Review of Accounting Studies

, Volume 13, Issue 1, pp 87–118

GAAP goodwill and debt contracting efficiency: evidence from net-worth covenants

Authors

    • Olin School of BusinessWashington University in St. Louis
  • Chandra Seethamraju
    • Mellon Capital Management
  • Tzachi Zach
    • Olin School of BusinessWashington University in St. Louis
Article

DOI: 10.1007/s11142-007-9045-0

Cite this article as:
Frankel, R., Seethamraju, C. & Zach, T. Rev Acc Stud (2008) 13: 87. doi:10.1007/s11142-007-9045-0

Abstract

We study the role of goodwill in promoting contracting efficiency and the effect of SFAS 141 and 142 on this role. We provide three main results. First, when a lending agreement contains some type of minimum net-worth covenant, the probability of a tangible net-worth covenant is decreasing in the borrower’s goodwill. Second, the use of tangible net-worth covenants has increased since the promulgation of SFAS 141 and 142. Finally, covenant slack is not significantly related to the use of tangible net-worth covenants relative to net-worth covenants. These results suggest that contracting parties realize efficiency gains by permitting borrowers’ actions to be restricted by the value of GAAP goodwill. However, recent changes in GAAP have reduced the contracting usefulness of goodwill.

Keywords

Contracting efficiencyGoodwillDebt covenants

JEL Classification

M41

1 Introduction

This paper addresses two questions: (1) whether incorporating goodwill in debt covenants is efficient; and (2) how SFAS 141 and 142 have affected the contracting role of goodwill. Our research design differs from the extensive literature that examines the effects of existing debt covenants on accounting method choice and firm value (For example, Defond & Jiambalvo, 1994, and Sweeney, 1994). Prior research looks for accounting and economic effects given pre-existing covenants. In contrast, we take an “ex ante” approach and assume covenants are not fixed. That is, we examine how the firm’s existing accounting methods and mandatory changes in those methods shape covenants. 1 Assuming covenants reflect the value-maximizing choices of management (Smith & Warner, 1979), an “ex ante” approach allows us to understand the efficiency implications of accounting methods and regulations.

Recorded goodwill affects contracting efficiency when lending agreements allow goodwill to restrict borrowers’ actions. Restrictions can occur when a lending agreement has a net-worth covenant and the borrower’s balance sheet contains goodwill. Under these circumstances, goodwill write-downs can lead to violation of the net-worth covenant or restrict the borrower’s ability to pay dividends (Beatty & Weber, 2006). If allowing goodwill changes to trigger or relax restrictions on borrowers is not efficient, the net-worth covenant can be neutralized with respect to goodwill changes by using tangible net worth instead of net worth as the covenant benchmark. The substitution of tangible net worth for net worth will occur (1) if allowing GAAP goodwill to affect net-worth restrictions increases agency costs arising from the conflicting incentives of borrowers and lenders and, (2) if additional negotiation, verification and record-keeping costs associated with excising goodwill are not significant.

With the release of SFAS 141 and SFAS 142, the Financial Accounting Standards Board shifted GAAP from the amortization of goodwill to the valuation of goodwill and, by eliminating the pooling method, increased the probability that acquirers will record goodwill in an acquisition. The FASB cited the need for “better information about intangible assets” as a reason for issuing SFAS 142, Goodwill and Other Intangible Assets. Yet Watts (2006) argues that the GAAP value of goodwill is typically not verifiable and that SFAS 141 and SFAS 142 are examples of the “soft accounting standards generated by standard setters” that will provide incentives for the creation of alternative accounting systems (For example, Standard and Poor’s “core earnings”). His argument implies that debt covenants will tend to exclude goodwill because allowing goodwill to affect covenant restrictions increases agency costs. To support his point, he cites research by Leftwich (1983), which provides evidence that debt covenants exclude goodwill when measuring total assets. This debate provides the motivation for our examination of the use of goodwill in debt covenants.

We investigate the use of goodwill in contracting by studying whether lending agreements include a minimum net-worth or a minimum tangible net-worth requirement, given that the borrower and lender have decided to include some type of net-worth requirement in the lending agreement. We use the Dealscan database to identify firms that have net-worth covenants and tangible net-worth covenants. The choice between these covenants depends on their relative net contracting benefits. We find that the probability of a tangible net-worth covenant is decreasing in the level of the borrower’s goodwill. This result implies that lenders believe goodwill is informative and provides an efficient means of limiting agency costs. 2

Our study also provides insight into the effects of the adoption of SFAS 141 and SFAS 142 on the contracting role of goodwill. Because SFAS 141 and 142 were implemented simultaneously, our before-and-after tests are of the joint effects of these standards. We find that the use of tangible net-worth covenants has increased since SFAS 141 and SFAS 142 came into effect. This result suggests that promulgation of SFAS 141 and SFAS 142 is associated with a reduction in the contracting efficiency of goodwill.

Comparing covenant slack associated with tangible net-worth covenants and net-worth covenants provides additional evidence on the relative costs and benefits of tangible net-worth and net-worth covenants. We measure covenant slack as the difference (scaled by total assets) between actual net worth in the fiscal year before a loan’s initiation and the minimum net worth required by the loan. Increased covenant slack provides the lender with less control and can therefore increase moral hazard costs. However, reduced covenant slack increases the likelihood of covenant violation and can lead to increased renegotiation costs. We find that the level of covenant slack is not significantly related to the use of tangible net-worth requirements. This result suggests that tangible net-worth covenants and covenant slack are neither complements nor substitutes.

In sum, our results offer some support for both views in the debate over the role of goodwill in promoting contracting efficiency. Our evidence suggests that recorded goodwill provides a useful input to the firm’s lending contracts. On the other hand, our results are consistent with the view that recent changes in goodwill accounting standards have reduced the net contracting benefits of goodwill.

1.1 Caveats

Our empirical models represent a reduced form of the underlying cost and benefit functions. Therefore, the interpretation of our results is limited. For example, we cannot identify the changes to specific costs associated with use of a tangible net-worth covenant. We discuss costs and benefits to highlight plausible underlying trade-offs faced by borrowers and lenders. Our tests regarding the effects of SFAS 141 and SFAS 142 are subject to the caveat that the standard-setting process is not modeled and therefore assumed to be exogenous (Wolfson, 1980).

2 Hypotheses development

In this section we develop our hypotheses on the choice between a net-worth and a tangible net-worth covenant. We begin by discussing the role of minimum net-worth-based covenants, in general. We then turn to the role that goodwill plays in the choice between a net-worth covenant and a tangible net-worth covenant, given that a net-worth-based covenant is selected. Next, we develop our hypothesis for the effects of SFAS 141 and SFAS 142 on the relative frequency of tangible net-worth covenants. Finally, we discuss our hypothesis for the relative difference in covenant slack expected in tangible net-worth compared to net-worth covenants.

2.1 Net-worth-based covenants

Net-worth-based covenants (or minimum worth covenants) reduce the agency costs of debt in several ways. First, these covenants limit the exposure of lenders to the possibility of a liquidating dividend. The payment of dividends by the borrowing corporation transfers wealth from the lender to the borrower. 3 When the debt is nearly risk-free, these wealth transfers are negligible, but the conflict between lenders and borrowers with regard to dividends grows as the equity cushion supporting the debt declines. The borrower will also be less inclined to make the investments necessary to maintain assets because additional investments may eventually be used to pay the lender’s claims, rather than providing returns to the borrower (Myers, 1984). By requiring a minimum level of shareholders’ equity and periodically evaluating compliance before the repayment of his loan, the lender protects (and the borrower bonds) himself against liquidating dividends (Kalay, 1982).

Second, establishing a minimum level of worth provides the lender with a minimum level of security, or collateral value, in the event that the borrower defaults. A violation of a net-worth covenant provides warning to the lender that the liquidation value of net assets is declining (Dichev & Skinner, 2002; Smith, 1993). It also allows the lender to renegotiate the loan before further impairment of his collateral.

Third, net-worth-based covenants reduce the borrower’s incentive to increase the firm’s asset-value variance. A higher variance increases the value of the call option granted by the lender to the borrower (Galai & Masulis, 1976), but a higher variance also increases the probability of covenant violation and, with it, cancellation of the option before debt maturity. Thus, while the borrower gains option value by increasing variance, he loses option value by reducing option life. 4

Finally, net-worth-based covenants reduce adverse selection arising from the information asymmetry between the lender and the borrower. A borrower can credibly convey his private information on the prospects of his firm by accepting a covenant because covenant violation is costly to the borrower. The costs arise because the lender can use his renegotiation option for private gains. For example, if market-wide interest rates have increased and a covenant is violated, the lender can choose to renegotiate even if there is little change in the borrower’s default risk. An example of this opportunism arises in the case of firms that delay their financial statement filings to review option backdating practices. Several investors who bought long-term public debt at prices below par value view these late filings as violations of debt covenants and demand either the immediate repayment of the debt or substantial fees to extend their default deadlines. 5 Covenant violations can also be costly to the borrower because they can lead to additional burdens on management. For example, managers may be forced to supply updated financial statements or justify forecasts and strategies to borrowers (Dichev & Skinner, 2002). Therefore, by accepting costly covenants, borrowers credibly convey their private information with regard to the eventual violation of the covenant. 6

In the discussion that follows, we assume that the benefits described above are provided by both tangible net-worth and net-worth covenants. However, these benefits can vary across these covenant types. We discuss why these benefits are likely to vary as well as the other important incremental costs that can affect the net benefits provided by a given covenant type.

