The Information Content of Loan Covenants



The Information Content of Bank Loan Covenants

by

Cem Demiroglu

and

Christopher James

Department of Finance, Insurance, and Real Estate

Warrington College of Business Administration

University of Florida

Gainesville, FL 32611

First Draft: November 30, 2006

The Information Content of Bank Loan Covenants

Abstract

This paper examines the relationship between the restrictiveness of covenants in bank loan agreements and subsequent operating performance of borrowing firms. Realized future operating performance is used as a proxy for the borrower’s private information concerning credit quality. The restrictiveness of financial covenants may be related to future operating performance because restrictive covenants act as a signaling device that permits high quality firms (e.g., firms with positive private information about their future performance) to distinguish themselves from observationally similar but lower quality firms. We measure the restrictiveness of financial covenants by how close the covenant is set relative to the level of covenant variable at the inception of the loan (covenant tightness) and also by number of covenants in the loan contract (covenant intensity). We find that tighter covenants are associated with improvements in the performance as measured by changes in the covenant variable. Consistent with lenders encouraging revelation of private information via covenant choice, we find that tighter covenants are associated with lower borrowing costs. We also find a larger stock price reaction to the announcement of loans with tighter covenants which is consistent with covenant choice signaling favorable private information. Finally, we find that covenant intensity is associated with better performance but only for the most informationally opaque firms. Overall, our results suggest that the borrower’s choice of the restrictiveness of covenants credibly conveys private information about future performance.

I. Introduction

Do borrowers agree to what appear to be onerous contract terms to credibly convey to the lender their confidence in future improvements in performance? In this paper we address this question by examining the relationship between the restrictiveness of bank loan covenants and subsequent changes in the operating performance of borrowing firms. We use realized future operating performance as a proxy for the borrowers’ private information concerning their creditworthiness at the time of loan. We also examine empirically the determinants of the design of loan contracts and the link between contract design and subsequent performance in the presence of agency costs and adverse selection.

The idea that covenants can serve as a signaling device is motivated by theoretical work by Chan and Kanatas (1985) and Besanko and Thakor (1987) on collateral requirement and more recently by Gerleanu and Zwiebel (2005) and Dessein (2005) on contract design and the allocation of control rights. These models suggest that borrowers (or entrepreneurs) use contract terms to credibly convey private information concerning their future prospects. The basic idea is that borrowers have private information about either their future performance or the costs of a potential covenant violation. Borrowers agree to restrictive covenants when they have favorable private information about their performance or when the costs of covenant violation are expected to be small (for example when ex-post the lenders are expected to have little incentive to enforce the covenant). In the signaling context, borrowers with negative private information (so called poor quality borrowers) will find it costly to mimic the contract choices of borrowers with favorable private information and hence opt for less restrictive covenants. By charging lower spreads to borrowers that choose tighter covenants and penalizing poor quality borrowers upon covenant violations, lenders encourage borrowers to reveal their private information. We refer to this conceptual framework as the Signaling Theory of Covenants (STC).[1]

Anecdotal evidence also suggests that the choice of covenants conveys information about the borrower’s expected future performance. For example, Zimmerman (1975) states that through the loan document and the covenants it contains “…the bank creates a clear understanding to the borrower as to what is expected”. Consistent with this view, when Lucent announced a new loan agreement on February 22, 2001, the Dow Jones News Service reported an analyst commentary that “… the financial conditions attached to the loan were the first hints of guidance Lucent has given since its first quarter conference call.” The analyst also stated that “... the minimum EBITDA requirements that Lucent agreed to reveal the company has committed to losing no more than $1.525 billion in the ongoing second quarter and $825 million in the third quarter.”

Prior empirical work on the importance of signaling in covenant selection is limited. One of the challenges in testing the STC is distinguishing between observable measures of firm quality (credit risk) and private information about borrower quality that may be conveyed through covenant choice. In particular, the most commonly cited reason for the existence of loan covenants and collateral requirements is that they serve to mitigate agency conflicts between shareholders and debtholders.[2] This explanation for the structure of covenants is often referred to as the Agency Theory of Covenants (ATC). Since the agency costs of debt are generally thought to be inversely related to the financial condition of the firm, covenants are expected to be more restrictive in loans to the least creditworthy borrowers (e.g., highly levered firms and to firms with significant growth opportunities).[3] This suggests that the relationship between covenant restrictions and potential proxies for private information such as future changes in operating performance and future loan defaults may be difficult to detect. In particular, since observationally riskier borrowers are more likely to experience poor future performance, if credit risk (based on observables) is measured with error then one may observe a negative relationship between the restrictiveness of covenants and future operating performance even though covenant selection conveys favorable private information. Finding, however, a positive and significant relationship between the restrictiveness of covenants and future operating performance is unambiguous evidence consistent with a signaling story.

The importance of private information and thus signaling is likely to vary inversely with the precision of the lender’s estimate of the creditworthiness of the borrower.[4] This suggests that the relationship between the restrictiveness of covenants and future performance is likely to be the strongest among more informationally opaque borrowers. Thus, the relationship between covenant choice and future performance should be stronger when information problems become more severe.

Another challenge in testing the STC is that the covenant choice set and hence the tightness of a covenant is likely to be related to the level of the covenant variable and thus underlying credit risk of the borrower. As a result, the informativeness of a firm’s covenant choice is likely to vary with credit risk. For example, consider a liquidity covenant that requires a firm maintain a minimum ratio of current assets to current liabilities. Firms with initially high levels of liquidity are likely to be able to choose from a broader menu of covenant levels than firms with initially low levels of liquidity. In this case the information content of the covenant choice depends on the initial level of the covenant variable. To address this problem we place borrowers in clusters based on their levels of the covenant variables at loan inception and then examine covenant choice within each cluster.

Our empirical analysis is based on a large sample of bank loans taken out by non-financial U.S. public firms between January 1995 and December 2001. The primary source for loan contract terms is the Dealscan database from the Loan Pricing Corporation.

We define the restrictiveness of covenants along two dimensions; covenant tightness and covenant intensity. For financial covenants we define covenant tightness by comparing the restrictiveness of each borrower’s covenant choice relative to the choices of borrowers that have similar covenant level choice sets (i.e. in the same cluster). Any covenant choice that is more restrictive than the cluster median is classified as tight. The advantage of using this tightness measure is that we can examine changes in performance along a narrow but presumably value relevant measure. Moreover, it seems reasonable to assume that the borrower’s private information is closely tied to future performance as measured by changes in the covenant variable.[5] For example, borrowers select coverage or liquidity covenants levels that they expect to achieve.

