only underperforming but also healthy borrowers may commit to covenants with increasing thresholds to convey their future prospects to lenders. However, because healthy borrowers do not need a grace period, to signal their type to lenders and separate themselves from weaker borrowers, they could commit to strict but constant covenant thresholds (e.g., Demiroglu and James, 2010). Covenants with constant thresholds are likely to be less costly for healthy borrowers relative to covenants with accelerated threshold values, due to the potentially higher probability of violations and renegotiations of the latter. Similarly, because healthy borrowers have a strong performance at loan initiation, they can also convey to lenders that their performance is not expected to deteriorate through interest increasing PPP (e.g., Manso et al., 2010). We therefore expect the tight trend feature in financial covenants to be utilized primarily by borrowers underperforming at loan initiation. 3. Sample, covenant trend characteristics and descriptive statistics 3.1. Data sources and sample selection We obtain syndicated loan contracts by searching all the 10-K, 10-Q and 8-K filings with the Securities and Exchange Commission (SEC) on the EDGAR system over the 1996–2009 period (electronic filings are not consistently available in 6 Because the initial contract serves as the “disagreement point” in future renegotiations, Huberman and Kahn (1988) show that even if the parties tear up the initial contract in the future and then write a new one, the original contract will still affect the contractual parties’ bargaining power. 7 If more restrictive threshold values in tight trend covenants increase the likelihood of contract renegotiations relative to renegotiations of contracts with constant threshold covenants, lenders will also benefit from charging loan amendment fees. These fees are not trivial; they reach up to 100 basis points, depending on the size of the loan deal and the complexity of the contract amendment (e.g., Roberts and Sufi, 2009b; Thompson and Ruby, 2014). 8 Petersen and Rajan (1995) suggest that relationship banks provide credit to young and more credit rationed borrowers, but they are compensated by higher interest spreads when these borrowers enhance their performance in future periods. Similarly, Bolton et al. (2013) show that while relationship banks generally charged their borrowers a higher interest spreads, they offered lending to these borrowers at more favorable terms during the financial crisis. 9 Hold-up problems due to the incumbent lender’s information advantage are also documented by the relationship banking literature (e.g., Rajan, 1992; Hauswald and Marquez, 2003, 2006).

Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i

6

N. Li et al. / Journal of Accounting and Economics ] (]]]]) ]]]–]]]

EDGAR prior to 1996). We use a text-search program to scan these filings for loan contracts using the following keywords: “credit agreement”, “loan agreement”, “credit facility” and/or “event(s) of default” (all contracts define events that trigger default in a dedicated "Events of Default" section). After identifying filings that contain syndicated loan contracts, we manually match the contracts to the syndicated loans available in DealScan. There are 15,519 loan packages outstanding to public non-financial U.S. firms with Compustat data available. The manual match of the DealScan data with the SEC filings results in a sample of 9,999 packages from 4,033 firms. To ensure matching accuracy, we check that the borrower’s name, the size and date of the package and the name(s) of the lead arranger(s) stated in the loan contract in the SEC filing are exactly the same as in DealScan. We focus on five earnings-based covenants: interest coverage, fixed charge coverage, debt service coverage, debt to cash flows and minimum EBITDA. The interest coverage covenant (IC covenant thereafter) is typically defined as the ratio of an earnings number (e.g., EBITDA, EBIT, operating income, etc.) to interest expense. The fixed charge coverage covenant (FCC covenant thereafter) is computed as the ratio of an earnings number to fixed charges, which may include interest expenses, principal payments, lease payments, etc. The debt service coverage covenant (DSC covenant thereafter) is measured as the ratio of an earnings number to debt service (interest and principal payments). The debt to cash flows covenant (DCF covenant thereafter) is generally the ratio of a debt measure (e.g., funded debt, senior debt, etc.) to an earnings measure (e.g., EBITDA or EBIT). Finally, the minimum EBITDA covenant (Min. EBITDA covenant thereafter) requires a firm to maintain a minimum profitability level (EBITDA, EBIT, operating income, etc.). For each of these covenants, we code all threshold values and the corresponding timing of the changes in each threshold over the duration of the loan. Conditional on the availability of at least one earnings-based covenant in the loan package contract, our final sample contains 6,826 packages from 3,182 firms.10 Our regression analyses have smaller samples due to additional restrictions on the data available to calculate the variables employed in the multivariate tests.

3.2. Characteristics of a tight threshold trend To provide an example of a tight trend covenant, in Appendix A we present extracts from a loan agreement for Citadel Broadcasting. In this contract, the IC covenant becomes more binding over time. The covenant threshold is 1.5 at the end of the first four quarters of the loan contract and increases to 1.75 for the next three quarters. It subsequently increases to 2.00, then 2.25 and finally to 2.50 beyond the 11th quarter. The threshold values for the FCC covenant remain constant over the contract’s duration at 1.25, while the DCF covenant (or the leverage ratio) also has a threshold trend that becomes more binding over time. Panel A of Table 1 reports the frequency of the covenants in our sample, as well as the distribution of the different threshold trend types. The IC and DCF covenants are the most commonly used: an IC covenant is present in 45% of the sample contracts, while a DCF covenant is present in 65% of the sample. The DSC covenant is the least used (only 6% of the packages have this feature). In terms of threshold trends, covenants become tighter in 30%, 27%, 16%, 41% and 42% of the packages for the IC, FCC, DSC, DCF and Min. EBITDA covenants, respectively. The frequency of covenants that become looser is generally below 3%. Similarly, covenants with fluctuating trends are uncommon. The Min. EBITDA covenant is an exception; we find that about 18% of these covenants have a fluctuating trend.11 The remaining contracts (55%) require constant thresholds over the life of the loan contract. For earnings-based covenants with a tight trend, we report descriptive statistics on the number of thresholds specified in the threshold grid in Panel B of Table 1. The average number of thresholds ranges from 3 for the DSC covenant to 7 for the Min. EBITDA covenant. In Panel C, we report the steepness of the tight threshold trends. We measure the steepness of a trend as the average of the slopes of each threshold change per quarter, weighted by the time periods corresponding to each threshold (Appendix A explains in detail the estimation of this measure). The strictness of IC, FCC and DSC covenants increases on average by 4.6%, 4.6% and 6.5% per quarter, respectively (Panel C of Table 2). Because increasing thresholds for the DCF covenant indicate a looser covenant over time, we multiply its slope by
1; the strictness of this covenant increases on average by 3.8% per quarter. The mean slope for the Min EBITDA covenant is much larger at 27.8% per quarter; this is often due to this covenant’s very low thresholds at loan origination or to thresholds that change from negative to positive values. In Panel D of Table 1, we compare the initial and final thresholds across the constant and tight trend partitions for the IC, FCC and DCF covenants. We exclude from these analyses DSC covenants, due to their low frequency, and EBITDA covenants, because EBITDA thresholds vary significantly with the size of the firm and may be negative at loan initiation, making cross 10 A tight trend is sometimes also included in non-earnings-based covenants, but it is not common. Based on DealScan data, the frequency of this trend over our sample period ranges from 7% for the Min Current Ratio covenant to 14% in the Debt to Tangible Net Worth covenant (except for a very small percentage of fluctuating and loose trends, the remaining non-earnings-based covenants have constant thresholds). Due to the low frequency of the tight trend feature in non-earnings-based covenants, we do not incorporate them into our analyses. Also, limiting our tests to earnings-based covenants helps us keep our covenant hand collection and coding processes manageable. 11 The fluctuating trend in the minimum EBITDA covenant is primarily due to its measurement interval. We randomly check 50 contracts with these covenants and find that 25 (50%) use EBITDA measured on a quarterly basis, 15 (30%) use a cumulative EBITDA over two, three or more quarters, while the rest use a rolling four quarter EBITDA. Due to business seasonality, this measurement of the EBITDA leads to a fluctuating trend. The estimation of the other earnings-based covenants is not affected by seasonality because they use rolling four quarters EBITDA.

Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i

N. Li et al. / Journal of Accounting and Economics ] (]]]]) ]]]–]]]

7

Table 1 Covenant trend descriptive statistics This table reports the frequencies of the threshold trends (Panel A), the number of threshold values (Panel B), the slope of the threshold trends (Panel C), the initial and final covenant threshold values (Panel D), and the grace periods for the tight trends (Panel E) by covenant type. The slope is estimated only for covenants with a tight trend. nnn denotes significance at 1%. All variables are defined in Appendix B. Panel A: Frequency of covenant threshold trends by covenant type N

Constant

Tighter

Looser

Fluctuating

0.677 0.673 0.792 0.543 0.363

0.297 0.270 0.157 0.411 0.423

0.008 0.017 0.027 0.013 0.035

0.060 0.040 0.024 0.033 0.179

IC covenant 3,079 FCC covenant 2,914 DSC covenant 414 DCF covenant 4,458 Min EBITDA covenant 1,186 Panel B: Number of threshold values by covenant type N IC covenant FCC covenant DSC covenant DCF covenant Min EBITDA covenant Panel C: Covenant threshold slope

914 787 65 1,832 502 by covenant type N

IC covenant 914 FCC covenant 787 DSC covenant 65 DCF covenant 1,832 Min EBITDA covenant 502 Panel D: Covenant threshold value by covenant type

Mean

Std

P25

Median

P75

4.300 3.256 2.815 4.477 6.596

2.843 1.604 0.967 2.876 5.610

3.000 2.000 2.000 3.000 3.000

4.000 3.000 3.000 4.000 4.000

5.000 4.000 3.000 5.000 8.000

Mean

Std

P25

Median

P75

0.046 0.046 0.065 0.038 0.278

0.046 0.057 0.111 0.036 0.360

0.021 0.013 0.022 0.018 0.033

0.031 0.028 0.031 0.028 0.097

0.051 0.052 0.059 0.043 0.452

Constant threshold group

IC covenant Initial value Final value FCC covenant Initial value Final value DCF covenant Initial value Final value Panel E: Grace period of tight

Tight trend group

N

Mean

N

Mean

t-Statistic

2,084 2,084

2.87 2.87

917 917

2.15 3.02

20.8nnn
4.01nnn

1,956 1,956

1.56 1.56

786 786

1.23 1.57

13.8nnn
0.70

2,353 2,353 trend by covenant type

3.62 3.62

1,817 1,817

4.56 3.13


8.93nnn 5.18nnn

Loan date to the first threshold value change date In years

IC covenant FCC covenant DCF covenant DSC covenant Min EBITDA covenant

Statistical Difference

Loan date to the last threshold value change date

% of Maturity

In years

% of Maturity

Mean

Median

Mean

Median

Mean

Median

Mean

Median

1.13 1.20 1.04 1.19 0.59

0.97 1.00 0.92 1.04 0.44

26.0 29.1 24.8 26.5 20.8

20.5 24.3 19.6 27.1 14.7

2.56 2.03 2.52 1.73 1.85

2.33 1.84 2.26 1.49 1.31

53.0 47.5 53.7 38.3 53.8

50.0 43.5 51.9 38.1 51.0

sectional comparisons difficult. We find that the tight trend’s initial thresholds are significantly less restrictive than the thresholds in the constant group (the IC and FCC thresholds are lower and the DCF thresholds are higher). But the final threshold significantly surpasses the constant threshold for the IC and DCF covenants, indicating that there is a potential trade-off between lower thresholds at loan initiation and more demanding thresholds by the time the loan matures. Panel E of Table 1 reveals that, except for the EBITDA covenant, the loose initial covenant thresholds last on average for about one year before the covenant threshold starts increasing (or decreasing in the case of the DCF covenant), which represents 20 to 30 percent of a loan’s maturity. For EBITDA covenants, this period is only six months. The period until the Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i