2.2 Tangible net worth versus net worth

Given that the lender and borrower agree to include a minimum worth requirement in the lending agreement, we now examine the benefits and costs of including a minimum net-worth requirement versus a minimum tangible net-worth requirement. We argue that this choice is related to the current and expected amount of goodwill in the firm’s balance sheet because the level of goodwill is an important driver of the costs and benefits. In this discussion we analyze (1) contracting benefits arising from the expected correlation between reported goodwill and the value of the debt, given moral hazard in the reporting decisions of managers; (2) the incremental expected implementation costs, which include initial negotiation and monitoring costs of a tangible net-worth requirement over a net-worth requirement; and (3) the incremental expected renegotiation costs of a net-worth requirement over a tangible net-worth requirement in case of a covenant violation.

2.2.1 Comparing incremental benefits

Debt covenants provide efficiency gains, by reducing moral hazard and adverse selection arising in the relations between borrowers and lenders. Covenants alert the lender to borrower actions that simultaneously increase the borrower’s expected payoffs and reduce the value of the lender’s claim (Smith & Warner, 1979). More important, when covenants are violated, the lender is granted specific rights, such as the ability to demand repayment. In this way, covenants mitigate potential lender/borrower conflicts, reducing contracting costs associated with borrowing and increasing the value of the firm.

Conflicts between borrowers and lenders may become more pronounced when the default probability is higher. Given that incentive conflicts become more pronounced as the lender’s expected payoff declines, the lender will use covenants to monitor credit quality and to gain additional rights when credit quality deteriorates. Thus, we assess the relative benefits of tangible net-worth and net-worth covenants in light of their usefulness in monitoring the lender’s expected payout.

A lender’s expected payoff depends on the value of the firm relative to the promised payment on the debt and the expected payoff in the event of default. To the extent that goodwill provides additional information about the lender’s expected payoff, it provides monitoring benefits. For example, the write-down of goodwill can indicate that the value of the firm is approaching the promised payment on the debt. In this case, including a net-worth rather than a tangible net-worth covenant improves monitoring efficiency because it allows the reduction of goodwill to influence the covenant.

On the other hand, if goodwill value is uncorrelated with firm value and goodwill has no value in a bankruptcy, it would be inefficient to include goodwill in debt covenants. Allowing a goodwill write-down to create additional lender rights results in a potential obligation for the borrower with no offsetting ex ante reduction in moral hazard or adverse selection. If creating these rights is costly, then a lending contract using a net-worth requirement will be less efficient that a lending contract using a tangible net-worth requirement. If a tangible net-worth covenant provides a less noisy measure of the lender’s expected payoff and avoids the costs associated with the arbitrary and costly creation of lender rights, it would offer incremental benefits.

2.2.2 Comparing incremental costs

Two costs arise from the process of administering debt covenants. The first is the cost of implementing the contract. 7 The costs of implementing a tangible net-worth covenant can be higher than those of a net-worth covenant. The initial negotiations surrounding the computation of non-GAAP terms are likely to be more complex than those associated with negotiating GAAP terms. We find cross-sectional variation in the definition used in covenants for tangible net worth suggesting that a widely accepted definition does not exist and that incremental negotiation costs can arise. For example, some contracts include unrealized gains on securities in the definition of intangible assets while other contracts do not. In addition, subsequent monitoring of these covenants can be incrementally costly. Some companies do not clearly identify intangible assets on their balance sheet. In Intel’s 2005 annual report, one must go to the footnotes to find the company has $654 million of “identified intangible assets.” On the balance sheet one only finds a line item for “deferred taxes and other assets.” Thus, removing intangible assets is not a mechanical process. The importance of these incremental monitoring and negotiation costs is an empirical question. The results of El-Gazzar and Pastena (1990) suggest that non-GAAP covenant terms are more costly to negotiate and monitor than GAAP covenant terms. They find that the exclusion of goodwill and other intangibles from equity is less likely in bank loans than in loans granted by insurance companies. The former are more likely to be syndicated, increasing the costs of monitoring or negotiating tailored accounting.

The second cost is the expected cost of renegotiating a loan when a covenant is violated. Expected renegotiation costs are increasing in the probability that a covenant will be violated. When firms have goodwill we argue that the expected renegotiation costs of a net-worth covenant are higher than those of a tangible net-worth covenant because a goodwill write-off increases the probability that a net-worth covenant will be violated but does not affect the probability of violating a tangible net-worth covenant (Beatty & Weber, 2006). 8 Because renegotiation is determined at the lender’s option, the decision to renegotiate reflects the lender’s desire to maximize the value of the debt, not the value of the firm. This ex post deviation from market-value maximization adds to expected renegotiation cost.

2.2.3 Comparing incremental costs and benefits

Given that neither covenant provides higher costs and lower benefits in all circumstances, the choice between covenants reflects the maximization of net benefits. The above discussion suggests that the incremental monitoring costs and initial negotiation costs of a tangible net-worth covenant exceed those of a net-worth covenant. Thus, the use of a tangible net-worth covenant implies that this incremental cost is offset by the contracting benefits. That is, excluding goodwill from the determination of covenant violation increases the correlation between covenant violation and deterioration in the borrower’s creditworthiness.

In summary, we expect the relation between the magnitude of goodwill and the incidence of tangible net-worth covenants to depend on the relative costs and benefits of each type of covenant. If goodwill provides no information about either collateral value or the probability of default, then we expect a positive relation between the probability of a tangible net-worth covenant and goodwill. In this case, goodwill adds noise to the covenant, and the negative effects of this noise are increasing in the amount of goodwill. For example, more goodwill implies that goodwill write-offs are more likely to lead to technical default, leading to higher expected renegotiation costs. Therefore, as the value of goodwill increases, benefits to removing this noise increase. Assuming the costs of removing goodwill rise less sharply than the benefits of removing goodwill, we expect the probability of a tangible net-worth covenant to increase in the level of goodwill. If, on the other hand, goodwill is correlated with the lender’s expected payout, a negative relation between the use of tangible net-worth covenants and the magnitude of goodwill is possible. In this case, the contracting benefits foregone by ignoring the value of goodwill (for example, using a tangible net-worth covenant) are increasing in the magnitude of goodwill. When goodwill is a small percentage of total assets, a covenant can ignore changes in goodwill without materially affecting contracting efficiency. However, as goodwill grows ignoring the value of goodwill significantly reduces the correlation between the covenant benchmark and the lender’s expected payout. This discussion leads to the following two-sided null hypothesis

H1

There is no relation between the magnitude of recorded goodwill and the propensity of a firm to use a tangible net-worth rather than a net-worth requirement in its lending agreements.

2.3 SFAS 142 and SFAS 141 and the propensity to use tangible net-worth covenants rather than net-worth covenants

2.3.1 SFAS 142

According to SFAS 142, goodwill has an indefinite life and therefore should not be amortized. SFAS 142 thereby alters prior GAAP, which required goodwill to be amortized over a period not exceeding 40 years. 9 Under SFAS 142, goodwill is written down when it is impaired. 10 While goodwill write-downs could occur under prior GAAP, SFAS 142 gives write-downs significantly more prominence by eliminating amortization and requiring a formal test of impairment. Moreover, SFAS 142 requires managers (and their auditors) to engage in the subjective practice of business-unit valuation. If SFAS 142 increases the frequency of goodwill write-offs (and thus increases expected renegotiation costs associated with net-worth covenants) without an offsetting increase in the benefits derived from improved correlation between goodwill carrying values and the expected payoff to lenders, then we would expect an increase in the use of tangible net-worth covenants relative to net-worth covenants after the implementation of the standard. If, on the other hand, the specific guidance provided by SFAS 142 for a substantive test of goodwill impairment increases the relevance of reported goodwill, its adoption can lead to an increase in the use of net-worth covenants.

2.3.2 SFAS 141

The accounting procedures used for acquisitions also have implications for the choice between tangible net-worth covenants and net-worth covenants. By proscribing use of the pooling method to account for acquisitions, SFAS 141 alters expected accounting procedures and can change the frequency of tangible net-worth covenants relative to net-worth covenants. The usual argument is that the purchase method lowers income by requiring amortization of goodwill. Therefore, firms seeking to relax pre-existing debt covenants will prefer the pooling method (for example, Aboody, Kasznik, & Williams, 2000; Ayers, Lefanowicz, & Robinson 2002; Weber, 2004). However, this argument assumes that the only metric that concerns firms with debt covenants is reported income. This argument does not apply to firms that have pre-existing net-worth covenants. 11 When the fair market value of the target’s net assets exceeds the book value of net assets, an acquisition accounted for under the purchase method adds this difference to the shareholders’ equity of the acquirer. The difference between the book value of net assets and their market value is not added to shareholders’ equity when the pooling method is used. Thus, a purchase-method acquisition has a greater potential to loosen net-worth covenants than a pooling-method acquisition.