A drawback of using the tightness measure is that lenders often make adjustments to GAAP numbers when defining covenants to account for specific needs or characteristics of the borrower (See Leftwich (1983)). As a result, definitions of covenant variables may differ from how those variables are reported or defined in Compustat. This implies that covenant tightness is likely to be measured with error. As we discuss later, we employ several procedures to minimize measurement error. We focus on two common and economically important financial covenants whose tightness we are able to measure most reliably; the current ratio covenant and the Debt/EBITDA covenant.

We also measure the restrictiveness of covenants by the number of covenants in the loan contract. Following Bradley and Roberts (2004) we define covenant intensity by the number of covenants included in the loan contract (hereafter we refer to this as the covenant intensity index). More specifically, the covenant intensity index equals the sum of six covenant indicators (collateral, dividend restriction, more than 2 financial covenants, asset sales sweep, equity issuance sweep, and debt issuance sweep). The index consists primarily of covenants that restrict borrower actions or provide lenders rights that are conditioned on adverse future events.[6] This suggests that if covenant intensity serves as a signal of borrower credit quality then it is likely to provide information concerning the likelihood or cost of adverse events. As a result, one way to measure future performance is by the frequency of subsequent covenant violations or defaults (see for example Jimenez et al. (2006)). The advantage of this performance measure is that it is a simple measure of credit risk (i.e. the likelihood the borrower is unable to perform according to the loan agreement). However, a potential problem with this approach is that default is an outcome that may be mechanically linked to the number of covenants. Moreover, even though borrowers may expect the likelihood of a covenant violation to increase with the number of covenants, higher quality borrowers may nevertheless choose more restrictive covenants because they expect the cost of covenant violations to be low.

To avoid these problems, when examining covenant intensity we use three broader measures of performance. The first measure is the frequency of CRSP delisting due to poor performance (liquidation, bankruptcy, or the borrowers failure to meet exchange listing requirements). This provides a measure of likelihood of poor operating performance (in the spirit of the frequency of defaults measure) that is not mechanically linked to the number of covenants. The second measure is changes in credit ratings; more specifically, the frequency of credit rating downgrades. This allows us to measure less extreme changes in credit quality. Our third and last measure is changes in Altman’s Z-score.

Overall, we find a positive and significant relationship between the tightness of financial covenants and improvements in operating performance of borrowers. In particular, controlling for credit risk and other determinants of covenant tightness, we find that tighter financial covenants are associated with significant improvements in the covenant variable in the three years following the inception of the loan. Consistent with the selection of tight covenants conveying favorable private information we find that stock returns on the announcement of loans are significantly higher for loans with tight covenants. Finally, controlling for the interdependence between the selection of tight covenants and borrowing costs, we find that borrowers significantly lower their interest costs by choosing tight rather than loose covenants. These results are consistent with covenant tightness conveying positive private information concerning the borrower’s unobservable credit quality.

Turning to the covenant intensity results for the entire sample, in univariate analysis, we find that the covenant intensity is positively related to the likelihood of CRSP delisting due to poor performance and credit rating downgrades. However, this is at least in part attributable to agency problems that lead to more intensive covenants for loans to observably riskier firms. To address this issue, in multivariate analysis, we examine the relationship between covenant choice and performance controlling for observable risk factors and proxies for the severity of information asymmetries. Controlling for these factors, we find that the covenant intensity is associated with improved performance where information asymmetries are most severe. We also find that covenant intensity is priced in the sense that selectivity adjusted spreads are lower when more intensive covenants are chosen.

Overall our empirical analysis indicates that covenant structure is related to observable risk measures as well as borrower’s private information concerning future changes in credit quality. Loans to observationally riskier borrowers have more restrictive covenants while at the same time borrowers appear to convey private information by their choice of covenants. To our knowledge this is the first paper to provide evidence on the importance of these potentially conflicting influences on contract choice.

The remainder of the paper is organized as follows. In section 2 we describe our sample of bank loans and our measures of covenant tightness and covenant intensity. In section 3 we outline our empirical tests and provide summary statistics. We present our empirical results in Section 4. Section 5 provides a summary and conclusion.

II. Sample and Measures of Restrictiveness of Covenants

A. Overall Loan Sample

Our primary data source on the terms of bank loan agreements is Loan Pricing Corporation’s (LPC) Dealscan database.[7] Our sample covers loans made from 1995 though 2001. We focus on dollar denominated bank loans of non-financial U.S. firms with publicly traded common stocks listed on CRSP files as of loan activation date.[8] We require that borrowers have financial information in Compustat for the fiscal year preceding the loan agreement. We eliminate debtor-in-possession (DIP) financings and subordinated loans. Finally, we eliminate loans with maturity (at loan inception) less than one year. We focus on long-term loans for two reasons. First, for very short term loans the loan renewal or roll over process serves as a substitute for covenants in controlling moral hazard. Second, we measure performance by yearly changes in operating performance. As a result, in the case of short-term loans there is uncertainty as to whether the restrictions in the loan contract are effective throughout the performance horizon.

The resulting loan sample includes 11,660 loans from 3,689 unique borrowers. Dealscan provides information on loan maturity, amount, type, purpose, syndication, covenants and pricing. We use Compustat to collect information on the financial condition of borrowers, CRSP for information on borrower common stocks, and I/B/E/S for analyst forecasts on borrower earnings. A more detailed description of the variables used in this study will be provided below.

The Dealscan database provides information on a number of financial and other affirmative and negative covenants. Financial covenants are restrictions on accounting variables (ratios) that must be maintained over the life of the loan. Dealscan reports information on 17 different financial covenants that fall into five broad covenant categories: coverage, leverage, liquidity, net worth, and capital expenditures.[9] About two-thirds of the loans include at least one financial covenant and the average loan includes 2.15 financial covenants. Debt/EBITDA and net worth (including tangible net worth) covenants are the most popular financial covenants (both are used in 52 percent of loans with financial covenants).[10] Dealscan also provides information on covenants restricting dividend payments (restricting dividends under certain conditions), collateral requirements and prepayments requirements (so called sweeps that mandate that a portion of the loan be repaid out of excess cash flows, debt and equity financings, or asset sales proceeds).[11]

Dealscan also provides information on the all-in-drawn spread. This is a measure of the borrowing cost per dollar of borrowing expressed as a basis point mark-up over the 6-month LIBOR and it includes recurring fees associated with the credit facility. LPC computes the spread for non-LIBOR based loans by converting index used to price the loan into a LIBOR equivalent using the historical relationship between the index and the LIBOR. All-in-drawn spread has been used as a measure of borrowing costs in a number of previous empirical studies on loan pricing (see for example, Bradley and Roberts (2004), Brav et al (2006), Guner (2006), Moerman (2005)).