N. Li et al. / Journal of Accounting and Economics ] (]]]]) ]]]–]]]

8

Table 2 Summary statistics This table reports summary statistics for the firm and loan characteristics used in the multivariate analysis. The Slope variable is estimated at the loan package level by averaging the individual covenant slopes across all earnings-based covenants in the package. We require that packages have at least one tight covenant threshold trend. A more detailed definition of Slope is in Appendix A; the other variables definitions are presented in Appendix B. Panel A reports descriptive statistics for the whole sample and Panel B compares firm characteristics across the tight threshold trend and constant threshold groups. nnn denotes significance at 1%. Panel A: Whole sample N

Mean

Covenant characteristics Tight Dummy 6,826 0.391 Slope 2,669 0.083 Loan characteristics Acquisition 6,826 0.169 Term Loan B 6,826 0.288 Loan Amount ($ millions) 6,826 332.57 Loan Amount (log) 6,826 4.905 Maturity (years) 6,725 3.760 Financial Covenants 6,826 2.358 Firm characteristics ROA 6,664 0.019 Loss 6,664 0.251 Interest Coverage 6,068 13.498 Cash flows 6,243 0.103 O-Score 6,389
4.710 Leverage 6,255 1.083 Rating 3,603 12.091 Cov Violation 5,868 0.149 Firm Size ($ millions) 6,743 2121.18 Firm Size (log) 6,743 6.334 Tangibility 6,714 0.328 Earnings Vol 6,448 0.081 Panel B: Firm characteristics by covenant trend feature

Std

P25

Median

P75

0.488 0.169

0.000 0.019

0.000 0.021

1.000 0.060

0.374 0.453 524.43 1.446 1.705 0.943

0.000 0.000 50.00 3.912 2.881 2.000

0.000 0.000 150.00 5.011 3.857 2.000

0.000 1.000 380.00 5.940 5.000 3.000

0.120 0.434 21.515 0.141 2.026 2.131 3.448 0.356 4733.09 1.639 0.247 0.140

0.000 0.000 2.344 0.029
5.948 0.108 10.000 0.000 177.48 5.179 0.126 0.018

0.038 0.000 5.542 0.078
4.779 0.492 12.000 0.000 549.48 6.309 0.256 0.035

0.074 1.000 13.375 0.145
3.574 1.083 14.000 0.000 1669.62 7.420 0.488 0.079

Non-tight-trend group

ROA Loss Interest Coverage Cash flows O-Score Leverage Rating Cov Violation Firm Size (log) Tangibility Earnings Vol

Tight trend group

Statistical difference

N

Mean

N

Mean

t-Statistic

4,089 4,089 3,680 3,833 3,892 3,913 2,094 3,565 4,131 4,100 3,961

0.032 0.193 16.783 0.116
5.085 0.802 11.118 0.113 6.471 0.338 0.077

2,575 2,575 2,388 2,410 2,497 2,342 1,509 2,303 2,612 2,604 2,487


0.001 0.344 8.434 0.081
4.126 1.553 13.441 0.205 6.128 0.311 0.089

10.89nnn
14.07nnn 15.04nnn 9.88nnn
18.99nnn
13.69nnn
21.16nnn
9.80nnn 8.67nnn 4.41nnn
3.42nnn

thresholds reach their final values ranges from 1.7 years for DSC covenants to 2.5 years for IC and DCF covenants. For most covenants, this period spans about half of the loan’s maturity. Following on our discussion in Section 2 of how a tight threshold trend allows lenders to issue longer term loans, we suggest that the length of the grace period may be indicative of what the maturity of loans of underperforming borrowers would be in the absence of the tight trend. If these borrowers were not able to commit to tight trend covenants, we posit that their loan maturity would have been similar to the time span between the loan origination date and the date of the first change in the covenant threshold. This is because lenders condition the continuation of financing on the improvement in borrower performance following this latter date. As we report in Panel E, lenders allow the loose initial covenant thresholds to last, on average, for about one year and require more demanding thresholds shortly after the beginning of the second year of the loan. Therefore, with a constant threshold covenant structure, lenders would most likely issue one-year loans and negotiate a loan renewal based on changes in the borrower’s performance. 3.3. Descriptive statistics We capture the presence of a tight trend covenant with Tight Dummy – an indicator variable that takes a value of one if the loan has at least one earnings-based covenant with a tight trend, and zero otherwise (see Appendix B for detailed Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i

N. Li et al. / Journal of Accounting and Economics ] (]]]]) ]]]–]]]

9

variable definitions).12 For loan packages that have at least one covenant with a tight trend, we also estimate the Slope variable, which captures the steepness of the tight trend at the package level. We measure Slope by averaging the trend steepness of all tight trend earnings-based financial covenants in the package. Table 2, Panel A presents the descriptive statistics. The mean values of Tight Dummy and Slope are 39% and 8.3%, respectively. Loan packages have an average size of $332.57 million. Given that a loan package typically has a number of tranches (loans) with different maturities, we compute a package’s weighted average maturity using the tranche sizes as weights and obtain an average maturity of 3.8 years. The average number of financial covenants for the sample loan packages is 2.4. About 17% of the sample loans are issued for the purposes of acquisition and 29% are institutional term loans (versus revolving lines of credit and banking term loans). We measure firm-specific characteristics in the quarter prior to the loan issuance. The average ratio of earnings before extraordinary items to total assets (ROA) is 1.9% and about 25% of the sample firms report negative earnings (Loss). The interest coverage ratio (Interest coverage) averages 13.5 and the ratio of operating cash flows to total sales (Cash flows) averages 10%. Sample firms have an average Ohlson’s (1980) bankruptcy score (O-score) of
4.71 (the higher score indicates higher credit risk) and an average Leverage, measured as the ratio of long-term debt to the book value of equity, of 1.08. For sample firms with an available credit rating, the mean S&P senior debt rating is BB. About 15% of the sample loans experience covenant violations in the year prior to the loan issuance. The sample firms are relatively large, with a mean value of total assets of $2,121 million. In Panel B, we perform a comparative analysis between borrower characteristics for the tight versus non-tight covenant sub-samples. Consistent with our prediction that underperforming borrowers are more likely to use the tight threshold trend, borrowers who obtain loans with a tight threshold feature show a significantly inferior financial performance at loan issuance. These borrowers are also more credit risky.

4. Results 4.1. Determinants of the threshold trend in earnings-based covenants We start our analyses by investigating the determinants of the presence of a tight threshold trend in earnings-based covenants. We expect these tests to shed light on whether borrowers with tight trend covenants experience inferior performance at loan initiation relative to borrowers without such covenants. We estimate the following Probit model: Tight Dummy ¼ p 1 þ p 2 Firm Controls þ p 3 Loan Controls þ p 4 Industry FE þ p 5 Year FE:

ð1Þ

Tight Dummy is defined as in previous tests. We include in the analyses a variety of firm and loan characteristics (the results of this and all subsequent tests are robust if we exclude loan controls). The model also includes industry and year fixed effects and we cluster the standard errors at the firm level (this applies to all the remaining analyses in the paper). The results reported in column 1 of Table 3 reveal that borrowers that have a weaker financial performance at loan initiation are more likely to commit to a tight trend feature. In particular, these borrowers experience losses and have lower interest coverage and operating cash flows. These results are also economically significant. Reporting a loss in the year prior to the loan’s issuance increases the probability that a tight trend is present by 11.8%, while a one standard deviation decrease in the interest coverage (cash flows) increases this probability by 6.5% (2.6%). In column 2, we augment our model with an indicator variable reflecting whether a borrower violated a financial covenant in the year prior to the loan’s issuance (this analysis is restricted to firms with available covenant violation data). Covenant violations are often associated with deterioration in a borrower’s financial performance (Roberts and Sufi, 2009a; Nini et al., 2009) and therefore may serve as an additional proxy for a borrower’s underperformance at loan initiation. We find that borrowers with recent covenant violations are more likely to commit to a tight trend. These results support our prediction that borrowers rely on tight trend covenants when they are experiencing an inferior performance at loan initiation and are in need of a grace period.13 We also find that borrowers with a higher O-score and leverage and lower asset size and tangibility have a significantly higher likelihood of tight trend covenants. This evidence suggests that more risky borrowers have stronger incentives to signal to lenders via a tight trend feature, consistent with lenders’ higher uncertainty about their future prospects relative to those of more creditworthy borrowers. In terms of loan controls, loans for acquisition purposes are more likely to be characterized by a tight trend.14 The coefficient on Term Loan B (term loan B or below – C, D, E, F) is positive, consistent with 12 We also compute an alternative measure of the use of tight trend covenants: the ratio of the number of earnings-based covenants with a tight threshold trend to the total number of earnings-based covenants in the loan package. This measure is highly correlated (0.92) with Tight Dummy, so we unsurprisingly obtain very similar results. 13 Our findings are robust when we exclude from the analyses loans with fluctuating or loose trends and when we consider a contract to have tight trend covenants only if all earnings-based covenants have a tight trend. 14 Acquirers often experience lower profitability due to the integration costs of the target firm, suggesting that they may not be able to meet strict thresholds following a loan’s initiation. Thus, to credibly convey to lenders that the acquisition will generate a stronger future performance, acquirers may need to rely on a tight trend feature.

Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i

10

N. Li et al. / Journal of Accounting and Economics ] (]]]]) ]]]–]]]

Table 3 Determinants of the presence of a tight threshold trend and Slope analysis This table presents the analyses of the determinants of the presence of a tight threshold trend (Tight Dummy) and the Slope of the tight trend. We estimate a Probit model for Tight Dummy (the reported numbers are marginal effects) and an OLS model for Slope. Regressions include year and industry fixed effects. Standard errors are clustered at the firm level. Coefficient t-statistics are in parentheses. nnn, nn and n denote significance at 1%, 5% and 10%, respectively. All variables are defined in Appendix B. Dependent variable: Tight Dummy

ROA Loss Interest Coverage Cash Flows O-Score Leverage

1

2

3

4


0.033 (
0.28) 0.118nnn (4.34)
0.003nnn (
5.37)
0.183nn (
2.08) 0.028nnn (3.70) 0.014nnn (2.88)


0.022 (
0.16) 0.099nnn (3.28)
0.003nnn (
5.26)
0.229nn (
2.44) 0.024nnn (2.92) 0.014nnn (2.79) 0.083nnn (3.03)
0.041nnn (
3.23)
0.074 (
1.48) 0.011 (0.12) 0.226nnn (9.18) 0.203nnn (10.06) 0.061nnn (4.53) 0.048nnn (7.27) 0.133nnn (11.99) 4,319 2,189 0.24


0.336nnn (
3.52) 0.036nnn (2.76) 0.000 (1.12)
0.088nn (
2.51)
0.002 (
0.67)
0.002 (
1.16)


0.277nnn (
2.56) 0.036nnn (2.68) 0.000 (1.18)
0.086nn (
2.36)
0.001 (
0.33)
0.001 (
0.64) 0.020n (1.71)
0.010n (1.77)
0.012 (
0.76) 0.015 (0.65)
0.002 (
0.36) 0.005 (0.73)
0.004 (
0.66)
0.014nnn (
5.34)
0.011nnn (
2.71) 1,655 1,089 0.19

Cov Violation Firm Size Tangibility Earnings Vol Acquisition Term Loan B Loan Amount Maturity Fin Covenants No. of obs. No. of firms R-squared

Dependent variable: Slope


0.044nnn (
3.78)
0.100nn (
2.17) 0.030 (0.38) 0.216nnn (9.46) 0.195nnn (10.67) 0.063nnn (5.11) 0.045nnn (7.50) 0.132nnn (12.93) 4,988 2,411 0.23