The relaxation of the net-worth covenant caused by a purchase-method acquisition is not accompanied by a strengthening of the lenders position. Cash acquisitions allow firms to distribute cash to target shareholders without affecting net worth. Moreover, when the acquisition gives rise to goodwill, an intangible asset displaces one with higher collateral value—cash—further weakening the security of prior debt. Therefore, a cash acquisition can reduce the expected value of the lender’s claim without violating a net-worth requirement. In a stock-based purchase-method acquisition, shareholders’ equity increases thereby adding slack to the net-worth covenant. However, a potentially large portion of the added shareholders’ equity is related to the write-up of assets and goodwill. Both sources have questionable collateral value. In this way, purchase-method acquisitions can impair the ability of a pre-existing net-worth requirement to reduce moral hazard and adverse selection costs. 12

A tangible net-worth requirement limits the ability of an acquisition to undermine net-worth requirements. Goodwill arising from a stock-based purchase-method acquisition does not add slack to the covenant while goodwill acquired in a cash-based acquisition leads to tighter covenants. Lenders and borrowers will anticipate the effects of expected future acquisitions on net-worth covenants when they negotiate lending agreements. Therefore, we expect the link between the use of tangible net-worth requirements and acquisitions to strengthen after SFAS 141, given that SFAS 141 rules out the use of pooling and thereby increases the probability that goodwill will arise in an acquisition. 13

In sum, both SFAS 141 and SFAS 142 could lead to an increase in the use of tangible net-worth covenants. Both these standards affect financial statements in fiscal 2002 and beyond. Given their effects coincide we do not attempt to distinguish between these standards. Instead we test their joint effect. Our hypothesis is as follows:

H2

The propensity of a firm to use tangible net-worth instead of net-worth requirements in its lending agreements is unchanged by the promulgation of SFAS 141 and SFAS 142.

2.4 Covenant slack in tangible net-worth covenants and net-worth covenants

We define covenant slack as the difference between a firm’s reported net worth (or tangible net worth) at the end of the fiscal year before the loan’s origination and the threshold level of net worth (or tangible net worth) that the firm must maintain to comply with the covenant. Lenders and borrowers can use slack to control contracting costs.

The choice of slack is a function of two costs. First, increasing slack reduces the covenant’s ability to serve as a warning device. More slack allows the borrower to transfer more wealth from the lender before covenant violation. If goodwill is less reliable and verifiable than other assets or if a net-worth covenant can be relaxed by future purchase-method acquisitions, then the marginal agency cost for one unit of slack is higher for a net-worth covenant than for a tangible net-worth covenant.

The second cost affected by slack is renegotiation cost. Renegotiation costs are a function of the probability of a covenant violation. A covenant with less slack is more likely to be violated and, thus, is associated with higher expected renegotiation costs. In Sect. 2.2, we argue that the probability of covenant violation can be reduced by using a tangible net-worth requirement. This argument suggests that, other things equal, tangible net-worth covenants can achieve a given level of renegotiation costs with less slack and therefore will be associated with less slack than net-worth covenants. In other words, for a given level of slack, net-worth covenants are more likely to be violated and therefore have higher expected renegotiation costs.

In summary, incremental renegotiation and moral hazard costs will likely have differing effects on the preference for slack in tangible net-worth covenants compared with net-worth covenants. Agency-cost considerations argue that tangible net-worth covenants should have more slack, while renegotiation-cost considerations suggest that net-worth covenants should have more slack. Whether we observe more slack in tangible net-worth covenants depends on the relative importance of these considerations. We note that the choices of covenant type and amount of slack are likely simultaneous. Therefore, we use instruments to proxy for the type of net-worth covenant. Our hypothesis stated in null form is as follows:

H3

The slack found in net-worth covenants is similar to the slack observed in tangible net-worth covenants.

3 Method

We test H1 and H2 by estimating the following logistic model: 14
$$ \begin{aligned} TNW=&\alpha +\beta _1 \ast GW+\beta _2 \ast GW\_ZERO+\beta _3 \ast OTH\_INTANG\\ &+ \beta _4 \ast RATING\_EXIST+\beta _5 \ast RATING+\beta _6 \ast ACQ+\beta _7 \ast MATURE\\ &+ \beta _8 \ast SYNDICATE+\beta _9 \ast POST+\beta _{10} \ast EBITDA\_COV\\ &+ \beta _{11} \ast E&B\_COV+\beta _{12} \ast OTHER\_COV+\beta _{13} \ast YIELD\\ &+ \beta _{14} \ast LNSIZE+\beta _{15} \ast BM+\beta _{16} \ast LEV+\beta _{17} \ast ROA+\beta _{18} \ast LOSS \\ &+ \beta _{19} \ast UNREAL\_GAIN, \\ \end{aligned} $$
(1)
where, TNW is an indicator variable equal to one if a debt facility contains a tangible net-worth covenant and zero if the facility contains a net-worth covenant. Tangible net-worth covenants are identified using the coding provided in the DealScan database. Facility-specific data items are measured in the year the facility is originated. Firm-specific data items are measured in the fiscal year before the origination of the facility.

Our main interest lies in the sign of the coefficient on GW, β1, which tests H1. GW is equal to goodwill as a percentage of total assets. 15 We expect the magnitude of other intangible assets (OTH_INTANG) to be associated with a tangible net-worth requirement in the same way as goodwill. Tangible net-worth definitions found in covenants occasionally add back unrealized gains on securities. Therefore, we include unrealized gains as a control variable (UNREAL_GAIN). We include several other variables to shed light on the costs and benefits associated with the choice between a tangible net-worth and a net-worth covenant. The value of acquisitions in the two-digit SIC code group of the borrower relative to the market capitalization of firms in that group in the year before the lending agreement is used as a proxy for expected acquisitions at the loan date (ACQ). We expect the anticipated level of future acquisitions to be associated with the existence of a tangible net-worth covenant based on the argument that future acquisitions add slack to net-worth covenants while not necessarily increasing the lender’s expected payout. The costs of implementing a tangible net-worth covenant are positively associated with the size of the syndicate (El-Gazzar & Pastena, 1990). Thus, we expect a negative association between the size of the syndicate (SYNDICATE) and the existence of a tangible net-worth covenant. To control for self selection in the use of net-worth-based covenants, we also check the robustness of our results to the use of a Heckman-two-stage procedure (Table 6).

In H2 we investigate whether the adoption of SFAS 141 and 142 is associated with the likelihood of including a tangible net-worth covenant in lending agreements. To test this hypothesis, we include in Eq. 1 an indicator variable, POST, equal to one for loans originated in 2002 or later. If SFAS 141 and 142 have reduced the usefulness of goodwill in lending agreements either by increasing renegotiation costs or increasing agency costs, we expect tangible net-worth covenants will be chosen more often and the coefficient on POST to be positive.

We also include as controls the yield spread of the loan (YIELD) and indicator variables for the existence of other covenants. Both yield spread and other covenants can serve as substitutes or complements to the use of tangible net-worth covenants. We group the other covenants into three types: EBITDA accounting covenants (EBITDA_COV), earnings and balance sheet-based accounting covenants (E&B_COV), and other covenants (OTHER_COV). All these covenants can act as correlated-omitted variables if they are substitutes or complements for tangible net-worth covenants. For a complete description of the covenants groups, refer to Table 5.

Other variables such as size, book-to-market, leverage, debt ratings, return on assets, and maturity are included because they are related to the agency costs of debt and thus can be related to the preference for tangible net-worth covenants. The agency costs of debt are hypothesized to increase in firms’ growth prospects because managers have more opportunities to increase the variance of their assets and thus transfer wealth from debt-holders to shareholders (Smith & Watts, 1992; Billett, King, & Meyer, 2007). Thus, we expect a negative correlation between agency costs and the book-to-market ratio. Because the use of tangible net-worth covenants can be efficient when agency costs are high, book-to-market could act as a correlated omitted variable in our regressions. Similar arguments can be made for measures of the firm’s financial health. The coefficient on loan characteristic variables such as SYNDICATE and MATURE should be interpreted with caution because these variables are likely chosen simultaneously with the covenants.