The basic unit of observation in Dealscan is a loan, also referred to as a “facility”. If a borrower enters into multiple loan agreements with the same lenders on the same date these loans are grouped together under a deal “package”. Loan covenants are typically drafted by package and hence all of the loans in a package are subject to the same covenants. Therefore, a loan-level analysis of covenants over-weights characteristics of borrowers taking out multiple facilities. This, however, is a less important concern in our study since we exclude all loans with maturity (at loan inception) less than one year. In particular, since most deal packages with multiple loans consist of one long-term and one short-term (i.e. less than 1-year) facility, after excluding short-term loans only about 25 percent of the loans in our sample belong to a deal package with multiple facilities. Moreover, when testing the relationship between covenant tightness and borrowing costs, a loan-level analysis is appropriate since loans that are in the same package may be priced differently. Therefore, the empirical results we provide in this paper are based on a loan-level analysis. Note that all of our results are robust to package-level analysis. Moreover, our multivariate results are robust to clustering standard errors at the package (or borrower) level.

B. Measuring the Restrictiveness of Covenants

Perhaps the most straightforward and potentially informative measure of the restrictiveness of financial covenants is covenant tightness. By covenant tightness we mean how close the covenant is set relative to the level of the covenant variable at the time the loan is made. Tighter covenants are more restrictive in the sense that the borrower has less flexibility with respect to decisions that may adversely affect the level of the covenant variable. Moreover, since lenders and borrowers presumably set covenants at levels they expect borrowers to achieve, where the covenants are set is potentially informative of the borrower’s expectation and confidence concerning future performance. Finally, since covenant tightness is defined based on the covenant variable, changes in the covenant variable provides a clear and presumably value relevant benchmark on which to measure change in performance.

Previous empirical studies on the determinants of covenant structure typically measure the restrictiveness of covenants by the presence of a particular covenant (for example, whether the loan is secured) or the number of covenants included in the loan contract. (See for example Booth and Booth (2006), Bradley and Roberts (2004), Billett, King and Mauer (2006) and Nash, Netter and Poulsen (2003)). Covenant restrictions are measured in this way in part because covenants such as collateral requirements are thought to be particularly important in controlling agency problems and partly because the existence and the restrictiveness of different covenants tend to be correlated with one another. The positive correlation among covenants probably arises because covenants restricting one activity affect borrower’s incentives to undertake other activities. While we examine the information content measuring covenant intensity, signaling through covenant intensity is likely to be more difficult to detect because one is forced to use broad performance benchmarks and because covenant intensity is likely to be closely linked to observable credit risk that may be measured with error.

C. Covenant Tightness Sample

A frequently cited advantage of bank loans is that the covenants can be designed and defined to meet the specific needs of the borrower.[12] However, this also means a lack of uniformity in how covenants are defined. As a result, covenant tightness calculated using Compustat financials and generic covenant names in Dealscan may suffer from significant measurement error.

According to Dichev and Skinner (2002) current ratio covenants have the most standard GAAP based covenant definitions among all financial performance covenants.[13] In addition, existing research on covenant violations indicates that current ratio covenants are among the most frequently violated covenants, which suggests that the tightness of current ratio covenants is economically important to borrowers (see e.g., Beneish and Press (1993), Chen and Wei (1993), and Sweeney (1994)). For these reasons, we include loans with current ratio covenants in our tightness sample.

Current ratio covenants principally condition lender control on changes in the borrower’s liquidity. Thus, current ratio covenants are different from financial covenants based on cash flows and leverage which are often cited as critical factors in the lending decisions. As a result 94 percent of Dealscan loans that include at least one financial covenant include a coverage covenant. Examining cash flow and leverage covenants requires getting around the measurement error associated with determining tightness and compliance. One way to get around the problem is to focus on a sample of loans where the covenants are defined so that tightness and compliance can be determined using financial data in Compustat (i.e. if there are no non-GAAP adjustments to the definition of the covenant). The covenant definitions and schedules are available for a sub-sample of Dealscan loans with Tearsheet information. Because of the high cost of hand collecting information on covenant definitions and quarterly covenant schedules, we use this methodology to study only the most commonly used financial covenant in bank loans, Debt/EBITDA covenant.

Loan sample for current ratio and Debt/EBITDA covenants consists of 956 and 559 loans, for 506 and 234 borrowers, respectively.

D. Measuring Covenant Tightness

One measure of covenant tightness is the slack (i.e. distance to default) which is simply the difference between the level of the covenant variable at the initiation of the loan and minimum or maximum level permitted by the contract. The greater the slack the less restrictive the covenant since the borrower’s financial condition can deteriorate more before triggering a covenant violation. While the slack provides a useful measure of covenant tightness when testing agency theories of covenant choice, slack is not a good measure of tightness when investigating the information content of covenant choice. The reason is that the slack is likely to be negatively correlated with credit risk since the level the covenant variable determines the menu of covenants applicable to a particular borrower. For example, when choosing a minimum current ratio covenant, a borrower with a current ratio of 1.50 has to choose a covenant below 1.50, or else he will be immediately in violation. The choice set of another borrower with a current ratio of 2.50 is obviously bigger. This borrower can choose a covenant anywhere between 0.50 (the lowest current ratio covenant in our sample) and 2.50. This heterogeneity of choice sets explains why firms with lower current ratios are more likely to choose covenants with a lower slack than firms with high current ratios.[14] This in turn implies that the slack measure will be correlated with observable credit risk measures.

To address this problem we compare the covenant choices of borrowers with the same choice sets and investigate whether their choices are correlated with subsequent performance conditional on credit risk. The fact that covenants are clustered at discrete numbers allow us to partition our covenant samples in order to form clusters in which borrowers have the same menu of covenant choices. Borrowers that have similar financial ratios are subject to similar covenant menus. Within each cluster we sort borrowers by their covenant choices. Then we classify covenants that are more restrictive than the cluster median as tight. Finally, we pool borrowers/loans from all clusters. Note that classifying covenant choices in discrete tight vs. loose categories serves to lower the measurement error problem associated with imperfect covenant benchmarks.