0.012nnn (
2.17)
0.001 (
0.08) 0.007 (0.31)
0.002 (
0.21) 0.008 (0.98)
0.005 (
0.80)
0.017nnn (
5.76)
0.011nnn (
2.71) 1,899 1,219 0.20

these loans typically being issued to more risky borrowers relative to bank term loans and revolvers.15 We also show that the presence of a tight trend is increasing with loan size, maturity and the number of financial covenants. In untabulated analyses we also control for credit rating at loan origination (when data is available) and the loan interest rate and find similar results. In addition, we rerun our tests by controlling for the initial covenant slack, as estimated by the Dichev and Skinner (2002) and Demiroglu and James (2010) slack measures,16 and accounting for the common adjustments to covenant ratios in debt contracts documented by Li (2015). We continue to find that a tight trend is more likely to be incorporated into the contracts of underperforming borrowers.17 In columns 3 and 4 of Table 3, we focus on the sub-sample of loans with a tight trend and examine the determinants of the tight trend’s steepness (Slope). We find that the “speed” with which the covenant thresholds become more restrictive is higher for borrowers with lower profitability and operating cash flows and for those experiencing losses. We interpret this evidence as suggesting that among the borrowers that signal with tight trend covenants, borrowers with a relatively more

15 This result also suggests that a tight trend is not driven by the need to mechanically update thresholds when a portion of a loan gets repaid. For bank term loans and revolvers with a front-end-loaded repayment structure that involves the amortization of the loan’s principal, as leverage decreases due to early payments, covenants may become loose, rendering them ineffective. In contrast, term loans B have a back end loaded repayment structure. 16 The first slack measure is estimated as the distance between the covenant variable and the initial threshold, scaled by the standard deviation of the covenant variable over the previous quarters. The second measure is based on the borrower’s covenant slack choice relative to the choices of borrowers with similar slack choice sets. To estimate the slack at the package level, we average the slack measures across all earnings-based covenants in the contract. 17 Although we take into account covenant adjustments suggested by Li (2015), note that the slack of earnings-based covenants can only be estimated with substantial measurement error because lenders’ adjustments to GAAP numbers often vary both across different covenants in the same contract and across different contracts, which requires tracking the precise definition of each covenant. Even if precise covenant definitions are hand collected from the contracts, some adjustments are not replicable, as the detailed data required for their estimation is not available in publicly available financial statements.

Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i

N. Li et al. / Journal of Accounting and Economics ] (]]]]) ]]]–]]]

11

inferior financial performance at the time of a loan’s origination commit to a faster acceleration of the threshold values to credibly communicate their expected performance improvement to lenders.18 4.2. Initial and final threshold values To provide support for our prediction that a tight trend endows underperforming borrowers with a grace period at loan initiation, we next examine the initial threshold values of tight trend covenants. We expect the initial thresholds in tight trend covenants to be less demanding than the corresponding threshold values used in the constant threshold covenants. We run the following model for IC, FCC and DCF covenants: Initial Threshold Value ¼ p 1 þ p 2 Tight Covenant þ p 3 Firm Controls þ p 4 Loan Controls þ p 5 Industry FE þ p 6 Year FE:

ð2Þ

Initial Threshold Value is the initial threshold in the covenant specification (for the constant threshold covenants the initial values do not change over the loan’s duration). Tight Covenant is an indicator variable equal to 1 if the covenant has a tight trend, 0 if the covenant has a constant threshold. We use several variables to reflect a borrower’s financial performance and creditworthiness at loan initiation, including the return on assets, an indicator variable reflecting losses, interest coverage, operational cash flows, the O-score and leverage. We also control for loan-specific characteristics likely to be associated with threshold values. Consistent with the univariate analyses we present in Table 1, Panel D, we find a negative (positive) relation between Tight Covenant and the initial thresholds of the IC and FCC (DCF) covenants, suggesting that the initial thresholds are significantly less restrictive in tight trend covenants than in constant threshold covenants (columns 1–3, Table 4). These results are also economically significant. The initial thresholds for IC and FCC covenants are lower by 0.4 and 0.15, respectively, while the DCF covenant’s initial threshold is higher by 0.65. These differences represent 14%, 9.6% and 18% of the mean threshold values of the respective constant thresholds. The results are consistent with tight trend covenants providing underperforming borrowers with a grace period following the loan’s initiation. Next, we examine the magnitude of final thresholds in tight trend covenants. If borrowers commit to a tight trend to signal improvement in their future performance, we expect threshold values to gradually increase over the life of the loan, potentially reaching or even surpassing the constant threshold values. We re-estimate model 2 with the final thresholds – Final Threshold Value – as the dependent variable. Across all three covenants, these thresholds are significantly more demanding than constant thresholds are (columns 4–6, Table 4). The final thresholds for the IC and FCC covenants are higher by 0.33 and 0.15, respectively, and lower by 0.25 for the DCF covenant. These differences represent 11.5%, 9.6% and 6.9% of the mean threshold values of the respective covenants for the constant threshold group. These more demanding final thresholds in tight trend covenants, coupled with the less restrictive initial thresholds, are consistent with our proposition that lenders entering into tight trend covenant contracts trade off lower covenant protection over the grace period with stronger control rights in the following periods.19 4.3. A tight threshold trend and changes in future financial performance We next focus on the realized future performance of borrowers with tight trend covenants. We expect these tests to provide further support for our two primary predictions that: 1) tight trend covenants are utilized by underperforming borrowers who need a grace period following a loan issuance and 2) tight trend covenants allow underperforming borrowers to credibly convey to lenders that they expect improvements in future performance. With respect to the first prediction, we analyze the financial performance of borrowers with tight trend covenants in the first year following the loan issuance, relative to the performance of borrowers without such covenants. The decrease in the tight trend borrowers’ performance (or no performance improvement), coupled with their underperformance at loan initiation, would indicate that these borrowers cannot commit to “standard” constant thresholds at loan initiation and need a grace period. With respect to the second prediction, we examine the financial performance of tight trend borrowers over the subsequent two year period. An improvement in their performance over this period would be consistent with these borrowers committing to tight trend covenants in order to signal their expectations of improved future performance to lenders. We focus on the changes in the profitability (Profitability),20 interest coverage (Interest Coverage) and debt to EBITDA (Debt to EBITDA) ratios of borrowers with tight trend covenants relative to those without such covenants. The univariate analysis in Table 5, Panel A indicates that over the first year after the loan issuance, tight trend covenant borrowers experience a decrease in Profitability and Interest Coverage and an increase in Debt to EBITDA. Relative to the control group of 18 Due to the high magnitude of the Min EBITDA covenants’ slope relative to other covenants, in unreported tests we repeat the Slope tests after excluding Min EBITDA covenants from the analyses. Our findings are robust. 19 In addition, by committing to more demanding final thresholds in a tight trend, borrowers who genuinely expect to improve their performance increase the signaling costs for borrowers who do not expect to improve their performance but who may want to mimic the tight trend signaling strategy. Because of the high cost of covenant violations (e.g., Chava and Roberts, 2008), stricter final thresholds make tight trend covenants costlier for borrowers who do not expect performance improvements, likely deterring them from mimicking this signaling strategy. 20 We measure Profitability as EBITDA scaled by total assets to avoid a mechanical relation between future profitability and interest payments triggered by the loan. In untabulated analyses, we also examine the performance measure based on income before extraordinary items and find similar results.

Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i

12

N. Li et al. / Journal of Accounting and Economics ] (]]]]) ]]]–]]]

borrowers, tight trend borrowers report a significantly higher increase in the debt to EBITDA ratio. However, over the second and third years following a loan’s issuance, these borrowers show a significant improvement in their interest coverage and debt to cash flow ratios, although there is still a minor decrease in profitability (Table 5, Panels B and C). These performance changes reflect a significant improvement in tight trend borrowers’ performance relative to other borrowers. To control for firm and loan characteristics that could potentially explain the ex-post performance, we estimate the following OLS model: Financial Perf ormance Change ¼ p 1 þ p 2 Tight Dummy þ p 3 Firm Controls þ p 4 Loan Controls þ p 5 Industry FE þ p 6 Year FE:

ð3Þ

Financial Performance Change is a change in one of the three future performance measures discussed above. We control for the same firm and loan characteristics as in eq. (1). In Table 6, Panel A, we show that, consistent with the univariate evidence, borrowers with tight trend covenants experience deterioration in performance in the first year following the loan issuance across all performance measures relative to borrowers who do not commit to covenants with a tight trend. Coefficients on Tight Dummy are significant, albeit only at the 10% level for Profitability and Interest Coverage. These differences are economically significant: for example, the change in profitability is less favorable by 0.8 percentage points for borrowers with a tight trend, representing 7% of the average profitability prior to the loan’s issuance.21 In Panel B, we show that borrowers with tight trend covenants subsequently improve performance across all performance metrics, as measured either over a one year period from the first to the second year following a loan’s issuance or over a two year period from the first to the third year following the issuance. The improvement in performance becomes stronger over the two year period. We view these findings as supporting our prediction that underperforming borrowers utilize a tight trend to credibly convey favorable information to lenders. In terms of control variables, we find that borrowers with a higher O-score are more likely to show improvements in their interest coverage after the loan issuance, while borrowers reporting losses prior to a loan’s issuance experience an improvement in the debt to EBITDA ratio. Borrowers with acquisition related loans significantly underperform other borrowers in the first year following the loan’s issuance, but subsequently improve their interest coverage ratio. In additional unreported analyses, we also control for the tightness of the initial covenant threshold using both the Dichev and Skinner (2002) and Demiroglu and James (2010) covenant slack measures; we find that our findings and inferences are unchanged. 4.4. A tight threshold trend and a borrower’s access to credit and loan pricing As we discuss in Section 2, we expect tight threshold trend covenants to mitigate credit rationing for borrowers experiencing inferior performance at loan initiation. To provide direct empirical evidence in support of this prediction, the ideal test would involve: 1) the analysis of loan applications and denials and the loan negotiation process and/or 2) a survey of loan officers or field experiment that would reflect on lenders’ decisions to grant credit. Because we cannot conduct these tests, we provide a number of descriptive analyses. First, we find that a significant proportion of borrowers with tight trend covenants cannot commit to constant threshold covenants because they would likely immediately violate these covenants at loan initiation. For borrowers with tight trend covenants, we estimate the “benchmark” constant threshold values that would be required by lenders, given these borrowers’ fundamentals at loan initiation, and compare them to their actual financial ratios at loan initiation. We regress the constant threshold values of the IC, FCC and DCF covenants on firm fundamentals, including O-score, size, earnings volatility, tangibility and leverage, as well as year and the twelve Fama-French industry fixed effects (as with initial and final threshold value analyses, we exclude DSC and Min EBITDA covenants from the analyses).22 We then use the estimated coefficients from these regressions to predict the benchmark constant threshold values for borrowers with tight trend covenants, given their fundamentals at loan initiation. We classify a covenant as being violated if the “benchmark” constant threshold values for these borrowers exceed their actual financial ratios at loan initiation.23 We find that 42%, 20% and 43% of IC, FCC and DCF covenants would have been violated immediately at loan initiation had the covenants received constant 21 In contrast to tight trend borrowers, borrowers utilizing tight covenant slack in constant threshold covenants to signal their type should improve or at least maintain their performance in the first year after the loan issuance. As in a classical signaling setting (e.g., Spence, 1973), strong borrowers that signal their “good” type show better future performance than borrowers who do not make such a commitment. In support of this argument, Demiroglu and James (2010) find that borrowers that commit via tight covenant slack show an improvement in financial ratios over the first four quarters following the loan issuance relative to borrowers without tight slack covenants. Similarly, Manso et al. (2010) show that borrowers with PPP experience a better performance, as measured by credit ratings changes, over the first year following the loan’s issuance, relative to borrowers with loans without PPP. 22 In the estimation of the benchmark constant threshold values, we do not account for a borrower’s profitability, interest coverage and cash flow ratios, because accounting for these performance measures could bias the benchmark threshold downward for borrowers who are temporarily underperforming at loan initiation. 23 We acknowledge the potential measurement error in the estimation of the “benchmark” constant threshold values. The measurement of a borrower’s actual ratios is also subject to a measurement error. Although we estimate these ratios by incorporating the most frequent adjustments to financial ratios in syndicated loan contracts, as documented by Li (2015), we cannot fully take into account all accounting adjustments that are used to calculate these ratios.

Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i

N. Li et al. / Journal of Accounting and Economics ] (]]]]) ]]]–]]]

13

thresholds.24 We view this evidence as suggesting that a tight threshold trend feature facilitates the access to credit for underperforming borrowers. Second, we find that the majority of underperforming borrowers commit to tight trend covenants, suggesting that lenders view this commitment as an important consideration when granting credit. We split borrowers into quartiles based on their performance measured by the IC, FCC and DCF ratios at loan initiation. We find that 64.1%, 51.5% and 76.3% of the borrowers with the weakest performance at loan initiation based on the IC, FCC and DCF ratios commit to tight threshold trend IC, FCC, and DCF covenants, respectively. This evidence further indicates that lenders are more willing to issue credit to underperforming borrowers when borrowers provide credible commitment to improve their performance in subsequent periods. Third, for each loan with tight trend covenants, we examine whether this feature is also included in other loans issued by the borrower within the 12 months period centered on the current loan’s issuance date. We find that if the borrower’s current loan has tight trend covenants, its other loans will also have these covenants in 69.9% of the cases. This consistent reliance on tight trend covenants highlights that they are the prevailing contracting mechanism for borrowers that obtain syndicated credit during a period of inferior performance. Although we cannot observe the counterfactual – underperforming borrowers who were denied credit because they could not commit to the tight trend covenants in the loan contract – we believe that the above descriptive analyses indicate that tight trend covenants facilitate access to credit for borrowers experiencing a temporarily poor performance at loan initiation. We next investigate whether borrowers would pay a higher interest rate in the absence of a tight trend feature in their loan contracts. We acknowledge that we are facing the same challenge as with the credit rationing proposition, as we cannot observe what interest rate spread borrowers would have been charged had they not been able to commit to tight trend covenants. Therefore, to shed light on whether tight trend covenants allow underperforming borrowers to obtain less expensive credit, we examine whether the covenant’s initial threshold values are priced differently for borrowers with tight trend covenants relative to those of borrowers with constant threshold covenants.25 We estimate the following OLS model: Loan Spread ¼ p 1 þ p 2 Tight Covenant þ p 3 InitialThreshold Value þ p 4 Tight Covenant nInitial Threshold Value þ p 5 Firm Controls þ p 6 Loan Controls þ p 7 Industry FE þ p 8 Year FE

ð4Þ

Our main variable of interest is the interaction term between Tight Covenant and Initial Threshold Value; both of these variables are defined as in the previous tests. We estimate the model for the IC, FCC and DCF covenants and predict a negative coefficient on the interaction term for the IC and FCC covenants and a positive coefficient for the DCF covenant, which would suggest that the initial level of covenant protection is priced more favorably if covenants have a tight trend feature. The model includes the same controls as in our previous analyses. The results in Table 7 are consistent with our predictions. The coefficients on Tight CovenantnInitial Threshold Value are negative and significant in columns 1 and 2. Economically, a one unit increase in the initial threshold value of IC (FCC) covenant reduces the interest spread by 24 (46) basis points for a tight threshold trend IC (FCC) covenant, relative to an 11 (30) point interest spread reduction for the constant threshold IC (FCC) covenant. Consistently, the coefficient on Tight CovenantnInitial Threshold Value is positive and significant in column 3. A one unit decrease in the initial threshold value of DCF covenant will reduce the interest spread by 14 basis points for a tight trend DCF covenant, relative to a 7 point reduction in the interest spread for the constant threshold DCF covenant. Overall, the evidence that the initial level of covenant protection is priced more favorably when borrowers receive tight threshold trend covenants supports our prediction that these borrowers would have paid a higher interest rate if they had not committed to the tight threshold trend feature in their loan contracts. 4.5. A tight threshold trend and covenant renegotiations While in the previous analyses we primarily focus on the benefits that tight trend covenants bring to borrowers, in this section we examine these covenants from lenders’ perspective. Our findings of significantly higher final thresholds in tight trend covenants relative to the constant thresholds (Table 4) provide evidence consistent with lenders obtaining stronger bargaining power in periods subsequent to the grace period, as more restrictive covenants endow them with more power in ex-post renegotiations (e.g., Huberman and Kahn, 1988; Aghion et al., 1994). We supplement this evidence by exploring the relation between the tight trend feature and covenant renegotiations, predicting that tight trend covenants are more likely to be renegotiated relative to constant threshold covenants, potentially providing lenders with an opportunity to extract rents in ex-post renegotiations.26 24 The percentage of covenant violations we observe for the FCC covenant is relatively low because the actual FCC ratios that we estimate are likely to be significantly overstated. The principal debt payments are typically included in the denominator of the FCC ratio, but this data is not available in Compustat, causing us to overstate the actual FCC ratios we use in covenant violation tests. 25 We thank an anonymous reviewer for suggesting this test. 26 We acknowledge that we cannot provide more direct evidence with respect to these lenders’ bargaining power and the division of surplus when the contract is renegotiated. One way to address this issue would be to collect a comprehensive sample of loan renegotiations and compare the outcomes of

Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i

N. Li et al. / Journal of Accounting and Economics ] (]]]]) ]]]–]]]

14

Table 4 A tight threshold trend and covenant threshold values This table compares the initial and final covenant threshold values of covenants with a tight trend and constant thresholds at the covenant level using OLS regressions. Tight Covenant is a dummy taking the value of 1 if a covenant’s thresholds become tighter over time, zero otherwise. Regressions include year and industry fixed effects. Standard errors are clustered at the firm level. Coefficient t-statistics are in parentheses. nnn, nn and n denote significance at 1%, 5% and 10%, respectively. All variables are defined in Appendix B. Initial threshold value

Tight Covenant ROA Loss Interest Coverage Cash Flows O-Score Leverage Tangibility Firm Size Earnings Vol Acquisition Term Loan B Loan Amount Maturity Fin Covenants No. of obs. No. of firms R-squared

Final threshold value

IC 1

FCC 2

DCF 3

IC 4

FCC 5

DCF 6


0.411nnn (
8.50) 0.831nn (2.12)
0.157nnn (
2.74) 0.002 (1.29) 0.378nn (2.09)
0.053nnn (
2.57)
0.022nn (
2.02)
0.342nnn (
2.81)
0.013 (
0.51) 0.208 (0.88) 0.002 (0.06)
0.107nn (
2.55) 0.064nnn (2.61)
0.008 (
0.69) 0.031 (1.15) 2,217 1,215 0.30


0.153nnn (
6.84) 0.220 (1.43)
0.110nnn (
3.58) 0.002nnn (2.59) 0.363nnn (2.56)
0.015 (
1.54)
0.008nn (
2.10) 0.036 (0.54) 0.077nnn (4.16)
0.003 (
0.04)
0.026 (
0.96)
0.163nnn (
7.85)
0.020 (
1.10)
0.013 (
1.61)
0.010 (
0.85) 2,029 1,186 0.25

0.654nnn (14.33)
0.119 (
0.33) 0.265nnn (3.49)
0.003nnn (
4.05)
0.148 (
0.70) 0.107nnn (6.39) 0.042nnn (3.04) 0.200n (1.69) 0.099 (3.38)
0.540nnn (
3.07) 0.138nnn (2.61) 0.024 (0.51) 0.085nnn (2.69) 0.087nnn (5.16)
0.004 (
0.15) 3,112 1,311 0.42

0.332nnn (6.44) 0.499 (1.20)
0.160nn (
2.54) 0.002n (1.79) 0.289 (1.51)
0.058nn (
2.71)
0.037nn (
2.22)
0.021 (
0.80)
0.326nn (
2.54) 0.332 (1.36) 0.030 (0.67)
0.065nn (
1.48) 0.072nnn (2.84)
0.004 (
0.33) 0.048 (1.62) 2,217 1,215 0.19

0.145nnn (5.51) 0.314nn (2.02)
0.084nn (
2.51) 0.002nn (2.51) 0.318nn (2.11)
0.016 (
1.48)
0.007 (
1.39) 0.083nnn (4.26) 0.042 (0.60) 0.032 (0.37)
0.015 (
0.51)
0.172nnn (
7.58)
0.023 (
1.18)
0.015n (
1.81)
0.013 (
1.02) 2,031 1,187 0.19


0.247nnn (
6.02) 0.677nn (2.12) 0.161nn (2.32)
0.005nnn (
7.07) 0.002 (0.01) 0.091nnn (6.32) 0.019 (1.63) 0.097nnn (3.77) 0.272nnn (2.63)
0.616nnn (
4.46) 0.061 (1.40)
0.096nn (
2.47) 0.059nn (2.15) 0.038nnn (2.66)
0.075nnn (
3.25) 3,112 1,665 0.29

Our analyses rely on a hand collected sample of loan renegotiations from Dou (2012). We obtain 1,751 contracts after matching Dou’s (2012) data with our sample. About 65% of the loans in this matched sample have their earnings-based covenants renegotiated. We caveat the measurement of the occurrence of covenant renegotiations because Dou (2012) collects data only on the first renegotiation. Therefore, if covenants were renegotiated over the life of the contract but were not amended during the first renegotiation, the dataset does not reflect this subsequent covenant renegotiation. To mitigate this concern, our renegotiation tests are based on the sample that includes contracts with earnings-based covenants amended in the first renegotiation or contracts whose earnings-based covenants were not renegotiated over the entire loan duration. We estimate the covenant renegotiation model using the following Probit regressions, where we employ the same firm and loan controls as in previous analyses: Covenant Renegotiation ¼ p 1 þ p 2 Tight Dummy þ p 3 Firm Controls þ p 4 Loan Controls þ p 5 Industry FE þ p 6 Year FE:

ð5Þ

Covenant Renegotiation is an indicator variable equal to one if a loan’s earnings-based covenant is renegotiated, zero otherwise. The coefficient on Tight Dummy reported in column 1 of Table 8 indicates that for contracts with tight trend covenants, the probability of an earnings-based covenant renegotiation is higher by 13% relative to contracts without tight trend covenants.27 (footnote continued) renegotiations for borrowers with and without tight trend covenants (and for tight trend covenant borrowers with and without performance improvement). 27 This higher probability of tight trend covenant renegotiations further suggests that committing to these covenants will be very costly for underperforming borrowers that do not expect to improve their performance. As a result, these borrowers will not mimic the tight trend signaling strategy of underperforming borrowers who genuinely expect to improve their performance.

Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i

N. Li et al. / Journal of Accounting and Economics ] (]]]]) ]]]–]]]

15

Because these analyses are performed at the loan level, we next aim to identify whether an earnings-based covenant with a tight trend or one without a trend is renegotiated. For the sample of contracts for which we have renegotiation data at the individual covenant level, we re-estimate model 5 for the IC, FCC or DCF covenants (we exclude the EBITDA and DSC covenants due to very limited renegotiation data we have available). In this analysis, the dependent variable is an indicator variable that equals one if the IC, FCC or DCF covenant is renegotiated, respectively, zero otherwise. We also substitute Tight Dummy with Tight Covenant, as previously defined, to reflect the existence of a tight trend in a specific covenant. We present the results in columns 2–4 of Table 8. Relative to covenants without tight trends, covenants with a tight trend are more likely to be renegotiated, reinforcing our findings and inferences from the loan-level analysis in column 1.28 We acknowledge that the choice of a tight trend feature is strongly associated with a borrower’s credit risk (Table 3). Because we cannot perfectly control for this risk, we cannot rule out the possibility that the significant relation between the existence of a tight trend and covenant renegotiations is attributed, at least partially, to the borrower’s riskiness. Although our findings of higher renegotiation probability of tight trend covenants suggest that lenders entering into tight trend covenant contracts are likely to benefit from stronger bargaining power following the grace period, one may argue that when a tight trend threshold exceeds the threshold typically demanded in constant threshold covenants, the borrower may threaten to exit the loan agreement by refinancing the loan with another lender. However, an important consideration in ex-post renegotiations is whether a borrower has credible alternative financing options (e.g., Roberts and Sufi, 2009b). Without the credible threat of exiting the loan agreement, a tight trend borrower would have low bargaining power in expost renegotiations even if its performance significantly improves. For instance, in times of a tight credit supply in the loan market, borrowers may find it difficult to find alternative lenders despite their better performance. In addition, the prior literature argues that the incumbent bank has an informational advantage over other potential lenders by virtue of its established relationship with the borrower; this enables the bank to prevent other banks from taking over the borrower (e. g., Dell’Ariccia et al., 1999). Thus, even if another lender is willing to step up by offering financing to the tight trend borrower, this lender may still demand a relatively high constant threshold to enhance its protection because it does not know the borrower as well as the original lender. As a result, the incumbent lender retains at least some of his bargaining power and higher renegotiation surplus. This situation is analogous to the hold-up problem documented by the relationship banking literature (e.g., Rajan, 1992; Hauswald and Marquez, 2003, 2006). To provide some evidence on the availability of alternative sources of credit financing for tight trend borrowers, we conduct untabulated analyses that focus on their loan issuance activity following the current loan initiation, conditioning on changes in their performance. We measure performance changes across three measures – Profitability, Interest Coverage and Debt to EBITDA – as defined in previous analyses. First, we find that across all these measures, only around 20% of tight trend borrowers obtain new loans following performance improvements.29 Further, more than 90% of this new financing is provided by the original lender (i.e., by the same lead arranger as in the original tight trend covenant loan). Second, we find that the new loan issuance activity and the proportion of new financing provided by the original lender do not differ significantly in almost all of our specifications across tight threshold trend borrowers with and without performance improvements. This evidence suggests that tight threshold trend borrowers do not seem to switch to alternative lenders even when they enhance their performance (either due to the hold-up problem or the tightness in the credit supply or both). Thus, their threat of exiting the existing loan agreement is relatively weak. In sum, the evidence in this section suggests that lenders’ weaker covenant protection over the grace period is likely to be compensated by an opportunity to extract rents in ex-post renegotiations over the following periods due to their stronger bargaining power and the higher renegotiation probability. This stronger bargaining power, together with an opportunity to issue longer term loans to underperforming borrowers and non-trivial amendment fees charged on loan renegotiations, explain lenders’ incentives to enter into tight trend covenant contracts. 4.6. Complementary and robustness analyses 4.6.1. The tight threshold trend and other loan terms Because not only earnings-based covenants but also other loan terms affect the allocation of control rights and bargaining power in loan contracts, we next explore the relation between the tight trend and PPP and net worth covenants with income-based escalators. We acknowledge that these analyses are impacted by endogeneity related to the joint determination of loan contractual terms and are therefore mainly descriptive. We run the following Probit model: Loan Term ¼ p 1 þ p 2 Tight Dummy þ p 3 Firm Controls þ p 4 Loan Controls þ p 5 Industry FE þ p 6 Year FE:

ð6Þ

28 In untabulated analyses, we also control for the effect of macroeconomic changes on the renegotiation probability, such as changes in GDP growth, aggregate bank leverage, stock market return, and the credit risk premium (Roberts and Sufi, 2009b). The results are unchanged. The results are also similar when we restrict the analyses to borrowers with credit ratings available to better control for credit risk or when we control for covenant slack at loan initiation. 29 We identify tight trend borrowers improving performance from the first to the second (from the first to the third) year following the loan issuance and examine their loan issuances in the third (fourth) year following the issuance.

Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i

N. Li et al. / Journal of Accounting and Economics ] (]]]]) ]]]–]]]

16

Table 5 A tight threshold trend and performance changes – univariate results This table presents the univariate results of performance changes across firms with and without tight threshold trend covenants over various time horizons. Panel A reports performance changes over the first year following the loan issuance, Panel B reports changes from the first year to the second year following the loan issuance and Panel C reports changes from the first year to the third year following the loan issuance. nnn and nn denote significance at 1% and 5%, respectively. Panel A: Year 0 to Year 1 Non-tight-trend group

Profitability Interest Coverage Debt to EBITDA Panel B: Year 1 to Year 2

Tight trend group

N

Mean

N

Mean

t-Statistic

3,882 3,432 3,287


0.012
3.705 0.437

2,393 2,575 1,898


0.009
3.015 1.071


0.98
1.53
4.84nnn

Non-tight-trend group

Profitability Interest Coverage Debt to EBITDA Panel C: Year 1 to Year 3

Tight trend group

Statistical difference

N

Mean

N

Mean

t-Statistic

3,639 3,280 3,036


0.008
0.720 0.076

2,190 2,085 1,756


0.005 0.682
0.385


1.12
4.65nnn 3.82nnn

Non-tight-trend group

Profitability Interest Coverage Debt to EBITDA

Statistical difference

Tight trend group

Statistical difference

N

Mean

N

Mean

t-Statistic

3,268 2,909 2,683


0.015
0.284 0.197

1,943 1,839 1,548


0.006 1.597
0.338


2.19nn
4.31nnn 3.02nnn

Loan Term is an indicator variable for the presence of interest increasing (decreasing) PPP or net worth covenants with income escalator. We use the same firm and loan controls as in prior tests. We find that when earnings-based covenants with a tight trend are present in the contract, lenders are less likely to employ interest increasing PPP (column 1, Table 9), which increases the interest spread when a borrower’s performance deteriorates following the loan date. Because a tight trend in threshold values requires increasingly stricter thresholds following the loan issuance, the deterioration in performance will trigger covenant violations and subsequent contract renegotiations, actually rendering the interest increasing provision ineffective for borrowers with a tight trend. In contrast, we find a positive association between the tight trend and the interest decreasing provision (column 2). This finding suggests that borrowers benefit from a reduction in loan pricing if they successfully meet more demanding threshold values. To corroborate this inference, we focus on contracts with PPP based on the DCF ratio, which is the most commonly used metric in PPP. We identify 1,132 contracts that include both an interest decreasing provision based on DCF and a DCF covenant with a tight trend. We find that in 67% of the contracts, the interest decreasing grid overlaps with the tight trend range; the initial DCF covenant threshold ratio is typically higher by 0.75 than the highest threshold on the pricing grid (remember that higher values of the DCF ratio indicates looser thresholds).30 Thus, borrowers who meet increasingly demanding thresholds over the life of the loan benefit from a decrease in the interest spread. For the remaining contracts, the interest decreasing pricing grid starts immediately below the DCF covenant’s final threshold, suggesting that these borrowers start benefiting from the interest spread reduction only after they meet the final threshold value. We next examine net worth covenants with income-based escalators, whose thresholds are also linked to a borrower’s performance (Beatty et al., 2008). There are two important differences between the tight trend and income-based escalator features. First, a tight trend imposes pre-determined stricter threshold values that a borrower must satisfy irrespective of its future performance, while the income escalator increases the net worth threshold by a percentage of net income, and only if the reported net income is positive. Second, with a tight trend feature, the timing of when a borrower has to achieve a specific threshold is precise and determined ex-ante. In contrast, the income-based escalator of the net worth covenant does not stipulate when the threshold values should be achieved, as they change as a function of a borrower’s future profitability. Only 35% of the loans in our sample of contracts with earnings-based covenants impose a net worth or tangible net worth covenant, pointing out that lenders infrequently utilize both earnings-based and net-worth-based covenants in a loan contract. However, using the sub-sample of contracts with both types of covenants, we find a positive and significant relation between Tight Dummy and the income-based

30

For our sample, DCF-based provisions represent 56.6% (91.1%) of all performance pricing provisions (financial-ratio-based pricing provisions).

Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i

N. Li et al. / Journal of Accounting and Economics ] (]]]]) ]]]–]]]

17

Table 6 A tight threshold trend and performance changes – multivariate analysis This table presents the multivariate results of performance changes across firms with and without tight trend covenants over various time horizons. Panel A reports performance changes over the first year following the loan issuance. Panel B reports changes over the one year (two year) period from the first year following a loan’s issuance to the second (third) year following a loan’s issuance. All regressions include year and industry fixed effects. Standard errors are clustered at the firm level. Coefficient t-statistics are in parentheses. nnn, nn and n denote significance at 1%, 5% and 10%, respectively. All variables are defined in Appendix B. Panel A: Year 0 to Year 1 Change in Profitability 1

Change in Interest Coverage 2

Change in Debt to EBITDA 3


0.008n (
1.85)


0.088n (
1.78)
23.407nnn (
5.65) 1.042 (1.61)

0.720nnn (3.71) 2.472 (1.37)
2.206nnn (
5.85) 0.002 (0.58)
0.421 (
0.47)
0.014 (
0.22)
0.031 (
0.53)
0.196nn (
2.00) 0.391 (0.99) 0.610 (1.01) 0.787nnn (3.16) 0.219 (1.17) 0.217nn (2.03) 0.035 (0.62)
0.074 (
0.79) 4,058 2,005 0.07

Tight Dummy ROA Loss


0.001nnn (
5.40) Cash Flows
0.001 (
0.07) O-Score 0.011nnn (5.59) Leverage
0.002 (
1.27) Firm Size 0.002 (0.92) Tangibility
0.000 (
0.01) Earnings Vol 0.016 (0.73) Acquisition
0.012nnn (
2.75) Term Loan B
0.010nn (
2.26) Loan Amount
0.005nn (
2.13) Maturity 0.001 (0.81) Fin Covenants
0.004n (
1.82) No. of obs. 4,746 No. of firms 2,291 R-squared 0.10 Panel B: Year 1 to Year 2 and Year 1 to Year 3 Interest Coverage

Change in Profitability

Tight Dummy

O-Score Leverage Firm Size Tangibility Earnings Vol Acquisition

Change in Debt to EBITDA

Year 1 to Year 3 2

Year 1 to Year 2 3

Year 1 to Year 3 4

Year 1 to Year 2 5

Year 1 to Year 3 6

0.007n (1.65)

0.012nn (2.28)

0.752n (1.95)
9.812nnn (
3.14) 0.379 (0.71)

1.471nnn (2.63)
7.906n (
1.69) 1.762nn (2.13)


0.000 (
1.31)
0.056nnn (
2.69) 0.001 (0.27)
0.001 (
0.55) 0.006nnn (2.76) 0.010 (1.27)
0.005 (
0.21) 0.006


0.0002n (
1.65)
0.051 (
2.15) 0.002 (1.03)
0.000 (
0.34) 0.006nn (2.03) 0.009 (0.79) 0.004 (0.12) 0.003