To address H3 and investigate the association between the degree of slack allowed in minimum worth covenants and the type of covenants used, we estimate the following model:
$$ \begin{aligned} SLACK=&\alpha +\beta _1 \ast IND\_TNW+\beta _2 \ast GW+\beta _3 \ast GW\_ZERO \\ & + \beta _4 \ast OTH\_INTANG+\beta _5 \ast RATING\_EXIST+\beta _6 \ast RATING \\ &+ \beta _7 \ast ACQ+\beta _8 \ast POST+\beta _9 \ast IND\_MATURE \\ & + \beta _{10} \ast IND\_SYNDICATE+\beta _{11} \ast LNSIZE+\beta _{12} \ast BM \\ &+ \beta _{13} \ast LEV+\beta _{14} \ast ROA+\beta _{15} \ast LOSS+\beta _{16} \ast UNREAL\_GAIN, \\ \end{aligned} $$
(2)
where SLACK is computed as the actual level of net worth (or tangible net worth) minus the level of net worth (or tangible net worth) stated in the covenant, all divided by total assets.

4 Data and results

4.1 Data

We obtain our sample from the Dealscan database available from the Loan Pricing Corporation between 1992 and 2003. 16 Financial statement data is obtained from Compustat. Acquisition data is obtained from the Securities Data Corporation database. Table 1 outlines our sample selection filters. We begin with 28,448 debt facilities that are available in the Dealscan database. We exclude 3,468 facilities that could not be matched to Compustat as well as 4,480 facilities that do not have sufficient information to compute the variables used in our tests. Next, we exclude the facilities of firms with negative book values. 17 Finally, we exclude all facilities that do not contain a net-worth or a tangible net-worth covenant. Our final sample consists of 4,096 facilities obtained for 1,662 unique firms.
Table 1

Sample selection. The sample of lending agreements is drawn from the Dealscan database provided by the Loan Pricing Corporation. The sample includes facilities from 1992–2003

Number of facilities on Dealscan

 

28,448

  Less facilities that cannot be matched to Compustat

3,468

 
  

24,980

  Less facilities that do not have Compustat data for variables used in subsequent tests

4,480

 
  

20,500

  Less facilities of firms with book values less than zero

781

 
  

19,719

  Less facilities that do not have net-worth-type covenants

15,623

 

Final Sample

 

4,096

4.2 Summary statistics

Table 2 reports summary statistics for our sample. 18 Our sample contains slightly more facilities with tangible net-worth covenants (52.6 percent) than with net-worth covenants (47.4 percent). The market capitalization (SIZE) of our sample firms averages $1.5 billion but is skewed with a median of $250 million. Goodwill (GW) constitutes about 6 percent of total assets, on average, but more than 50 percent of the sample firms have no goodwill on their balance sheet. Approximately 19 percent of sample firms reported losses in the year before originating loan facilities (LOSS). Approximately 27 percent of the observations in our sample have a non missing bond rating (RATING) in Compustat, and of those, 41.5 percent have an investment-grade bond rating. The median term (MATURE) of the loans in our sample is about 36 months (natural log equal to 3.58) and the median number of lenders is approximately three.
Table 2

Descriptive statistics

Variable (N = 4,096)

Mean

Median

Std Dev

Min.

Max.

TNW

0.526

1

0.49

0

1

SIZE

1,537.9

250.3

6,936.2

0.268

139,092.7

BM

0.82

0.52

8.06

0.001

514.9

ASSETS

2,310.5

305.7

18,086.9

1.26

668,641

TOTAL EQUITY

535.6

127.8

1,937.8

0.283

40,395

ROA

0.02

0.04

0.15

−3.49

0.37

LEV

1.74

0.66

30.17

0.00

1,925

GW

0.06

0

0.11

0

0.71

OTH_INTANG

0.005

0

0.03

0

0.57

LOSS

0.19

0

0.39

0

1

RATING (N = 1,111)

0.415

0

0.492

0

1

MATURE

3.48

3.58

0.71

0

5.48

SYNDICATE

1.23

1.09

1.12

0

4.61

SLACK (N = 3,846)

−0.369

0.062

7.749

−253.5

0.826

SLACK*(N = 3,012)

0.125

0.085

0.125

0.000

0.826

Variable Definitions:

N is the number of observations

TNW is equal to one if the firm has a tangible net-worth covenant and zero if the firm has a net-worth covenant

SIZE is the market value of equity defined as stock price at the end of the fiscal year multiplied by the number of shares (data#199*data#25)

BM is the book-to-market ratio (data#60/data#199*data#25)

ASSETS is the total assets of the firm in millions of dollars (data#6)

TOTAL EQUITY is shareholders’ equity in millions of dollars (data#60)

ROA is net income deflated by total assets (data#18/data#6)

LEV is the leverage of the firm measured as total liabilities (data#181) divided by market value of equity

GW is the total amount of goodwill on the balance sheet (data#204) as a percentage of total assets

OTH_INTANG is the amount of intangibles other than goodwill (data#352) as a percentage of total assets

LOSS is a dummy variable equal to one if net income for the year was less than or equal to zero and zero otherwise

RATING is a dummy variable equal to one if the firm’s bond rating is investment grade or above and zero otherwise (data#280). The variable is only reported for 1,111 facilities of firms that have a valid rating (i.e. they have publicly rated debt)

MATURE is the log of the maturity of the loan in months.

SYNDICATE is the log of the number of lenders involved in the loan

SLACK is equal to the difference between the actual level of net worth (or tangible net worth) in the fiscal year before the origination of the credit facility and the net worth (or tangible net worth) threshold stated in the credit facility, all divided by total assets. The data is available for 3,846 observations

SLACK* is computed only for observations for which SLACK > 0 (N = 3,012)

Facility-specific data items are measured in the year the facility originated. Firm-specific data items are measured in the fiscal year before the origination of the facility and are obtained from Compustat. Facility-level data is obtained from Dealscan between 1992 and 2003

Table 3 reports univariate correlations between several variables of interest. These estimates indicate that the incidence of a tangible net-worth covenant is negatively associated with the magnitude of goodwill and other intangible assets. Consistent with El-Gazzar and Pastena (1990), we find that tangible net-worth covenants are less likely to be used as the number of lenders increases (SYNDICATE). This result suggests that tangible net-worth covenants are associated with higher implementation costs. The negative correlation between the existence of a tangible net-worth covenant and RATING also indicates that tangible net-worth covenants are more likely to be used as default risk increases. A tangible net-worth covenant is more likely to be chosen over a net-worth covenant when ROA is low. These results are consistent with the argument that tangible net-worth covenants are more likely to be used when an increased risk of default focuses a lender’s attention on the collateral value of assets and magnifies agency concerns.
Table 3

Correlations

 

TNW

SIZE

BM

ROA

LEV

GW

OTH_INTANG

LOSS

RATING

MATURITY

SYNDICATE

SLACK

TNW

 

−0.06*

0.02

−0.06*

0.01

−0.27*

−0.08*

0.06*

−0.10*

−0.15*

−0.29*

0.02

SIZE

−0.21*

 

−0.01

0.03*

−0.00

−0.01

0.01

−0.03*

−0.28*

−0.05*

0.23*

0.01

BM

0.12*

−0.45*

 

−0.01

−0.04

0.00

−0.00

0.06*

−0.18*

−0.02

−0.02

0.01

ROA

−0.03*

0.26*

−0.34*

 

−0.01

0.01

−0.20*

−0.53*

0.20*

0.13*

0.12*

0.30*

LEV

−0.04*

−0.16*

0.56*

−0.47*

 

0.01

−0.00

0.05*

−0.13*

−0.02

−0.02

−0.04*

GW

−0.22*

0.06*

−0.06*

0.03*

0.00

 

0.01

−0.03*

−0.08*

0.09*

0.15*

−0.01

OTH_INTANG

−0.08*

0.12*

−0.01

−0.05*

0.05*

0.09*

 

0.05*

−0.00

−0.02

0.05*

−0.21*

LOSS

0.06*

−0.29*

0.17*

−0.68*

0.16*

−0.04*

0.03*

 

−0.18*

−0.13*

−0.20*

−0.06*

RATING

−0.10*

0.55*

−0.14*

0.20*

−0.21*

−0.04

−0.04

−0.18*

 

−0.26*

0.25*

0.10*

MATURE

−0.18*

0.08*

−0.03*

0.10*

0.02

0.09*

−0.07*

−0.15*

−0.19*

 

0.21*

0.02

SYNDICATE

−0.29*

0.62*

−0.13*

0.09*

0.15*

0.15*

0.09*

−0.20*

0.29*

0.24*

 

−0.04*

SLACK

0.14*

0.07*

0.04*

0.12*

−0.34*

−0.08*

0.02

0.01

0.20*

−0.06*

−0.13*

 

Pearson correlations are displayed above the diagonal and Spearman correlations are displayed below the diagonal. The number of observations for all variables, except those related to SLACK and RATING, is 4,096. RATING is based on 1,111 observations, and SLACK is based on 3,846 observations. * indicates significance at at least the 0.05 level. Variables are as defined in Table 2

The first three years in our sample period (1992–1994) contain very few observations (about 1.3 percent), consistent with the fact that Dealscan’s coverage during those years is less complete (see page 1099 in Dichev and Skinner, 2002). Beginning in 1995, the number of observations is fairly evenly distributed across years. The final year, 2003, contains slightly fewer observations. Also, there is no pronounced time-trend in the relative use of tangible net-worth versus net-worth covenants. 19

Table 4 reports the frequency of debt facilities across 14 industry groups. We combine several SIC codes into broad industry groups based on the economic similarity of firms. We use these industry categories as control variables in our regression analysis. 20 As can be seen in Table 4, we find considerable cross-sectional variation in the relative use of tangible net-worth covenants across industries. For example, 82.8 percent of loan facilities associated with the mining & construction industry use tangible net-worth requirements, compared with only 25.8 percent of facilities associated with the chemicals industry. This variation suggests that our industry classification captures underlying variation in factors related to the use of tangible net worth covenants. However, multivariate tests are necessary to assess whether this explanatory power is incremental to other firm-specific controls.
Table 4

Frequencies of Net-Worth and Tangible Net-Worth Covenants by Industry. The table reports the frequencies of net-worth and tangible net-worth covenants in our sample based industry groups. The SIC codes of each group are also reported

Industry group

SIC codes included

% of sample

Net worth

Tangible net worth

N

Pct.