For the current ratio covenant sample we use quarterly Compustat files to calculate the ratio of current assets (QDI # 40) to current liabilities (QDI # 49). Current ratio covenants in our sample range between 0.50 and 3.50 and are mainly clustered at discrete levels 0.25 from each other. Also, the current ratios of our sample firms range between 1.00 and 9.30. Therefore, we place borrowers with current ratios below 3.50 into 10 clusters. Each of these ten clusters has a cluster width of 0.25. For example, the first cluster includes firms with current ratios between 1.00 and 1.24; the second cluster includes firms with current ratios between 1.25 and 1.49, and henceforth. All borrowers with current ratios greater than or equal to 3.50 are placed in an 11th cluster.[15] In each cluster, borrowers that choose covenants greater than or equal to the cluster median covenant are classified as choosing tight covenants. After making the tight/loose covenant classifications in each cluster, we pool the data from clusters.

We measure the tightness of the Debt/EBITDA covenant choice in a similar manner. As mentioned before, Debt/EBITDA covenant sample consists of loans with Tearsheet information in Dealscan. Tearsheets includes a summary of covenant definitions as well as the covenant schedules (e.g. changes in covenant levels over time). After reading all Tearsheets, we eliminated loans with non-GAAP adjustments to Debt/EBITDA covenant.[16] We define Debt/EBITDA as the ratio of 4-quarter moving average debt (defined by using long-term debt (QDI # 51) and debt in current liabilities (QDI # 45)) to four quarter moving sum EBITDA (QDI # 21). We use a 4-quarter moving average because Debt/EBITDA covenants are most frequently defined as an average of prior quarters. The Debt/EBITDA covenants range from 2.00 to 7.00 with clustering at discrete intervals of 0.25. As a result we partition the data in 21 even clusters. The bottom cluster includes firms with Debt/EBITDA ranging from 0 and 2.00 and the top cluster includes firms with Debt/EBITDA greater than 6.75. Borrowers in the bottom cluster have the greatest menu of covenant options. In each cluster, borrowers that choose covenants less than or equal to the cluster median covenant are classified as choosing tight covenants.

While measuring covenant tightness based on clusters controls for differences in credit risk we also conducted the analysis based on slack. As discussed below the main results are similar when we use this alternative measure of covenant tightness.

E. Measuring Covenant Intensity

As discussed earlier covenant intensity refers to the number of covenant restrictions contained in the loan document. Our aggregate measure of the covenant structure is Bradley and Robert’s (2004) covenant intensity index. The index equals the sum of six covenant indicators: collateral, dividend restrictions, asset sales sweep, debt issuance sweep, equity issuance sweep, and the existence of more than 2 financial covenants. Therefore, the value of the intensity index ranges between 0 and 6. When the value of one of the indicators is missing we set the index to missing and eliminate the loan from the intensity analysis. There are 3,704 loans from 1,564 borrowers in our covenant intensity sample.

Note that the covenant intensity index equally weights each covenant and hence implicitly assumes that each covenant is equally restrictive (or important) for borrowers. As a robustness check we modify the intensity index by counting only once the existence of a sweep (so that the maximum value of the index is 4) and by including in the count all categories of financial covenants (so that the maximum count is 10). Our results are similar when we use these alternative measures.

III. Empirical Tests and Summary Statistics

A. Empirical Methodology

Under the STC, covenant selection provides a signal of the borrower’s private information concerning his credit quality. The basic idea is that covenant violations involve costs to borrowers so that the selection of more intensive covenants by higher quality borrowers is costly for lower quality borrowers to mimic.[17] Higher quality borrowers have an incentive to choose more restrictive covenants if, controlling for observable risk characteristics, more restrictive covenants are associated with lower borrowing costs.

Testing the STC requires a proxy for the borrower’s private information regarding credit quality. Our empirical tests assume that the borrower’s private information concerning credit quality is correlated with realized future operating performance. The STC predicts that borrowers with favorable private information select more restrictive covenants. Thus, finding a positive relationship between the restrictiveness of covenants and future operating performance is consistent with covenants choice serving as a signal. We specify our empirical model with future performance as an explanatory variable because future performance is assumed to be a proxy for the private information the borrower has at the time terms of the loan agreement are negotiated.[18] In testing the STC, controlling for observable risk characteristics is critical since agency based theories of covenant choice predict that riskier borrowers will be subject to more restrictive covenants and therefore we would expect future operating performance to be negatively correlated with observable credit risk measures (i.e. riskier customers are expected to perform worse, on average).

When examining the choice of covenant tightness we measure performance by changes in the ratio that the covenant is written on. In particular, we examine changes in the current ratio and Debt/EBITDA from the end of the first quarter after loan inception to the end of 5th, 9th, and 13th quarters. Since covenant intensity is measured by the number of covenants that cover a potentially broad set of indicators of financial health we use three broad based measures of performance. The first measure is the frequency of delisting due to poor performance (liquidation, bankruptcy, and the borrowers failure to meet exchange listing requirements) in the three years following the inception of the loan.[19] The second measure is frequency of rating downgrades. For firms without a credit rating, we measure downgrades by declines in estimating credit ratings sufficient to move the borrower into a lower alpha based credit rating. We estimate credit ratings using an ordered probit model developed by Blume, Lim, and MacKinlay (1998). A detailed explanation of the rating model is provided in Appendix B. Our final broad measure of performance is changes in Altman’s Z-score as defined in Sufi (2006).

Jimenez et al (2006) argue that the importance of collateral as a signal of borrower quality varies with the degree of information asymmetries. The basic idea is that private information about credit quality is more important when lenders evaluate opaque (i.e. less transparent) borrowers. This argument suggests that the relationship between covenant choice and future performance will vary with the opacity of the borrower.

If the restrictiveness of covenants conveys favorable private information concerning the borrower credit quality then we would expect that controlling for observable risk differences, tighter and more intensive covenants to be associated with relatively lower loan spreads. Note, however, that the prediction of a negative relationship between borrowing costs and the covenant restrictiveness is not unique to the signaling story. For example, the ATC also predicts that more restrictive covenants will be associated with lower borrowing costs.