3.160n (1.81) 0.631nnn (3.51)
0.210nnn (
3.37) 0.153 (0.77)
1.479n (
1.90)
2.457 (
1.57) 1.147nnn

6.816nnn (2.84) 0.846nnn (3.23)
0.247nn (
2.25) 0.392 (1.40)
3.021 (
2.62)
5.320nn (
2.39) 1.826nnn


0.457nnn (
2.89) 1.204 (1.12)
0.721nn (
2.52)
0.009nn (
2.17) 0.178 (0.26)
0.091 (
1.61) 0.007 (0.14)
0.066 (
0.72) 0.077 (0.22) 0.429 (0.79)
0.307


0.909nnn (
3.50)
0.528 (
0.23)
1.013nn (
2.38) 0.001 (0.20)
0.203 (
0.15) 0.054 (0.64)
0.097 (
1.32)
0.210 (
1.51) 0.392 (0.61)
0.474 (
0.53) 0.334

Loss

Cash Flows

Change in Interest Coverage

Year 1 to Year 2 1

ROA

Interest Coverage

6.991nnn (3.06) 3.630nnn (12.80)
0.656nnn (
7.42) 1.604nnn (5.19)
2.464nn (
2.00)
12.956nnn (
4.21)
6.718nnn (
8.43)
1.988nnn (
3.48)
0.468 (
1.33) 0.317n (1.90) 0.389 (1.35) 4,673 2,263 0.21

Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i

N. Li et al. / Journal of Accounting and Economics ] (]]]]) ]]]–]]]

18

Table 6 (continued ) Panel B: Year 1 to Year 2 and Year 1 to Year 3 Change in Profitability

Term Loan B Loan Amount Maturity Fin Covenants No. of obs. No. of firms R-square

Change in Interest Coverage

Change in Debt to EBITDA

Year 1 to Year 2 1

Year 1 to Year 3 2

Year 1 to Year 2 3

Year 1 to Year 3 4

Year 1 to Year 2 5

Year 1 to Year 3 6

(1.27) 0.003 (0.75)
0.005nn (
2.01)
0.001 (
0.81)
0.001 (
0.48) 4,355 2,087 0.06

(0.42) 0.003 (0.52)
0.003 (
1.06) 0.000 (0.24)
0.000 (
0.16) 3,927 1,892 0.07

(2.81) 0.530 (1.42)
0.021 (
0.10)
0.024 (
0.21)
0.129 (
0.61) 4,409 2,115 0.05

(3.18) 0.351 (0.64)
0.618nn (
2.08)
0.079 (
0.48)
0.026 (
0.08) 3,917 1,879 0.07

(
1.56) 0.267 (
1.52) 0.073 (0.73) 0.024 (0.54) 0.119 (1.47) 3,674 1,795 0.03

(1.01)
0.233 (
0.84) 0.209 (1.35) 0.036 (0.45) 0.170 (1.16) 3,267 1,603 0.03

Table 7 A tight threshold trend, initial threshold values and loan spreads This table presents the effect of a tight trend in covenant threshold values and the initial covenant threshold on loan interest spreads. The dependent variable is loan spreads in basis points. Initial Threshold in column 1 (2, 3) are the initial threshold values for the IC (FCC, DCF) covenant. Tight Covenant is a dummy taking the value of one if a covenant’s thresholds become tighter over time, zero if the threshold is constant. All regressions are OLS regressions and include year and industry fixed effects. Standard errors are clustered at the firm level. Coefficient t-statistics are in parentheses. Intercepts are not reported. nnn, nn and n denote significance at 1%, 5% and 10%, respectively. All variables are defined in Appendix B.

Tight Covenant Initial Threshold Value Tight CovenantnInitial Threshold Value ROA Loss Interest Coverage Cash Flows O-Score Leverage Tangibility Firm Size Earnings Vol Acquisition Term Loan B Loan Amount Maturity Fin Covenants No. of obs. No. of firms R-squared

IC Covenant 1

FCC Covenant 2

DCF Covenant 3

69.94nnn (4.22)
11.43nnn (
3.82)
12.38nn (
2.15)
173.64nnn (
4.02) 24.41nnn (3.16)
0.210n (
1.86) 3.39 (0.15) 3.53n (1.81) 2.93nnn (2.68) 22.55nn (2.02)
7.11nn (
2.31) 58.13nnn (3.10) 22.02nnn (4.12) 61.87nnn (10.56)
17.58nnn (
4.98) 0.264 (0.17) 15.62nnn (5.62) 2,132 1,186 0.58

46.37nnn (3.49)
30.41nnn (
7.18)
15.13n (
1.81)
55.16 (
1.61) 37.70nnn (5.71)
0.400nnn (
4.08)
37.33 (
1.52) 6.44nnn (3.65) 1.96n (1.78) 14.31 (1.25)
6.72nn (
2.00) 38.99nn (2.08) 25.93nnn (4.86) 48.20nnn (9.13)
10.35nnn (
2.81)
2.58 (1.35) 14.31nnn (6.24) 1,965 1,162 0.52

11.79 (0.90) 7.13nn (2.35) 6.76n (1.85)
101.92nnn (
3.35) 34.45nnn (5.11)
0.25nnn (
3.37) 1.60 (0.08) 5.67nnn (3.44) 2.03nn (2.09) 16.34n (1.80)
11.67nnn (
4.92) 39.99nnn (2.68) 16.07nnn (3.71) 59.36nnn (13.76)
13.33nnn (
4.60)
4.54nnn (
3.33) 16.38nnn (7.34) 3,018 1,632 0.56

Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i

N. Li et al. / Journal of Accounting and Economics ] (]]]]) ]]]–]]]

19

Table 8 A tight threshold trend and covenant renegotiation This table presents the analyses of the effects of a tight trend in covenant threshold values on the likelihood of future earnings-based covenant renegotiation, using the sub-sample of contracts with available covenant renegotiation data. The dependent variable in column 1 is an indicator variable for whether one of the five earnings-based covenants is renegotiated during the life of the loan. The dependent variables in column 2 (3, 4) are indicator variables for whether the IC (FCC, DCF) covenant is renegotiated during the life of the loan. All regressions are Probit models. The reported numbers are marginal effects. All regressions include year and industry fixed effects. Standard errors are clustered at the firm level. nnn, nn and n denote significance at 1%, 5% and 10%, respectively. All variables are defined in Appendix B. Earnings-based covenant 1 Tight Dummy

Loss Interest Coverage Cash Flows O-Score Leverage Firm Size Tangibility Earnings Vol Acquisition Term Loan B Loan Amount Maturity Fin Covenants No. of obs. No. of Firms R-squared

FCC covenant 3

DCF covenant 4

0.150nn (2.46) 0.018 (0.03) 0.091 (1.09)
0.000 (
0.07)
0.177 (
0.81)
0.008 (
0.34) 0.024 (1.51)
0.042 (
1.16)
0.199 (
1.50) 0.079 (0.26) 0.019 (0.30) 0.041 (0.58) 0.041 (1.07) 0.084nnn (4.79) 0.004 (0.12) 498 389 0.14

0.163nnn (3.20) 0.014 (0.03) 0.058 (0.75)
0.002 (
1.12)
0.255 (
0.94)
0.003 (
0.15)
0.013 (
0.83)
0.080nn (
2.14) 0.089 (0.66)
0.151 (
0.89)
0.042 (
0.68) 0.035 (0.61) 0.063 (1.52) 0.036n (1.72) 0.058n (1.91) 473 368 0.11

0.100nnn (2.58) 0.238 (0.63) 0.011 (0.19)
0.001 (
0.61)
0.226 (
1.31)
0.014 (
0.85) 0.023n (1.88)
0.071nnn (
2.82)
0.000 (
0.00) 0.109 (0.70) 0.001 (0.01)
0.048 (
1.08) 0.086nnn (3.15) 0.050nnn (3.66) 0.040n (1.89) 833 619 0.09

0.128nnn (4.21)

Tight Covenant ROA

IC covenant 2


0.024 (
0.09) 0.030 (0.69)
0.000 (
0.27)
0.125 (
0.95)
0.012 (
0.95) 0.020nn (2.13)
0.065nnn (
3.47)
0.077 (
1.03) 0.028 (0.23)
0.053 (
1.43)
0.013 (
0.36) 0.069nnn (3.12) 0.048nnn (4.52) 0.060nnn (3.46) 1,395 934 0.10

escalator (column 3), suggesting that when lenders utilize both earnings-based and net-worth-based covenants, they are likely to require both sets of covenants to become increasingly demanding over time. Although our analyses in this section provides some preliminary evidence of how earnings-based covenants with a tight trend relate to PPP and net worth covenants, we leave it to future research to perform a more comprehensive analysis of the correspondence between the tight trend feature and the other debt contractual terms.

4.7. Sensitivity tests: exclusion of healthy borrowers While our focus so far has been on underperforming borrowers that use the tight trend feature to signal future performance improvements, it is possible that healthy borrowers may also rely on this feature to convey private information to lenders. Because healthy borrowers may substantially differ from borrowers who underperform at loan initiation, we attempt to identify these borrowers and test the sensitivity of our findings to their exclusion from the sample. We compare the initial thresholds of tight trend covenants with “benchmark” constant thresholds, estimated as we describe in Section 4.4. We classify each tight trend covenant as being less (more) initially restrictive than the constant threshold covenant if its initial threshold value is lower (greater) than the benchmark constant threshold. For packages (loans) that have more than one tight trend covenant, we aggregate covenant classifications at the package level. If all tight trend covenants in the package have initial thresholds below (above) the benchmark thresholds, we classify the covenant package as providing a grace period and thus characterizing an underperforming borrower (characterizing a healthy Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i

20

N. Li et al. / Journal of Accounting and Economics ] (]]]]) ]]]–]]]

Table 9 A tight threshold trend and other contractual terms This table presents the analyses of the association between the presence of a tight trend feature and the use of interest increasing performance pricing provisions, interest decreasing performance pricing provisions and net worth covenants with an income escalator. Regressions are Probit models. The reported numbers are marginal effects. When the use of an income escalator is the dependent variable (column 3), the regression is conditional on the subsample of contracts with net worth or tangible net worth covenants. All regressions include year and industry fixed effects. Standard errors are clustered at the firm level. nnn, nn and n denote significance at 1%, 5% and 10%, respectively. All variables are defined in Appendix B.