N

Pct.

Agriculture

0–999

0.3

6

50.0

6

50.0

Mining & construction

1000–1299

3.8

27

17.2

130

82.8

1400–1999

     

Food

2000–2111

3.1

59

46.8

67

53.2

Textiles & printing/publishing

2200–2790

6.3

149

58.2

107

41.8

Chemicals

2800–2824

2.3

69

74.2

24

25.8

2840–2899

     

Pharmaceuticals

2830–2836

1.8

26

36.1

46

63.9

Extractive

2900–2999

4.8

50

25.3

148

74.7

1300–1399

     

Durable manufacturers

3000–3569

22.8

430

46.1

502

53.9

3580–3669

     

Transportation

3680–3999

10.5

135

31.3

297

68.8

Utilities

7370–7379

3.4

75

54.0

64

46.0

3570–3579

     

3670–3679

     

Retail

4900–4999

1.3

32

59.3

22

40.7

Services

5000–5999

13.1

256

47.9

279

52.1

Computers

7000–7369

11.6

308

64.6

169

35.4

7380–9999

     

Financial Services

6000–6999

15.0

316

51.5

297

48.5

4.3 Tests of hypotheses one and two

The results of our investigation of the relation between the level of goodwill and the probability of observing a tangible net-worth covenant are shown in Table 5, where we estimate several versions of Eq. 1. Recall that this relation should be positive if lenders are mainly interested in the collateral value of the assets, if goodwill provides little indication of the lenders’ expected payoff, and/or if the incremental implementation costs of a tangible net-worth covenant are minimal. This relation is expected to be negative if goodwill is correlated with the value of debt and thus write-offs are expected to be meaningful events to the lenders. In Table 5 we report clustered standard errors (by firm) to mitigate independence problems from multiple observations per firm (see Petersen, 2005).
Table 5

Logistic estimation of the determinants of the existence of a tangible net-worth covenant. The table reports the results of four logistic models whose dependent variable is an indicator variable equal to one if a debt facility contains a minimum tangible net-worth requirement and equal to zero if the facility contains a minimum net-worth requirement. The indicator variable is determined based on the classification appearing in the Dealscan database. Facility-level data is obtained from Dealscan between 1992 and 2003. The models also include industry controls, which are not reported in the tables. Industry descriptions are reported in Table 4. Firm-clustered standard errors are reported (see Petersen, 2005). *, ** and *** indicate significance at the 10%, 5% and 1% level, respectively

Dependent variable = TNW

Model 1

Model 2

Model 3

Model 4

 

Pred. sign

Coeff.

Robust Std. Err

Coeff.

Robust Std. Err

Coeff.

Robust Std. Err

Coeff.

Robust Std. Err

INTERCEPT

 

1.832***

0.530

2.381**

0.586

2.277***

0.566

4.198***

0.691

GW

− / + 

−6.624***

−0.758

−6.501***

0.762

−6.550***

0.770

−6.565***

0.840

GW_ZERO

− / + 

−0.302**

0.146

−0.364**

0.148

−0.343**

0.148

−0.332**

0.153

OTH_INTANG

− / + 

−6.288***

1.627

−5.824***

1.563

−7.154***

1.729

−7.520***

1.877

RATING_EXIST

 

−0.209

0.174

0.008

0.175

−0.030

0.175

0.109

0.177

RATING*

 

0.077

0.239

−0.068

0.247

−0.031

0.250

−0.230

0.245

ACQ

 + 

0.377*

0.224

0.441*

0.232

0.496**

0.239

0.572**

0.240

MATURE

   

−0.297***

0.070

−0.285***

0.070

−0.272***

0.073

SYNDICATE

  

−0.395***

0.065

−0.399***

0.065

−0.375***

0.068

POST

+

    

0.331**

0.135

0.513***

0.141

EBITDA_COV

       

−0.436***

0.160

E&B_COV

       

−0.409

0.256

OTHER_COV

       

−0.396***

0.134

YIELD

       

−0.002**

0.001

LNSIZE

 

−0.200***

0.042

−0.054

0.047

−0.057

0.047

−0.126**

0.052

BM

 

0.223*

0.125

0.246**

0.123

0.229*

0.122

0.205*

0.118

LEV

 

−0.058*

0.033

−0.064**

0.033

−0.060*

0.032

−0.053*

0.031

ROA

 

−0.725

0.487

−0.822

0.511

−0.801

0.524

−0.747

0.516

LOSS

 

−0.185

0.172

−0.292*

0.175

−0.323*

0.177

−0.213

0.186

UNREALIZED_GAIN

 

1.225

5.894

0.508

5.814

0.407

5.747

0.318

5.252

14 industry controls based on 4-digit SIC codes

included

included

Included

Included

N

4,096

3,993

3,993

3,661

Pseudo-R-Squared

13.7%

16.3%

16.4%

18.0%

Variable Definitions (See Table 2 for additional variable definitions):

GW_ZERO is an indicator variable equal to one if goodwill is equal to zero. If goodwill does not equal to zero, then the variable equals zero

ACQ is the relative value of all acquisitions made in a firm’s 2-digit SIC code group in the year before the debt contract. The relative value of acquisitions is computed as the value of all acquisitions divided by the total market capitalization of all firms in that industry group

RATING_EXIST is an indicator variable equal to one if the debt rating (data#280) exists for that firm. The variable is equal to zero if the debt rating is missing

RATING* is a dummy variable equal to one if the firm’s bond rating is investment grade or above and zero otherwise (data#280). We equate the indicator variable to zero in cases where the debt rating variable is missing

POST is an indicator variable equal to one if the facility was originated in 2002 or after

LNSIZE is the natural logarithm of market value of equity defined as stock price at the end of fiscal year multiplied by the number of shares (data#199*data#25)

YIELD is the cost of the loan measured as the markup over LIBOR

UNREALIZED_GAIN is unrealized gains from marketable securities (data#238) divided by total assets

EBITDA_COV is an indicator variable equal to one if a debt contract contain at least one of the covenants from the group of EBITDA covenants in the table below

E&B_COV is an indicator variable equal to one if a debt contract contain at least one of the covenants from the group earnings or balance sheet-based covenants in the table below

OTHER_COV is an indicator variable equal to one if a debt contract contain at least one of the covenants from the group of “Other covenant” in the table below
 

 

EBITDA Covenants (EBITDA_COV)

Earnings and Balance Sheet Covenants (E&B_COV)

Other Covenants(OTHER_COV)

Fixed charge coverage

Current ratio

Asset sales sweep

Debt service coverage

Quick ratio

Debt issue sweep

Interest Coverage

Debt to tangible net-worth

Equity issue sweep

Cash interest coverage

Leverage ratio

Excess cash flow sweep

Debt to cash flow

Debt to equity

Maximum capital expenditures

Senior debt to cash flow

Dividend restrictions

 

Minimum EBITDA

  

In Model 1 of Table 5, we present results from a logistic estimation, which excludes variables that relate to the characteristics of the loan (MATURE and SYNDICATE) and the effective date of SFAS 141 and 142 (POST). We find that the level of goodwill in the fiscal year before the lending agreement (GW) is negatively associated with the probability of including a tangible net-worth covenant. This result is consistent with recorded goodwill being correlated with the lender’s expected payoff. The implied positive association between the preference for net-worth covenants and the magnitude of goodwill is consistent with lenders believing that goodwill levels are informative and an efficient means of limiting agency costs. Similarly, intangible assets other than goodwill (OTH_INTANG) are negatively associated with the incidence of a tangible net-worth covenant. The negative coefficient on GW_ZERO suggests a nonlinear relation between goodwill and tangible net-worth covenants. That is, for low levels of goodwill, net-worth covenants appear to be more likely. We emphasize that the negative coefficient on GW is not driven by outliers because a similarly negative coefficient appears when using the rank of goodwill or the existence of goodwill as alternative explanatory variables.