We examine the relationship between borrowing costs and covenant restrictions using a two-step procedure similar to the ones used by Booth and Booth (2006) and Bradley and Roberts (2004). The first step involves estimating a model of the determinants of covenant restrictions. The second step involves estimating a selectivity bias adjusted model of borrowing costs for both borrowers that select tight (intensive) and loose covenants under alternative covenant structures. We expect that the lower the likelihood of choosing restrictive covenants (according to the selection model) the lower the borrowing costs for borrowers that select tight or intensive covenant structures.

While better quality borrowers may select tight covenants, selecting tight covenants exposes the borrower to a greater likelihood of default for modest deteriorations in performance. Thus, tighter covenants may be associated with a higher probability of default. However, if tight covenants provide a positive signal of borrower quality then violating tightly set covenants may be associated with less severe consequences than if covenants are set loosely. To examine this issue, we investigate the relationship between the frequency of bankruptcy or delisting due to poor performance, covenant violations and how tightly covenants are set at loan inception.

Our final empirical test of the STC involves examining the relationship between the stock price reaction to loan announcements and the intensity and tightness of covenants. If covenant structure conveys favorable private information we would expect a more positive share price reaction associated with loans with tight or intensive covenant structures.

B. Summary Statistics.

Table 1 provides summary statistics for the borrowers and loans in overall sample of 11,660 loans as well as sub-samples of loans with Debt/EBITDA and current ratio covenants. While summary statistics for the overall sample include loans with Debt/EBITDA and current ratio covenants the statistical tests are based on comparing borrowers/loans in the Debt/EBITDA and current ratio samples to borrowers/loans without these covenants in their loan agreements. [20] Our summary statistics for the overall sample are very similar to those reported in related papers using Dealscan (see for example, Sufi (2006)).

The summary statistics in Table 1 suggest that smaller firms are more likely to have Debt/EBITDA and current ratio covenants in their loan contracts. The reported credit ratings are based on actual credit ratings for senior debt (i.e. historical S&P senior (or long-term) credit ratings in Compustat) when they are available. When ratings are not available we use estimated ratings from an ordered probit model developed by Blume, Lim and MacKinlay (1998). As shown, the mean (and median) credit ratings are significantly lower for firms in the Debt/EBITDA and current ratio samples.[21] In particular, notice that while the average letter rating for all firms is BB, mean and median numerical estimated credit rating is lower for the covenant sub-samples implying greater levels of credit risk for these firms. Consistent with the differences in credit rating, loans with Debt/EBITDA and current ratio covenants are associated with higher borrowing costs and higher likelihood of being secured. Finally, the mean (and median) Z-score for firms in the Debt/EBITDA and current ratio samples are somewhat higher than the overall sample suggesting these firms are less risky. This, however, is primarily due to the weight placed on working capital in the Z-score calculation and the fact that firms in the Debt/EBITDA and current ratio samples have significantly more liquidity.

The pre-loan leverage of firms with Debt/EBITDA covenants is only slightly higher than the leverage of firms without Debt/EBITDA covenants. However, firms with current ratio covenants have significantly lower (about 10 percent) pre-loan leverage than firms without current ratio covenants. There are two potential explanations for these findings. First, loan amount is significantly higher for loans with Debt/EBITDA and current ratio covenants, which suggests that the decision to include covenants may among other things be based on pro-forma leverage rather than pre-loan leverage. Second, because of high earnings volatility, low profitability, and opacity, firms with current ratio covenants may have limited access to outside capital and relatively lower debt capacity.[22] Therefore, the low pre-loan leverage of firms with current ratio covenants does not necessarily imply better credit quality.

Covenant structure may also be related to severity of information asymmetries because adverse selection problems may lead more restrictive covenants to control agency problems and because covenants may play a more important signal role for informationally opaque firms. We measure the information asymmetries using three proxies. The first proxy is the forecast error in analysts’ earnings estimates for the fiscal year prior to loan inception.[23] Following Christie (1987), we define the analyst earnings forecast error as the ratio of the absolute difference between predicted earnings and the actual earnings (per share) to the price per share at the beginning of the month of forecasts. Predicted earnings are defined as the mean monthly earnings forecast for the month preceding the last month of the fiscal year prior to the loan.[24] Higher forecast errors are assumed to be associated with greater information asymmetries between borrowers and lenders concerning future cash flows. Our second measure is the standard deviation of abnormal returns on quarterly earnings announcements in the five years before the initiation of the loan (see Dierkens (1991) and Krishnaswami and Subramaniam (1999)).[25] A significant stock market reaction to an earnings announcement suggests that the market did not anticipate the information and hence greater information asymmetries about a firm’s cash flows. Our final measure of asymmetry information is the age of the borrower. Older borrowers are likely to have a more established track record and hence lenders are likely to be better informed about their prospects. (See Jimenez et al (2006))[26]. As shown in Table 1, firms with current ratio covenants are significantly younger and more opaque than firms in the overall sample of firms. Firms with Debt/EBITDA covenants are also younger than the overall sample of firms though we have mixed evidence about the relative transparency of the firms in the two groups.

Bradley and Roberts (2004) and Billet et al (2006) find that the use of one type of covenant tends to be highly correlated with the use of other types of covenants. Based on these findings we would expect that loans in the Debt/EBITDA and current ratio samples have more intensive covenant structures than the overall sample of loans. As shown in Table 1 we find the mean and median covenant intensity index is higher for loans with Debt/EBITDA covenants. In contrast, the intensity index is lower for loans in the current ratio covenant sample. However, this appears to be due to the fact that current ratio covenants substitute for prepayment requirements (i.e. sweeps). For example, 76 percent of loans with Debt/EBITDA covenants contain asset sales sweeps versus 41 percent of loans without Debt/EBITDA covenants. In contrast, 39 percent of loans with current ratio covenants have asset sales sweeps versus 63 percent of loans without current ratio covenants. Otherwise, loans with Debt/EBITDA and current ratio covenants are more likely to include collaterals and restrictions on dividend payments than loans without these covenants.