Tight Dummy ROA Loss Interest Coverage Cash Flows O-Score Leverage Firm Size Tangibility Earnings Vol Acquisition Term Loan B Loan Amount Maturity Fin Covenants No. of obs. No. of firms R-squared

Interest increasing PP 1

Interest decreasing PP 2

Income escalator 3


0.064nnn (
4.78) 0.089 (1.24) 0.010 (0.59) 0.001nnn (3.65)
0.091 (
1.11)
0.012nnn (
2.69)
0.001 (
0.29)
0.001 (
0.15)
0.027 (
0.89)
0.030 (
0.62)
0.024 (
1.46)
0.075nnn (
5.45) 0.015n (1.85) 0.010nnn (2.80) 0.006 (0.95) 2,858 1,843 0.18

0.113nnn (4.77) 0.193 (1.29)
0.074nn (
2.39)
0.001nnn (
2.59) 0.037 (0.41)
0.010 (
1.14) 0.002 (0.33)
0.054nnn (
3.96) 0.051 (0.92)
0.018 (
0.22) 0.107nnn (3.59)
0.013 (
0.54) 0.126nnn (8.80) 0.023nnn (3.26) 0.078nnn (6.74) 2,858 1,843 0.14

0.081nnn (4.48) 0.240nn (2.12)
0.019 (
0.71)
0.001nnn (
2.94) 0.002 (0.03)
0.009 (
1.25) 0.006 (0.92)
0.034nnn (
3.12)
0.030 (
0.66) 0.008 (0.14) 0.035 (1.60)
0.023 (
1.13) 0.043nnn (3.56) 0.019nnn (2.97) 0.011 (1.01) 1,764 1,141 0.13

borrower that does not need of a grace period). We remove from the analysis loan packages that have tight trend covenants with initial thresholds that are both below and above the benchmark constant thresholds. We identify 1,466 packages as providing borrowers with a grace period, but only 415 as imposing stricter initial threshold values. These statistics reinforce our inference that a tight trend feature is used in contracts primarily by underperforming borrowers who are in need of a grace period at loan initiation. They also suggest that healthy borrowers are much less likely to signal their future performance with this feature, potentially because they can signal to lenders by committing to a strict constant threshold over the entire life of the loan. Using the packages identified above as belonging to underperforming borrowers in need of a grace period, we repeat the tests reported in Tables 5–9 by performing a comparative analysis of these borrowers and those with constant threshold covenants (Table 10). For these analyses, to obtain a more homogenous constant threshold covenant group that would better reflect the constant thresholds required by lenders, we eliminate observations with studentized residuals with an absolute magnitude above 2 in the constant threshold value regressions; this constant threshold group contains 3,070 packages.31 We present the coefficients and t-statistics (or marginal effects and z-statistics for Probit models) on our main variables of interest only. All specifications are estimated using the same control variables as in the previous tests; the control variable coefficients are similar to those previously presented and omitted for brevity. Overall, we find that our results continue to hold across all analyses: the coefficients on Tight Dummy (Tight Covenant) are statistically significant in most specifications and similar in magnitude to those reported in our primary

31 These studentized residuals are a standard approach for identifying the outliers in the regression setting. Observations with studentized residuals above 2 (below -2) in the constant threshold value regressions are classified as covenants with abnormally high (low) constant thresholds. We omit from the constant threshold group all loan packages that have constant threshold covenants with abnormally high or low thresholds. In unreported robustness tests, we employ alternative cut-offs for studentized residuals, such as 1.5 and 1, and find very similar results.

Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i

N. Li et al. / Journal of Accounting and Economics ] (]]]]) ]]]–]]]

21

Table 10 Robustness tests This table replicates tests in Tables 5–9 using a sub-sample restricted to contracts identified as having tight trend covenants with initial threshold values looser than the benchmark constant threshold values (1,466 contracts) and contracts with constant threshold values covenants, excluding contracts with abnormally high or low constant thresholds (3,070 contracts). See Section 4.6 for a detailed description of these samples. All analyses are estimated with the same set of control variables as in the previously tabulated tests, but the results for these variables are not tabulated. We present the results for the main variables of interest only. Panel A: A tight threshold trend and performance changes – univariate results Non-tight-trend group N

Tight trend group

Mean

N

Year 0 to Year 1 Profitability 2,919
0.015 1,352 Interest Coverage 2,591
4.385 1,248 Debt to EBITDA 2,597 0.473 1,088 Year 1 to Year 2 Profitability 2,729
0.009 1,235 Interest Coverage 2,461
0.847 1,175 Debt to EBITDA 2,362 0.086 999 Year 1 to Year 3 Profitability 2,465
0.017 1,101 Interest Coverage 2,190
0.338 1,048 Debt to EBITDA 2,089 0.238 893 Panel B: A tight threshold trend and performance changes – multivariate analysis

Year 0 to Year 1 Tight Dummy Year 1 to Year 2 Tight Dummy

Mean

t-Statistic


0.015
4.025 1.437


0.10
0.58
5.96nnn


0.002 0.856
0.575


1.88n
4.34nnn 4.69nnn


0.001 2.050
0.571


3.58nnn
4.57nnn 3.80nnn

Change in Profitability

Change in Interest Coverage

Change in Debt to EBITDA


0.008n (
1.78)


1.813nnn (
2.69)

0.958nnn (3.74)

0.011nn (2.34)

0.586 (1.30)


0.591nnn (
3.14)

1.291nn (1.99)


1.023nnn (
3.51)

Year 1 to Year 3 Tight Dummy

0.017nnn (2.95) Panel C: A tight threshold trend, initial threshold values and loan spreads IC covenant 1
13.05nnn (
3.86) Tight CovenantnInitial Threshold Value
13.56 (
1.36) Panel D: A tight threshold trend and covenant renegotiation

Initial Threshold Value

Earnings-based covenant 0.162nnn (4.41) Panel E: A tight threshold trend and other contractual terms

Tight Dummy (Tight Covenant)

Tight Dummy

Statistical difference

FCC Covenant 2

DCF Covenant 3


31.51nnn (
5.25)
42.71n (
1.77)

7.70n (1.93) 11.72nn (2.46)

IC covenant

FCC covenant

DCF covenant

0.150nn (2.24)

0.159nnn (2.66)

0.117nnn (2.71)

Interest increasing PP

Interest decreasing PP

Income escalator


0.083nnn (
4.84)

0.074nnn (2.64)

0.069nnn (3.24)

tests. This evidence supports the robustness of our main findings and inferences to the exclusion of potentially healthy borrowers that utilize tight trend covenants.

5. Conclusion In this paper, we investigate an important feature of earnings-based financial covenants in syndicated loan agreements — the presence of covenant threshold values that change over the life of the loan. We document that these dynamic covenant thresholds are used often and that they predominantly become tighter over the loan’s duration. We further find that borrowers that commit to tight trend covenants experience losses, lower interest coverage ratios and operating cash flows as well as recent financial covenant violations, consistent with poor performance at loan initiation. We also find that, relative to covenants with a constant Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i

22

N. Li et al. / Journal of Accounting and Economics ] (]]]]) ]]]–]]]

threshold, the initial thresholds in tight trend covenants are significantly less restrictive, providing underperforming borrowers with some time to enhance their performance over a grace period. Further supporting the need for a grace period at loan initiation, we find that borrowers that commit to tight trend covenants also experience performance deterioration in the first year following the loan issuance. However, consistent with the signaling prediction, tight trend borrowers show subsequent improvements in their performance over the subsequent two year period. We also provide descriptive evidence that tight trend covenants potentially facilitate access to credit for temporarily underperforming borrowers and show that they are associated with lower loan spreads. In addition, we examine why lenders agree to offer borrowers tight trend covenants that provide them with a relatively weak protection over the grace period. Our findings of more demanding final threshold values in tight trend covenants relative to constant thresholds, as well as the higher renegotiation probability of tight trend covenants, suggest that lenders trade off weaker covenant protection during the grace period with the rent extraction opportunity in ex-post renegotiations over the following periods. Overall, we show that tight trend covenants provide underperforming borrowers who expect performance improvements with a grace period at loan initiation and a mechanism for signaling their future prospects to lenders. Thus, our findings contribute to the understanding of the critical role of covenants in debt contracting. Our findings also highlight the important role played by earnings-based covenants in the ex-ante allocation of control rights in loan contracts. Tight threshold trend covenants facilitate a dynamic allocation of control rights between the borrowers and its lenders, thus enhancing the efficiency of debt agreements. Appendix A. Earnings-based covenant example and the computation of Tight Dummy and Slope measures The loan agreement between Citadel Broadcasting Company and its creditors signed on February 10, 2000 specifies three earnings-based financial covenants: an interest coverage ratio covenant, a fixed charge coverage ratio covenant and a debt to cash flow ratio covenant (called “leverage ratio” in this contract). We provide excerpts from the loan contract where these covenants are discussed. CONSOLIDATED INTEREST COVERAGE RATIO. Permit the Consolidated Interest Coverage Ratio for any period of four consecutive fiscal quarters, in each case taken as one accounting period, ended during any period set forth below to be less than the amount set forth opposite such period below: Period January 1, 2000 April 1, 2000 July 1, 2000 October 1, 2000 January 1, 2001 April 1, 2001 July 1, 2001 October 1, 2001 January 1, 2002 April 1, 2002 July 1, 2002 Thereafter

Through Through Through Through Through Through Through Through Through Through Through

March 31, 2000 June 30, 2000 September 30, 2000 December 31, 2000 March 31, 2001 June 30, 2001 September 30, 2001 December 31, 2001 March 31, 2002 June 30, 2002 September 30, 2002

Ratio 1.50 1.50 1.50 1.50 1.75 1.75 1.75 2.00 2.00 2.25 2.25 2.50

CONSOLIDATED FIXED CHARGE COVERAGE RATIO. Permit the Consolidated Fixed Charge Coverage Ratio for any period of four consecutive fiscal quarters, in each case taken as one accounting period, to be less than 1.25 to 1.00. MAXIMUM CONSOLIDATED LEVERAGE RATIO. Permit the Consolidated Leverage Ratio at any time during a period set forth below to be greater than the ratio set forth opposite such period below: Period January 1, 2000 April 1, 2000 July 1, 2000 October 1, 2000 January 1, 2001 April 1, 2001 July 1, 2001 October 1, 2001 January 1, 2002 April 1, 2002 July 1, 2002 October 1, 2002 Thereafter

Through Through Through Through Through Through Through Through Through Through Through Through

March 31, 2000 June 30, 2000 September 30, 2000 December 31, 2000 March 31, 2001 June 30, 2001 September 30, 2001 December 31, 2001 March 31, 2002 June 30, 2002 September 30, 2002 December 31, 2002

Ratio 7.25 7.00 6.75 6.50 6.25 6.00 6.00 5.75 5.50 5.25 5.00 4.50 4.00

Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i

N. Li et al. / Journal of Accounting and Economics ] (]]]]) ]]]–]]]

23

A.1 Computation of Tight Dummy In the case of Citadel Broadcasting Company’s loan package, the interest coverage and debt to cash flow covenants become tighter over time (while the fixed charge coverage covenant has a constant threshold). Therefore, the variable Tight Dummy is coded as 1 for this contract. A.2 Computation of Slope For each earnings-based covenant with a tight trend in the contract, we estimate the steepness of the covenant trend by first computing the threshold percentage change per quarter as follows: Slopei ¼

Valuei þ 1
Valuei Datei þ 1
Datei = Valuei 90

Value is the threshold value and Date is the starting date of a covenant threshold.32 We divide the threshold percentage change by the period (stated in quarters) over which the change applies. If the covenant has n changes in thresholds, we compute a weighted average slope measure at the financial covenant level based on the slopes of each threshold change, where the weights are the time periods corresponding to each slope: Covenant Slope ¼

n X 1 Slopei nðDatei þ 1
Datei Þ: Daten
Date1 i ¼ 1

Finally, we estimate the steepness of the threshold trend at the loan contract level by averaging the individual covenant slope variables across all earnings-based covenants with a tight trend (N) in the loan package: Slope ¼

N 1X Covenant Slopej : Nj¼1

To estimate the slope of the interest coverage covenant, we code the thresholds and their dates as follows: Value1 ¼1.5, Date1 ¼March 31, 2000; Value2 ¼1.75, Date2 ¼March 31, 2001; Value3 ¼2.00, Date3 ¼ December 31, 2001; Value4 ¼2.25, Date4 ¼June 30, 2002; Value5 ¼2.50, Date5 ¼December 31, 2002. The slope of the interest coverage covenant is computed as: P4   ð1:75
1:5Þ ð2
1:75Þ ð2:25
2Þ ð2:5
2:25Þ 90 i ¼ 1 ½Slopei  ðDatei þ 1
Datei Þ þ þ þ  ¼ 0:049: ¼ Date5
Date1 1:5 1:75 2 2:25 1005 The slope of the debt to cash flow covenant is calculated in a similar way and is equal to
0.047. Note that decreasing threshold values for the debt to cash flow covenant means a tighter covenant over time; thus we multiply its slope by
1. Finally, we estimate the steepness of the threshold trend at the loan contract level by averaging the individual covenant slopes across all earnings-based covenants with the trend in the package. Therefore, the Slope for this contract is (0.049þ 0.047)/2 ¼0.048, which means that at the contract level, the covenant trend gets tighter by 4.8% per quarter. Appendix B. Covenant and variable definitions B.1 Definitions of covenants Covenant

Description

Interest Coverage (IC) Covenant Fixed Charge Coverage (FCC) Covenant Debt Service Coverage (DSC) Covenant Debt to Cash Flow (DCF) Covenant Minimum EBITDA (Min. EBITDA) Covenant

Earnings (e.g., EBITDA, EBIT, operating income, etc.) divided by interest expense. Earnings (e.g., EBITDA, EBIT, operating income, etc.) divided by fixed charges (e.g., interest expense, principal payments, capital expenditures, dividend payments, etc.). Earnings (e.g., EBITDA, EBIT, operating income, etc.) divided by debt service (e.g., interest expense and principal payments). Debt (e.g., funded debt, senior debt, etc.) divided by earnings (e.g., EBITDA, EBIT, operating income, etc.).