The variables associated with bond ratings (RATING* and RATING_EXIST) do not appear to be associated with the incidence of a tangible net-worth covenant. ACQ has a positive, marginally significant association with the likelihood of a tangible net-worth covenant. This result is consistent with the discussion in Sect. 2.3, which argues that future acquisitions loosen net-worth covenants, making them less effective than tangible net-worth covenants for controlling agency costs.

In Model 2, we add variables related to characteristics of the loan facility. Results discussed for Model 1 also hold for our variables of interest in Model 2. The number of lenders (SYNDICATE) is negatively associated with the incidence of tangible net-worth covenants, which is consistent with tangible net-worth covenants being more costly to negotiate and monitor. In addition, the length of the loan (MATURITY) is negatively associated with the incidence of a tangible net-worth covenant. This result holds for all specifications and is surprising given that agency costs would seem to be positively related to maturity (Billett et al., 2007). As maturity increases, the lender’s need for an early warning is increased. Alternatively, this result suggests that tangible net-worth covenants and maturity are complementary mechanisms for controlling agency costs. That is, high agency costs lead to reduced maturity and an increased use of tangible net-worth covenants. This explanation implies that maturity and covenant type are jointly determined. Given the potential for endogeneity in the relation between facility-specific characteristics (that is, maturity, syndication and other covenants) and the use of tangible net-worth covenants, these coefficients should be interpreted with caution. Facility-specific characteristics are included to assess the robustness of our estimates on experimental variables GW and POST.

In Model 3, we add an indicator variable, POST, which is equal to one after the effective date of SFAS 141 and 142 to examine whether these goodwill standards affect debt contracts. If these standards reduce the usefulness of goodwill in contracts, we should observe an increase in the likelihood of including a tangible net-worth covenant in debt contracts originated after the adoption of SFAS 141 and 142 and the coefficient on POST will be positive. The results from Model 3 are consistent with this interpretation. We find that the indicator variable, POST, is positive and significant. That is, the incidence of tangible net-worth covenants has increased in the years following the adoption of SFAS 141 and 142, controlling for other economic determinants of the covenants. 21

In Model 4 we add three indicator variables to control for other covenants that are included in the debt contracts. The first indicator variable equals one if the debt contract contains EBITDA-based accounting covenants (EBITDA_COV). The second indicator variable equals one if the debt contract contains earnings or balance sheet-based accounting covenants (E&B_COV). The third indicator variable equals one if the debt contract contains other types of covenants (OTHER_COV). (For a full list of covenants, refer to Table 5.) We also add the variable YIELD, which measures the cost of the loan over LIBOR. We include these variables to investigate whether our main result is sensitive to controlling for other terms of the loan including pricing and other covenants. The pricing of the loan and inclusion of other covenants are jointly determined with other features of the contract, including the existence of a tangible net-worth covenant. Therefore, we are reluctant to interpret these coefficients. After including these variables, the GW variable continues to be negatively associated with the incidence of tangible net-worth covenants.

The sample included in Models 1–4 is restricted to facilities that contain minimum worth covenants. Coefficient estimates will be biased if sample selection is associated with the independent variable of interest because such an association implies that the error term is correlated with the independent variable. To address this concern we estimate a two-stage Heckman procedure and report its results in Table 6. In the first stage, we model the incidence of a minimum worth covenant as a function of several firm characteristics. We estimate the first stage on all firms with lending agreements available in the Dealscan database and with sufficient information to compute all variables in the first stage (14,148 observations). The pseudo-R2 from the first stage estimation is 6.7 percent, suggesting that we have difficulty controlling for the sample-selection bias. However, the coefficient on GW is insignificant suggesting that the magnitude of goodwill is not related to the selection of minimum worth requirements.
Table 6

Two-stage logistic estimation of the determinants of the existence of a tangible net-worth covenant. The table reports the results of a two-step Heckman selection procedure. In the first stage, we estimate the relation between an indicator variable that is equal to one if a minimum worth requirement exists and a set of independent variables chosen to explain cross-sectional variation in use of minimum net-worth requirements. The second stage’s dependent variable is the one used in models 1–4 in Table 5. Firm-clustered standard errors are reported. *, ** and *** indicate significance at the 10%, 5% and 1% level respectively. Facility-level data is obtained from Dealscan between 1992 and 2003

Model 5

Model 5

1ststage of a Heckman Selection model

2ndstage of a Heckman Selection model

Dependent variable = Net worth

Dependent variable = Tangible net worth

 

Coeff.

Robust Std. Err

 

Coeff.

Robust Std. Err

INTERCEPT

0.118

0.227

INTERCEPT

1.101

1.111

GW

0.119

0.097

GW

−1.054***

0.178

LNSIZE

−0.143***

0.016

GW_ZERO

−0.047

0.034

BM

0.042**

0.018

OTH_INTANG

−1.211***

0.085

LEV

−0.058***

0.012

RATING_EXIST

0.019

0.019

ROA

0.191

0.133

RATING*

−0.012

0.064

LOSS

−0.069

0.053

ACQ

0.079

0.107

UNREALIZED_GAIN

1.359

1.257

MATURE

−0.062***

0.014

VOLATLITY

0.027

0.138

SYNDICATE

−0.092***

0.022

PPE

−0.007

0.043

POST

0.057***

0.018

INT_COVERAGE

0.002

0.006

LNSIZE

−0.029

0.101

   

BM

0.028

0.048

   

LEV

−0.012

0.049

   

ROA

−0.094

0.171

   

LOSS

−0.050

0.056

   

UNREALIZED_GAIN

0.366

1.213

   

Inverse Mills Ratio

0.118

1.030

14 industry controls based on 4-digit SIC codes

included

 

included

N

14,148

 

3,993

Pseudo-R-Squared

6.7%

 

16.5%

Variable Definitions (See Table 2 for additional variable definitions):

GW_ZERO is an indicator variable equal to one if goodwill is equal to zero. If goodwill does not equal to zero, then the variable equals zero

ACQ is the relative value of all acquisitions made in a firm’s 2-digit SIC code group in the year before the debt contract. The relative value of acquisitions is computed as the value of all acquisitions divided by the total market capitalization of all firms in that industry group

RATING_EXIST is an indicator variable equal to one if the debt rating (data#280) exists for that firm. The variable is equal to zero if the debt rating is missing

RATING* is a dummy variable equal to one if the firm’s bond rating is investment grade or above and zero otherwise (data#280). We equate the indicator variable to zero in cases where the debt rating variable is missing

POST is an indicator variable equal to one if the facility was originated in 2002 or after

LNSIZE is the natural logarithm of market value of equity defined as stock price at the end of the fiscal year multiplied by the number of shares (data#199*data#25)

UNREALIZED_GAIN is unrealized gains from marketable securities (data#238) divided by total assets

VOLATILITY is the annual stock return volatility calculated based on daily returns in the year before the debt contract

PPE is property, plant and equipment (data#8) deflated by total assets

INT_COVERAGE is the interest coverage ratio defined as interest expense (data#15) divided by income before interest, taxes and depreciation and amortization

In the second stage, we estimate Model 3 including the inverse-Mills ratio from the first stage. The coefficient on the inverse-Mills ratio is insignificant. The negative (positive) relation between goodwill (POST) and the incidence of a tangible net-worth covenant is robust to the Heckman procedure.

4.4 An alternative sample

To investigate the sensitivity of our results to the use of Dealscan data, we employ an alternative sample. To construct the alternative sample, we search 10-K filings for firms whose debt contracts included net-worth and tangible net-worth covenants during the years 1998–2002. Our search included the following keyword searches:
  1. 1.

    The tangible net-worth sample (628 observations): Maintain (near5) “tangible net worth”

     
  2. 2.

    The net-worth sample (618 observations): Maintain (near5) “net worth” (not) “tangible net worth”

     

Our final alternative sample, with sufficient information to run our regression models, contains 863 observations—422 with net-worth covenants and 441 with tangible net-worth covenants.

4.4.1 Overlap between the two samples

The final Dealscan sample contains 4,096 facilities with sufficient information to estimate the regression models. Of those, 169 facilities (of 118 different firm-years) are shared with the alternate sample. Thus, of the 843 observations in the alternate sample, about 20 percent are shared with the Dealscan sample. For the 169 observations that are common between the two samples, we find agreement between the two samples with regard to the classification of net-worth covenant type in 143 cases. These results suggest an estimated upper bound for the potential error rate in the Dealscan data set is approximately 15 percent.

We inspect the 26 cases where Dealscan’s and the alternate sample’s net-worth-covenant classifications disagree. We find that in 18 cases, the classification of our alternate sample was correct, in 2 cases the classification in Dealscan was correct, and in 6 cases we could not determine which source is correct.

In unreported results, we find that the firms in the alternate sample are smaller (in terms of total assets and sales). We find no significant differences in the levels of goodwill or leverage between the firms in the two samples. Thus we conclude that Dealscan is likely to include larger firms in its database.