Conceptually, the relationship between syndication and covenant choice is unclear. On the one hand, syndication involves multiple lenders, which increases hold out problems and other impediments to debt restructuring. To mitigate these problems the syndicated loans may have fewer and looser covenants similar to what one sees in public debt contracts. Alternatively, since covenant violations are used to trigger monitoring on the part of lenders (see Rajan and Winton (1995)), free-rider problems associated with monitoring may lead to more restrictive covenants when loans are syndicated. Drucker and Puri (2006) argue that loans that are expected to be sold contain more restrictive covenants which may imply that syndicated loans have more restrictive covenants. As shown in Table 1, we find that loans with Debt/EBITDA covenants on average have more lenders than the loans in the overall sample. However, we find that the loans with current ratio covenants have fewer lenders, on average, and are less likely to be syndicated. One reason for this is that, as shown in Table 1, loans with current ratio covenants are taken out by much smaller borrowers and are, on average, much smaller (in dollar terms) than loans in the overall sample.

Table 2 provides summary statistics for our samples of loans with current ratio and Debt/EBITDA covenants grouped by the tightness of the covenants. As shown, all-in-drawn spreads are significantly higher for loans with tighter covenants suggesting these loans are riskier. The greater credit risk appears to stem from lower earnings since we find no significant differences in leverage based on covenant tightness. It is important to remember that since tightness is measured using clusters of firms that have similar levels of the covenant variables, we would find much larger differences in credit spreads and observable risk measures than those in Table 2 if we defined tightness by the distance to default (i.e. slack). For example, smaller, less profitable, more highly levered firms, as well as firms with high investment spending are significantly more likely to agree to current ratio and Debt/EBITDA covenants with lower distance to default (not reported).

One reason firms may choose tighter covenants is that they have less volatility in earnings or liquidity. Consistent with this argument, mean of the quarterly standard deviation of EBITDA/Sales is significantly less for firm with tight coverage covenants and the mean and the median of the standard deviation in quarterly current ratios is significantly lower for firms with tight current ratios.[27]

As shown in Table 2, we find no systematic relationship between the tightness of covenants and information asymmetries. For example, while the median analyst forecast error is significantly larger for the firms with tight covenants borrowers in the tight Debt/EBITDA covenant sample have significantly lower mean and median volatility in earnings announcement returns. In addition, there are no significant differences in the age of borrowers based on the tightness of covenants.

Finally, notice that the frequency of loans to finance takeovers is higher in the tight covenant sample. This suggest that tight covenants may be one way borrowers credibly commit to lenders to undertake changes designed to improve cash flows or liquidity associated with the acquisition.

Table 3 provides summary statistics based on the intensity of covenant structure. Firms with more restrictive covenants in their loan agreement are significantly riskier and more informationally opaque than firms with no or few covenants in their loan agreements. For example, firms in the no covenant sample have significantly higher earnings, use less leverage and have significantly lower earnings volatility than firms in the high intensity sample. Credit risk also appears to increase with the intensity of the covenant structure. In particular, leverage is significantly higher and earnings are significantly lower and more volatile for firms in the high intensity than for firms in the low intensity sample.

As shown in Table 3, firms with loans that contain more intensive covenants are on average younger and have higher analyst earnings forecast errors than firms with less intensive structures. Overall, these results indicate that information asymmetries and observable credit risk are important determinants of covenant intensity. Comparing the findings in Table 2 to those in Table 3 it may appear as though differences in covenant intensity are associated with larger differences in credit risk than are differences in covenant tightness. However, once again, it is important to remember that our tightness measure accounts at least partly for differences in the credit risk. In particular, by using clusters we are comparing differences in covenant choice for firms with similar liquidity or coverage ratios and that are therefore likely to have similar observable credit risk characteristics.

IV. Empirical Results

A. Univariate Analysis of Covenant Choice and Operating Performance

Table 4 provides the results of a univariate analysis of covenant choice and future performance. Panels A and B provide a comparison of industry adjusted changes in current ratios and Debt/EBITDA by the tightness of the current ratio and Debt/EBITDA covenants, respectively, at the initiation of the loan.[28] Industry adjusted changes are computed as the difference between the change for each firm and the change in the median firm within the same four-digit SIC code as the sample firm.[29] Consistent with the STC, we find significant increases in mean and median current ratio for firms with tight current ratio covenants. The percentage increase in current ratios for firms with tight covenants is significantly greater than the change for firms with loose covenants and firms without liquidity covenants. Indeed, the median change in current ratios for firms in the loose covenants is negative indicating a reduction in liquidity after the loan is made.

As shown in Panel B, the performance of firms with tight Debt/EBITDA covenants also improves. In particular, the mean and median Debt/EBITDA decrease significantly relative to the firms with loose covenants and relative to firms without coverage covenants. While the performance of firms with tight Debt/EBITDA covenants improves, the performance of firms with loose covenants deteriorates (i.e. the amount of debt relative to earnings increases).

In contrast to the improvements in performance for firms with tight covenants, we find that the overall performance of firms with more intensive covenants deteriorates. As shown in Table 5, firms in the high intensity sample are more likely to be delisted due to poor performance and are more likely to experience a credit downgrade in the three years following the initiation of the loan. Moreover, the mean and median Altman’s Z-score decreases more for firms with intensive covenants than for firms with less restrictive covenants in their loan agreements.

Overall, the univariate analysis provides no support for the signaling story concerning the choice of covenant intensity. The negative relationship between covenant intensity and performance may reflect the role of covenants in controlling agency problems. In particular, agency problems may lead to a positive relationship between covenant intensity and observable credit risk. Under the signaling hypothesis, covenant choice allows lenders to distinguish between firms that are indistinguishable based on observable risk characteristics. In next section we present a multivariate analysis of covenant choice that attempts to control differences in firm characteristics.

We also examine the relationship between covenant tightness and the broad performance measures reported in Panel C (not reported for brevity). Overall, we find no systematic relationship between covenant tightness and declines in overall performance. In particular, we find no significant difference in the frequency of credit rating downgrades or in changes Z-scores between the firms in the loose and tight covenant samples. We do however find that the frequency of a delisting in the three years following the loan agreement is higher for firms in the tight covenant sample. For example, the frequency of delisting through year 3 is 8.31 percent and 10.49 percent for firms in the tight current ratio and Debt/EBITDA covenant samples, respectively. In contrast, the frequency of a delisting for firms with loose current ratio and Debt/EBITDA covenants is 5.23 percent and 2.08 percent, respectively. As we discuss in the next section, the higher frequency of delisting appears to be related to higher observable credit risk for firms with tight covenants.