32

Minimum earnings level (e.g., EBITDA, adjusted EBITDA, operating income, etc.).

We scale Valuei þ 1
Valuei by the absolute value of Valuei if V aluei is negative. The slope is not well defined if Valuei equals zero.

Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i

N. Li et al. / Journal of Accounting and Economics ] (]]]]) ]]]–]]]

24

B.2 Definitions of variables used in the empirical tests

Covenant variables Final Threshold Value Initial Threshold Value Tight Covenant Tight Dummy Slope

Control variables Acquisition Cash Flows Covenant Violation Covenant Renegotiation Debt to EBITDA Decreasing PP Earnings Vol Financial Covenants Firm Size Income Escalator Increasing PP Interest Coverage Leverage Loan Amount Loss Maturity O-Score

Profitability Rating

ROA Tangibility Term Loan B

Description The last covenant threshold value as specified in the loan contract. The first covenant threshold value as specified in the loan contract. An indicator variable equal to one if the covenant has a tight trend, zero otherwise. An indicator variable equal to one if at least one earnings-based covenant in the loan contract has a tight trend, zero otherwise. The mean of the slopes of the earnings-based covenants with a tight trend in the loan package. The slope of each covenant is calculated as ½ðV aluen
Valuen
1 ÞÞ=V aluen
1 þ ðValuen
1
Valuen
2 ÞÞ=V aluen
2 þ … þ ðValue2
Value1 ÞÞ=Value1   90=ðDaten
Date1 Þ , where Valuei is the ith threshold value, Datei is the first date using Valuei , and n is the number of threshold values. See Appendix A for a more detailed description. Description An indicator variable equal to one if the loan purpose is acquisition financing (acquisition line and takeover), zero otherwise. Operating cash flows scaled by sales. An indicator variable equal to one if the firm violates any financial covenant within one year prior a loan’s issuance date, zero otherwise. Indicator variable equal to one if one of a loan’s earnings-based covenants is renegotiated, zero otherwise. The ratio of total debt to EBITDA, calculated for firms with positive EBITDA. An indicator variable equal to one if the loan package contains an interest decreasing performance pricing provision, zero otherwise. Standard deviation of the annual return on assets over the five year period prior to a loan’s issuance date. Number of financial covenants in the loan agreement. Natural logarithm of total assets. An indicator variable equal to one if an income escalator feature in net worth or tangible net worth covenants is present in a loan contract, zero otherwise. An indicator variable equal to one if the loan package contains an interest increasing performance pricing provision, zero otherwise. EBITDA divided by total interest expense, where EBITDA is calculated as earnings before extraordinary items plus the interest expense, tax expense and depreciation and amortization expenses. It is measured using four quarter rolling data. Long-term debt divided by the book value of equity. Natural logarithm of the loan package amount. An indicator variable equal to one if the firm reported negative earnings, zero otherwise. Weighted average of the maturities (in years) across all tranches in the loan package, weighted by the size of each respective tranche. Olsson’s (1980) score: O-Score¼–1.32–0.407  log(total assets/GNP price-level index) þ6.03  (total liabilities/total assets)
1.43  (working capital/total assets)þ 0.076  (current liabilities/current assets)
1.72  (1 if total liabilities 4total assets, else 0) – 2.37  (net income/total assets)
1.83  (funds from operations/total liabilities)þ 0.285  (1 if net loss for the last two years, else 0)
0.521  (net income
lag net income)/ (|net income|þ |lag net income|). The ratio of EBITDA (income before extraordinary items plus interest, tax, depreciation and amortization expenses) to total assets, measured using four quarter rolling data. The numerical equivalent of firms’ senior debt rating: 1 for the highest and 23 for the lowest. Ratings are obtained from the Standard and Poor's (S&P) historical database. If S&P’s ratings are not available, we retrieve the Moody’s, Fitch or Duff and Phelps senior debt rating from the Mergent Fixed Income Securities Database (FISD). For borrowers with missing ratings on S&P and FISD, we collect ratings from the Internet-based version of DealScan. Return on assets computed as the ratio of net income before extraordinary items to total assets, measured with four quarter rolling income data. Net properties, plant, and equipment scaled by total assets. An indicator variable equal to one if the loan package contains a term loan B or below (C, D, E and F), 0 otherwise.

References Aghion, P., Bolton, P., 1992. An incomplete contracts approach to financial contracting. Review of Economic Studies 59, 473–494. Aghion, P., Dewatripont, M., Rey, P., 1994. Renegotiation design with unverifiable information. Econometrica 62 (2), 257–282. Asquith, P., Beatty, A., Weber, J., 2005. Performance pricing in bank debt contracts. Journal of Accounting and Economics 40, 101–128. Beatty, A., Webber, J., Yu, J., 2008. Conservatism and debt. Journal of Accounting and Economics 45, 154–174. Berlin, M., Mester, L.J., 1992. Debt covenants and renegotiation. Journal of Financial Intermediation 2, 95–133. Besanko, D., Thakor, A., 1987. Collateral and rationing: sorting equilibria in monopolistic and competitive credit markets. International Economic Review 28, 671–689. Bolton, P., Freixas, X., Gambacorta, L., and Mistrulli P.E. 2013. Relationship and transaction lending in a crisis. Working Paper. Bradley, M., Roberts, M.R., 2004. The Structure and Pricing of Corporate Debt Covenants. Working Paper. Duke University. Chan, Y., Kanatas, G., 1985. Asymmetric valuations and the role of collateral in loan agreements. Journal of Finance 17, 84–95.

Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i

N. Li et al. / Journal of Accounting and Economics ] (]]]]) ]]]–]]]

25

Chava, S., Roberts, M.R., 2008. How does financing impact investment? The role of debt covenants. Journal of Finance 63, 2085–2121. Christensen, H., Nikolaev, V., 2012. Capital versus performance covenants in debt contracts. Journal of Accounting Research 50, 75–116. Christensen H., Nikolaev V. and Wittenberg-Moerman R. 2015. Accounting information in financial contracting: an incomplete contract theory perspective. Working Paper. DellíAriccia, G., Friedman, E., Marquez, R., 1999. Adverse selection as a barrier to entry in the banking industry. RAND Journal of Economics 30, 515–534. Demerjian, P., 2011. Accounting standards and debt covenants: has the ‘balance sheet approach’ led to a decline in the use of balance sheet covenants? Journal of Accounting & Economics 52, 178–202. Demiroglu, C., James, C., 2010. The information content of bank loan covenants. Review of Financial Studies 23, 3700–3737. Dewatripont, M., Tirole, J., 1994. A theory of debt and equity: diversity of securities and manager-shareholder congruence. Quarterly Journal of Economics 109, 1027–1054. Diamond, D., 1991. Monitoring and reputation: the choice between bank loans and directly placed debt. Journal of Political Economy 99, 688–721. Diamond, D., 1993. Seniority and maturity of debt contracts. Journal of Financial Economics 33, 341–368. Dichev, I., Skinner, D., 2002. Large-sample evidence on the debt covenant hypothesis. Journal of Accounting Research 40, 1091–1123. Dou, Y., 2012. The debt-contracting value of accounting numbers, renegotiation, and investment efficiency. Working Paper. New York University. Drucker, S., Puri, M., 2009. On loan sales, loan contracting, and lending relationships. Review of Financial Studies 22, 2835–2872. Fang, S., 2011. State Independent Contractual Adjustments in Financial Covenants. Working Paper. University of Rochester. Freudenberg, F., Imbierowicz, B., Saunders, A., Steffen, S., 2012. Covenant violations, loan contracting, and default risk of bank borrower. Working Paper. NYU. Garleanu, N., Zwiebel, J., 2009. Design and renegotiation of debt covenants. Review of Financial Studies 22, 749–781. Hart, O., 2001. Financial contracting. Journal of Economic Literature 34, 1079–1100. Hart, O., Moore, J., 1988. Incomplete contracts and renegotiation. Econometrica 56, 755–785. Hauswald, R., Marquez, R., 2003. Information technology and financial services competition. Review of Financial Studies 16, 921–948. Hauswald, R., Marquez, R., 2006. Competition and strategic information acquisition in credit markets. Review of Financial Studies 19, 967–1000. Huberman, G., Kahn, C., 1988. Limited contract enforcement and strategic renegotiation. American Economic Review 78, 471–484. Leftwich, R., 1983. Accounting information in private markets: evidence from private lending agreements. Accounting Review 58, 23–42. Li, N., 2015. Performance Measures in Earnings-Based Financial Covenants in Debt Contracts. Working Paper. University of Texas at Dallas. Manso, G., Strulovici, B., Tchistyi, A., 2010. Performance-sensitive debt. Review of Financial Studies 23 (5), 1819–1854. Murfin, J., 2012. The supply-side determinants of loan contract strictness. Journal of Finance 67, 1565–1601. Nini, G., Smith, D.C., Sufi, A., 2009. Creditor control rights and firm investment policy. Journal of Financial Economics 92, 400–420. Nini, G., Smith, D.C., Sufi, A., 2012. Creditor control rights, corporate governance, and firm value. Review of Financial Studies 25, 1713–1761. Ohlson, J., 1980. Financial ratios and the probabilistic prediction of bankruptcy. Journal of Accounting Research 19, 109–131. Petersen, M., Rajan, R., 1994. The benefits of lending relationships: evidence from small business data. Journal of Finance 49, 3–37. Petersen, M., Rajan, R., 1995. The effects of credit market competition on lending relationships. Quarterly Journal of Economics 110, 407–443. Rajan, R., 1992. Insiders and outsiders: the choice between informed and armslength debt. Journal of Finance 47 (4), 1367–1400. Roberts, M., Sufi, A., 2009a. Control rights and capital structure: an empirical investigation. Journal of Finance 66, 1657–1695. Roberts, M., Sufi, A., 2009b. Renegotiation of financial contracts: evidence from private credit agreements. Journal of Financial Economics 93, 159–184. Roberts, M. 2015. The role of dynamic renegotiation and asymmetric information in financial contracting. Journal of Financial Economics (116): 61-81. Thompson, J., Ruby, C.L., 2014. Global trends in leveraged lending, 2nd edition Global Legal Group, London, UK chapter 5.

Please cite this article as: Li, N., et al., Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics (2015), http://dx.doi.org/10.1016/j.jacceco.2015.07.004i