4.4.2 Results using the alternative sample

In Table 7 we report the results of estimating our basic model 1 using the alternate sample. We cannot estimate more complete versions of the models using the alternate sample because we do not have information on maturity, the number of lenders, and other covenants for this sample. For comparison purposes, we also report the results from the Dealscan sample (previously reported in Table 5). While the alternate sample has significantly fewer observations, the coefficient on GW remains negative and significant. Interestingly, we find that firms for which bond ratings exist on Compustat (presumably firms with public debt) are less likely to include a tangible net-worth covenant in their contracts. Also, we find that given the existence of a debt rating, firms with higher ratings are more likely to be associated with a tangible net-worth covenant.
Table 7

Logistic Estimation of the Determinants of the Existence a Tangible Net Worth Covenant (Robustness tests). The table reports the results of logistic models whose dependent variable is an indicator variable equal to one if a debt facility contains a minimum tangible net-worth requirement and equal to zero if the facility contains a minimum net-worth requirement. The indicator variable is determined based on the classification appearing in the Dealscan database. The first model is the base model that is reported in Table 5. The second model reported in the table is estimated using an alternative sample that is collected using a keyword search of 10-K’s, which identified contracts that contained net-worth and tangible net-worth in the years 1998–2002. The third model is similar to model 3 in Table 5 and is estimated using one random observation for every firm in the sample. Its standard errors, therefore, are not clustered, controlling for firm-effects. The models also include industry controls, which are not reported in the tables. Firm-clustered standard errors are reported for the first and second model. *, ** and *** indicate significance at the 10%, 5% and 1% level, respectively. Facility-level data is obtained from Dealscan between 1992 and 2003

Dependent variable = TNW

Model 1

Model 1

Model 3

 

Pred. Sign

Deal scan sample

Alternate sample

Random observations

 

 

Coeff.

Robust Std. Err

Coeff.

Robust Std. Err

Coeff.

Std. Err

INTERCEPT

 

1.832***

0.230

2.379

1.635

2.109***

0.776

GW

− / + 

−6.624***

0.758

−5.283***

1.147

−5.937***

0.850

GW_ZERO

− / + 

−0.302**

0.146

−0.390**

0.212

−0.285*

0.159

OTH_INTANG

− / + 

−6.288***

1.627

0.542

2.204

−5.920***

2.363

RATING_EXIST

 

−0.209

0.174

−0.812***

0.252

0.071

0.189

RATING*

 

0.077

0.239

0.608*

0.362

−0.133

0.250

ACQ

+

0.377*

0.224

0.211

0.333

0.561**

0.240

MATURE

     

−0.496***

0.093

SYNDICATE

    

−0.509***

0.077

POST

+

    

0.164

0.167

LNSIZE

 

−0.200***

0.042

−0.195***

0.049

−0.003

0.049

BM

 

0.223*

0.125

0.051

0.036

0.171

0.106

LEV

 

−0.058*

0.033

−0.025

0.017

−0.044

0.028

ROA

 

−0.725

0.487

−0.471

0.393

−0.888*

0.529

LOSS

 

−0.185

0.172

0.006

0.207

−0.423**

0.185

UNREALIZED_GAIN

 

1.225

5.894

−1.455

3.096

−4.626

4.467

14 industry controls based on 4-digit SIC codes

included

included

included

N

4,096

863

1,511

Pseudo-R-Squared

13.7%

13.2%

17.0%

Variable Definitions (See Table 2 for additional variable definitions):

GW_ZERO is an indicator variable equal to one if goodwill is equal to zero. If goodwill does not equal to zero, then the variable equals zero

ACQ is the relative value of all acquisitions made in a firm’s 2-digit SIC code group in the year before the debt contract. The relative value of acquisitions is computed as the value of all acquisitions divided by the total market capitalization of all firms in that industry group

RATING_EXIST is an indicator variable equal to one if the debt rating (data#280) exists for that firm. The variable is equal to zero if the debt rating is missing

RATING* is a dummy variable equal to one if the firm’s bond rating is investment grade or above and zero otherwise (data#280). We equate the indicator variable to zero in cases where the debt rating variable is missing

POST is an indicator variable equal to one if the facility was originated in 2002 or after

LNSIZE is the natural logarithm of market value of equity defined as stock price at the end of the fiscal year multiplied by the number of shares (data#199*data#25)

UNREALIZED_GAIN is unrealized gains from marketable securities (data#238) divided by total assets

Finally, the last columns in Table 7 report the results of an observation-independence robustness test in which we estimate model 3 from Table 5 limiting each firm in the sample to only one randomly chosen debt facility. The number of observations in this specification falls to 1,511. The results for H1 regarding goodwill do not change as a result of this estimation. However, the coefficient on POST is no longer significant.

4.5 Test of hypothesis three

The results of our investigation of the association between the degree of slack allowed in minimum worth covenants and the type of covenants used are reported in Table 8. We estimate two versions of equation (2). The dependent variable, SLACK, is computed in two ways: (i) in Model 1 slack is computed as the actual level of net worth (or tangible net worth) minus the level of net worth (or tangible net worth) stated in the covenant, all divided by total assets, and (ii) in Model 2 slack is computed as the natural logarithm of the actual level of net worth (or tangible net worth) divided by the level of net worth (or tangible net worth) stated in the covenant. In both models, we exclude observations where SLACK is negative because these observations have a greater likelihood of containing data errors. 22 Because slack, the type of net-worth covenant, and other debt facility features (for example, the number of lenders and maturity) are likely to be jointly determined, we use instruments to isolate the portion of the independent variables that is pre-determined. In particular, we define IND_TNW, IND_SYNDICATE and IND_MATURE as the averages of all other facilities that originated in the same year in the firm’s industry-group. 23 In both models, we find evidence that slack is increasing in situations where agency costs are reduced. The coefficient on RATING is positive indicating that higher rated debt is associated with more slack. In addition, the facilities of larger firms are associated with more slack. However, we do not find evidence supporting the hypothesis that the level of slack chosen in a debt contract is related to the type of minimum worth covenant. In Model 1, the coefficient on IND_TNW is not significant. This result is robust to the alternative definition of slack used in Model 2.
Table 8

Regression Estimates of the Variables Associated with Covenant Slack. The table reports the results from estimation of two OLS regression models whose dependent variable is either (i) SLACK, which is equal to the difference between the actual level of net worth (or tangible net worth) in the fiscal year before the origination of the credit facility and the net worth (or tangible net worth) threshold stated in the credit facility all divided by total assets or (ii) Log(SLACK_RATIO), which is equal to the natural logarithm of the actual level of net worth (or tangible net worth) in the fiscal year before the origination of the credit facility divided by the net-worth (or tangible net-worth) threshold stated in the credit facility all divided by total assets. Facility-level data is obtained from Dealscan between 1992 and 2003. Actual net-worth is defined to be consistent with the net-worth benchmark in the covenant. The models are estimated for all facilities for which SLACK is positive. The models include industry controls, which are not reported in the tables. Firm-clustered standard errors are reported. *, ** and *** indicate significance at the 10%, 5% and 1% level, respectively.

 

Model 1

Model 2

Dependent variable = 

SLACK

Log(SLACK_RATIO)

 

Coefficient

Robust Std. Errors

Coefficient

Robust Std. Errors

INTERCEPT

0.174**

0.788

0.545

0.459

IND_TNW

−0.027

0.025

−0.100

0.174

GW

−0.025

0.032

−0.276

0.186

GWZERO

0.031***

0.008

0.049

0.052

OTH_INTANG

−0.010

0.085

−0.689**

0.270

RATING_EXIST

−0.042***

0.008

−0.004

0.064

RATING*

0.033***

0.011

0.222**

0.089

ACQ

0.007

0.009

−0.133*

0.079

POST

0.022**

0.011

0.004

0.067

IND_MATURE

−0.024

0.019

−0.144

0.112

IND_SYNDICATE

−0.026

0.016

−0.030

0.112

LNSIZE

0.012***

0.003

0.045***

0.012

BM

0.035***

0.007

−0.029

0.040

LEV

−0.009***

0.002

0.007

0.010

ROA

−0.062

0.041

−0.231

0.189

LOSS

0.008

0.011

0.071

0.055

UNREALIZED_GAIN

0.068

0.189

−0.171

0.759

14 industry controls based on 4-digit SIC codes

Included

Included

N

2,973

2,961

Adjusted-R-Squared

11.6%

4.2%

Variable Definitions (See Table 2 for additional variable definitions):

SLACK_RATIO is the ratio of the actual level of net worth (or tangible net worth) in the fiscal year before the origination of the credit facility divided by the net worth (or tangible net worth) threshold stated in the credit facility

IND_TNW is equal to the proportion of facilities taken by firms in the same industry group as the examined firm (excluding the examined firm) that included a tangible net-worth covenant during the same year as the examined facility

GW_ZERO is an indicator variable equal to one if goodwill is equal to zero. If goodwill does not equal to zero, then the variable equals zero