B. Multivariate Analysis of Covenant Choice

We control for credit risk and factors other than signaling that may affect covenant choice by estimating a multivariate model of covenant choice. For covenant tightness, we estimate probit models relating the likelihood of tight covenant selection to loan characteristics, borrower characteristics, proxies for information asymmetries and changes in the covenant variable in the year following the loan agreement. We also estimate the model using performance during the two and three years following the loan was made and get similar findings (not reported for brevity). For covenant intensity, we estimate a Poisson regression with the covenant intensity index as the dependent variable.[30]

Estimates of the probit model relating the likelihood of choosing tight covenants to loan and borrower characteristics are presented in Table 5. We provide estimates for three specifications of the model based on three different proxies for information asymmetries (age, volatility in earnings announcement returns, and analyst earnings forecast errors). Each regression includes an intercept, in addition to industry (based on one-digit SIC codes), year, loan type, and loan purpose fixed effects (unreported). As shown in Panel A, smaller firms and firms with lower earnings are more likely to have tight current ratio covenants. This result is consistent with riskier borrowers receiving more restrictive covenants. We also find a positive and statistically significant relationship likelihood of choosing tight current ratio covenants and the amount of investment spending (i.e. capital expenditures plus R&D). This result is consistent with tight current ratio covenants being used to mitigate agency cost of debt for high growth firms. Specifically, restrictions on changes in liquidity may serve to limit asset substitution problems that are particularly acute among high growth firms. Finally, we find a negative but insignificant relationship between tightness and the historical volatility of borrowers’ current ratios.

Consistent with the signaling hypothesis, we find a positive and statistically significant relationship between covenant tightness and future change in the borrower’s current ratio (our proxy for the borrower’s private information at the time the loan is negotiated). This is finding is consistent with the argument that borrowers select tight current ratio covenants when they expect their performance to improve.

As shown in Panel B of Table 5, riskier borrowers are more likely to agree to tight Debt/EBITDA covenants. In particular, notice that tighter Debt/EBITDA covenants are more likely the lower the earnings of the borrower and the greater the amount of leverage used. This suggests that, not surprisingly, tight coverage covenants focus on claims dilution problems arising from the issuance of additional debt. Finally, the likelihood of tight covenants is less likely the more volatile the borrower’s Debt/EBITDA prior to the origination of the loan. This suggests that covenants are set looser the more difficult it is to forecast the covenant variable.

The negative and statistically significant relationship between tight Debt/EBITDA covenants and subsequent changes in the borrower’s industry adjusted Debt/EBITDA is consistent with the signaling hypotheses. In other words, borrowers that expect their performance to improve in terms of reductions in the amount of debt relative to earnings (i.e. increased debt coverage) select tighter Debt/EBITDA covenants.

Finally, we find no relationship between covenant tightness and the severity of information asymmetries. Note that in Panels A and B of Table 5 the estimated coefficients on our information proxies are generally not statistically significant. Moreover, we find no evidence that the importance of covenant tightness as a signal varies with the severity of information problems. Specifically, when we interact performance changes with our asymmetric information proxies (not reported) we find that the interaction effects are not statistically significant.

We also estimate the probit models of covenant tightness using our broad measures of performance. Using a broad based measure of performance as a proxy for the borrowers’ private information is likely to cause attenuation bias if the private information conveyed in the financial covenant choice relates narrowly to the covenant variable. Overall, we find no significant relationship between covenant tightness and the changes Altman Z-score, the likelihood of credit rating downgrades, or the frequency of delistings due to poor performance. We also find no significant relationship between covenant tightness and these performance measures interacted with our proxies for information asymmetries. Thus, the private information conveyed through covenant choice appears to pertain narrowly to the expected future performance of the covenant variable.

Turning to our analysis of the determinants of covenant intensity, Panels A, B and C in Table 6 provide estimates of Poisson regressions relating covenant intensity to firm and loan characteristics and our three proxies for the borrower’s private information. Because delistings are infrequent in the first year following the loan, we report the results using delistings within three years of the initiation of the loan.[31]

Overall, the results reported in Table 6 indicate that, consistent with previous empirical studies, agency problems are important determinants of the covenant structure of bank loan agreements (see, for example, Bradley and Roberts (2004)). In particular, we find a positive and significant relationship between covenant intensity and leverage and the negative relationship between covenant intensity and earnings. These results are consistent with the ATC argument that the agency costs of debt are expected to be greater the riskier the borrower. We also find that covenant intensity is positively related to the maturity of the loan. This result is consistent with the argument that covenants serve to mitigate agency problems by reducing the effective maturity of the loan (see Billett et al (2006)). Moreover, for most of the specifications, we find a positive and significant relationship between covenant intensity and investment spending. This is consistent with the argument that covenants serve to mitigate asset substitution problems associated with growth firms. Finally, information asymmetries appear to be important determinants of covenant intensity. On average, younger firms and firms with less predictable earnings have loans with more restrictive covenants.

If we ignore the cross effects of the information proxies with our performance measures, we find covenant intensity is associated with deteriorations in future operating performance. In particular, as shown in Table 6, we find a positive relationship between covenant intensity and the frequency of rating downgrades and delistings and a negative relationship between covenant intensity and changes in Altman’s Z-scores. One explanation for this is that covenant intensity is positively related to observable credit risk and that our credit risk measures do not completely control for differences observable credit quality. Again, this is more likely to be a problem when examining covenant intensity than covenant tightness because we measure covenant tightness by first clustering based on covenant variable.

To address this problem, we include in the model cross effect variables defined as our performance measures interacted with the various information proxies. The idea is that since private information regarding credit quality is likely to be more important when lending to younger and more informationally opaque borrowers, the use of covenant intensity to signal credit quality will be more prominent among opaque rather than transparent borrowers.[32] Consistent with this argument, the coefficient estimates for the cross effects variable shown in Table 6 are negative and significant when analyst earning forecast errors or earnings announcement returns are used as information proxies and the downgrades or delistings performance measures are used and positive and statistically significant when performance is measured by changes in Altman’s Z-scores. These results are consistent with the argument that the importance of covenants as a signal increases as information asymmetries become more severe. The results using age as an information proxy are also consistent with signaling. In particular, since information asymmetries are assumed to decrease with the age of the borrower, the positive and significant coefficient for the cross-effects with downgrades and delistings and the negative coefficient for cross-effects for changes in Altman’s Z-score are consistent with a signaling hypothesis. Overall, these results are consistent with covenants playing a role in mitigating adverse selection problems.

C Are Covenant Restrictions Priced?