ACQ is the relative value of all acquisitions made in a firm’s 2-digit SIC code group in the year before the debt contract. The relative value of acquisitions is computed as the value of all acquisitions divided by the total market capitalization of all firms in that industry group

RATING_EXIST is an indicator variable equal to one if the debt rating (data#280) exists for that firm. The variable is equal to zero if the debt rating is missing

RATING* is a dummy variable equal to one if the firm’s bond rating is investment grade or above and zero otherwise (data#280). We equate the indicator variable to zero in cases where the debt rating variable is missing

POST is an indicator variable equal to one if the facility was originated in 2002 or after

IND_MATURE is the average log of the maturity of loans in months taken by firms in the same industry-group as the examined firm (excluding the examined firm) during the same year as the examined facility

IND_SYNDICATE is the average log of the number of lenders involved in loans taken by firms in the same industry group as the examined firm (excluding the examined firm) during the same year as the examined facility

LNSIZE is the natural logarithm of market value of equity defined as stock price at the end of the fiscal year multiplied by the number of shares (data#199*data#25)

UNREALIZED_GAIN is unrealized gains from marketable securities (data#238) divided by total assets

5 Conclusion

We study how the magnitude of goodwill and how changes in goodwill accounting rules relate to the choice of minimum net-worth covenants. We take an “ex ante” view of covenants in that we treat covenants as a choice that is shaped by accounting procedures and regulations. We also assume that covenant terms are chosen to maximize contracting efficiency. These assumptions imply that lenders and borrowers will not allow covenants to be affected by an accounting number if the number can be excluded at minimal cost and if including the number would lead to the arbitrary and costly reallocation of rights between the contracting parties. Therefore, by studying whether an accounting number is allowed to affect covenant terms, we gain insight into the contracting role of that number as well as the costs of excluding it from the contracts. Moreover, by studying how the contracting use of a number changes around a change in accounting standards, we understand how the accounting standard affects contracting efficiency.

We investigate how the magnitude of goodwill relates to whether a lending agreement includes a net-worth covenant or a tangible net-worth covenant. We also examine whether the adoption of SFAS 141 and SFAS 142 is associated with a change in the propensity to use tangible net-worth covenants. The choice between these covenants depends on the relative implementation and renegotiation costs as well as the agency benefits associated with each of these covenants. We conjecture that the relative agency benefits of a tangible net-worth covenant and a net-worth covenant depend on the correlation between goodwill and the borrower’s expected payout. If this correlation is non existent, a tangible net-worth covenant provides incremental agency benefits by reducing noise and opportunism associated with the covenant. We expect that these costs and benefits vary with goodwill. Hence, investigating the level of goodwill in relation to these alternative covenants provides insight into whether goodwill accounting aids or reduces contracting efficiency.

We find that the probability of a tangible net-worth covenant is decreasing in the level of the borrower’s goodwill. This result suggests that the level of goodwill is informative and provides an efficient way of limiting agency costs. The result also suggests that the collateral value of assets is not the sole factor determining the net-worth covenant type.

We also investigate changes in the choice of net-worth covenant type after the adoption of SFAS 141 and SFAS 142. We argue that, if the adoption of these standards reduced the salutary affect of goodwill on contracting costs, we would expect an increased use of tangible net-worth covenants after adoption. Indeed, we find some evidence that the use of tangible net-worth covenants has increased since the adoption of SFAS 141 and SFAS 142, after controlling for other economic factors. This result implies that these standards have reduced the contracting usefulness of goodwill.

Finally, we investigate the covenant slack associated with tangible net-worth and net-worth covenants to gain insight into the costs related to these covenants. We find that the magnitude of covenant slack is not significantly related to the use of tangible net-worth covenants relative to net-worth covenants. Therefore, tangible net-worth covenants appear to be neither a complement nor a substitute for slack.

Our investigation is an initial step in understanding the contracting costs and benefits of goodwill. Our empirical models represent a reduced form of the underlying cost and benefit functions. Therefore, the interpretation of our results is limited. Future research could attempt to better isolate measures of the underlying costs and benefits and provide a richer analysis of the trade-offs faced by the borrower and lender given existing accounting standards and firm characteristics.

Footnotes
1

See Leftwich (1983) for an early paper that can be viewed as taking an “ex ante” approach. More recent papers include Frankel and Litov (2007) and Demerjian (2007).

 
2

Beatty, Weber, and Yu (2006) also find a negative relation between goodwill and use of tangible net-worth covenants.

 
3

This discussion designates both managers and shareholders by the term “borrower.” In so doing, we assume shareholders interests are aligned with managers in their dealings with lenders. However, incentive alignment can vary across firms (Begley & Feltham, 1999).

 
4

Minimum net-worth requirements are not unique in this respect. Any covenant whose requirements are linked to the performance of the firm and is evaluated before the maturity of the debt can reduce the borrower’s incentives to increase variance by acting to shorten the expected life of the option.

 
5

“Hedge funds play hardball with firms filing late financials,” Wall Street Journal, August 29, 2006.

 
6

Smith (1993) notes that lenders may eschew opportunistic use of this renegotiation option to attract future borrowers. If so, this would reduce the value of the renegotiation option and thereby limit its usefulness as a screening mechanism.

 
7

We use the term “implementation costs” to include both the costs of the initial negotiation of the covenant as well as the ongoing monitoring of the covenants.

 
8

This assumes that the slack in net-worth covenants is not set higher than in tangible net-worth covenants to decrease the likelihood of covenants violation. We investigate whether covenant slack varies between tangible net-worth and net-worth covenants. We find that the median slack in contracts that contain net-worth (4.9%) is significantly lower than the slack used in contracts with tangible net-worth covenants (7.2%).

 
9

SFAS 142 must be applied in fiscal years beginning after December 15, 2001. Prior GAAP was derived from Accounting Principles Board Opinion No. 16 Business Combinations (APB 16).

 
10

See Kieso, Weygandt, & Warfiesd, 2004, for details.

 
11

Nor does this argument apply to most coverage-based covenants because those covenants are often computed using earnings before depreciation and amortization is deducted.

 
12

Faced with the possibility of future acquisitions by the borrower, parties can agree to require lender approval of all acquisitions. However, this alternative is costly when future information asymmetry between borrowers and lenders gives rise to renegotiation costs (e.g., Myers & Majluf, 1984; Myers, 1984). Moreover, when bonds are selling below par, such covenants give lenders the opportunity to force borrowers to compensate them for allowing acquisitions that do not hurt the value of the lenders’ claims.

 
13

SFAS 141 provisions must be applied to business combinations initiated after June 2001.

 
14

See Tables 2 and 5 for a detailed description of the variables.

 
15

To account for potential nonlinearity in the relation between goodwill and debt covenants, we also include an indicator variable (GW_ZERO) which equals one if the firm has no goodwill on the balance sheet in the year before the initiation of the loan. Results are qualitatively similar when goodwill is measured as log of (1+GW).

 
16

As a robustness analysis and to check the validity of the DealScan data, we construct an alternative sample. The nature of this sample and its results are reported in Sect. 4.4.

 
17

We identified 70 observations that include a tangible net-worth or a net-worth covenant and had a negative book value. We manually examined these observations. In 40% of the observations, the covenant threshold is negative at the inception of the loan. About 75% of the observations seem to be in violation at inception. We exclude these observations from our estimations because they confound our attempts to control for growth opportunities, a major factor affecting the agency costs of debt.

 
18

Please refer to Table 2 for precise definitions of the variables.

 
19

The frequency of observations across years is available from the authors upon request.

 
20

We also used the 48 industry groups as proposed by Fama and French (1997), with no qualitative change to the conclusions of the study. The use of 48 industry groups significantly reduces the power of our tests given the large number of industries relative to our sample size. We find a reduction in the significance of some of our control variables. Our use of a smaller number of industries reflects an attempt to develop an empirical model with sufficient power while, at the same time, providing meaningful industry controls.

 
21

In an alternative specification, which we do not report, we add an interaction variable to examine whether the slope coefficient on GW is different in the period after the new standards. We find no evidence of a difference while the other results remain unchanged.

 
22

The results for the sample reported in Table 8 appear to be more powerful because many variables, including control variables, turn significant once the observations with negative SLACK are excluded. The coefficient on TNW remains insignificant when models are estimated on a sample that includes negative slack observations.

 
23

We find a significant difference in the means and medians of the three instruments across the groups of firms that use a tangible net-worth covenant compared with the group of firms that use net-worth covenants. However, the ability of the three instruments to explain the use of tangible net-worth covenants is low. The pseudo-R2 of a logistic model whose dependent variable is TNW on the three instruments is 2%.

 

Acknowledgements

We gratefully acknowledge the comments of two anonymous referees, Jack Hughes (the editor), Jerry Zimmerman and seminar participants at Columbia University, New York University and Washington University in St. Louis.

Copyright information

© Springer Science+Business Media, LLC 2007