Borrowers will have an incentive to signal private information concerning credit quality by choosing more restrictive covenants if they are rewarded for doing so through lower costs of borrowing. Thus, borrowing costs and the choice of covenant tightness (intensity) are interdependent. We account for this interdependence using a two step selectivity adjustment procedure described in Lee (1978) and Heckman (1979) and employed recently by Booth and Booth (2006) to examine loan pricing. This methodology involves first estimating a probit model of covenant choice. In the case of covenant tightness the dependent variable equals one if a tight current ratio or Debt/EBITDA covenant is chosen. For covenant intensity we convert the count variable to a binary intensity variable and then estimate the probit model of covenant choice. The binary intensity measure equals to one if the covenant index is greater than 4 and zero if the covenant index is zero. As a robustness check, we also define the binary intensity variable as one if the index is greater than 4 and zero if the index is zero or one. In the probit model, we assume that the choice of tight or intense covenants is a function of the same loan and borrower characteristics described in Tables 5 and 6, respectively.

We use the linear predictors of the first step probit model to compute the inverse Mills ratios. The inverse Mills ratio is calculated as φ (ψ) / Φ (ψ) when tight (intensive) covenants are selected, and φ (ψ) / (1 - Φ (ψ)) when the loan agreement contains loose (less intensive) covenants. Here φ is the standard normal density function, Φ is the standard normal cumulative distribution function and ψ is the estimated linear predictor from the first-stage probits. The second step involves estimating (via OLS) the relationship between borrowing costs and observed loan and borrower characteristics conditional on the selection of tight (intensive) or loose (less intensive) covenant structures. We include in the borrowing cost regression the inverse Mills ratio as a selectivity variable. Intuitively, the inverse Mills ratio provides a measure of the lenders updated beliefs regarding credit quality based on the choice of covenants. A negative coefficient for the inverse Mills ratio implies that the choice of tight (intense) covenants, on the margin, reduce borrowing costs by increasing the lenders perception of the credit quality of the borrower. Therefore, we expect the estimated coefficient on the inverse Mills ratio to be negative for borrowers that choose tight (intense) covenants and positive or insignificant for borrowers that select loose (less intense covenants).

Table 7 and 8 provide estimates of selectivity corrected loan spread regressions relating borrowing costs to loan and borrower characteristics as well as the inverse Mills ratio calculated from the first step probit model.[33] We measure borrowing costs by log of the all-in-drawn spread. Consistent with the argument that tighter or more intensive covenants lead to lower borrowing costs the estimated coefficient on the inverse Mills ratio is negative and statistically significant for the sample of loans with tight or intense covenants. This is consistent with borrowers having an incentive to signal favorable private information through the covenant choice. In contrast, the estimated coefficient on the inverse Mills ratio for loans with loose or less intense covenants is positive (and for most specifications statistically significant). Note that the positive coefficient on the selectivity variable is consistent with the ATC in that borrowers that select loose covenants appear to trade greater flexibility for higher borrowing costs.

E. Covenant Tightness and Covenant Violations

For a given change in operating performance, tighter covenants expose borrowers to a greater likelihood of violation. However, a finding of more covenant violations among borrowers that choose tight covenants is not necessarily inconsistent with the signaling hypothesis. In particular, a borrower may select tight covenants even if they are more likely to violate if they expect the costs of a violation to be lower when covenants are set tightly. The costs of a violation in turn depend how lenders react to a violation. Lenders’ reactions to violations can vary in severity from a simple waiver to a demand for immediate repayment of the loan. We investigate the relationship between the potential consequences of covenant violations and the tightness of covenants by first identifying firms that violated either the current ratio or Debt/EBITDA covenants, then examining whether violations are more likely to lead to bankruptcy, delisting due to performance, or the lenders not granting waivers when covenants are set tightly.

We focus on violations of financial covenants because using financial data we can determine when a violation is likely to have occurred. For the current ratio covenant sample, we identify potential covenant violations by comparing the level of the covenant variable to the minimum threshold required by the loan agreement. We refer to these as potential violations because of the measurement problems associated with determining the tightness of financial covenants and because loan agreement contains adjustments to the covenant level over the life of the loan.[34] Moreover, subsequent amendments to the loan agreement may change the covenant threshold required by the initial loan agreement. Amended loan agreements are unlikely to impose more restrictive covenant thresholds than the initial loan agreement, however. If a covenant appears to have been violated then we search SEC quarterly filings (i.e. 10-Qs and 10-Ks) for any discussion of covenant violations or defaults from the first quarter after loan inception up to three years following the violation or until the maturity date of the loan. For the Debt/EBITDA covenant sample, because the measurement error problem may potentially be more severe, we identify covenant violations by searching quarterly SEC filings for discussion of covenant violations for all of the loans in the sample. Unfortunately, in most cases when SEC filings indicate a covenant violation without providing a detailed discussion of the specific covenants that were violated. We assume, since the firm is out of compliance with either the current ratio or Debt/EBITDA covenant, the violation reported in the SEC filing pertains to Debt/EBITDA or current ratio covenants. If there is not a potential violation (i.e. the firms appeared to be in compliance with the Debt/EBITDA and/or the current ratio covenants according to our quarterly tightness measure) or if there is a potential violation but violation is not reported in the SEC filings, we assume the covenant was not violated.

We collected information on whether the firms in our sample filed for bankruptcy, were delisted because of poor performance, or had covenant violations that were not cured through a waiver or amendment to the loan agreement. We obtained this information in SEC filings discussing the covenant violation and from the CRSP delisting files. If violations are more costly when covenants are set loosely we expect that these so called bad outcomes are less frequent when firms violate tight rather than loosely set covenants.

Panel A of Table 9 provides estimates of a probit model relating the frequency of covenant violations to covenant tightness and firm risk characteristics.[35] Loans are defined as having tight covenants if there is tight Debt/EBITDA and/or current ratio covenants. We include in the probit models the firm leverage and the other credit risk measures used in our previous analyses. Table 9 shows that covenant violations are significantly more likely when covenants are set tightly and when more covenants are included in the loan agreement. This result is not particularly surprising since, as shown in Tables 4 and 5, credit risk and covenant tightness and intensity are positively related.

Of greater interest is the relationship between so called “bad outcomes” and covenant violations when covenants are set tightly. To examine this question we estimate a probit model relating the frequency of bad outcomes to covenant violations. The estimates of the probit model are presented in Panel B of Table 9.[36] Consistent with the argument that the costs of violation are inversely related to how tightly covenants are set at loan inception, we find a negative and statistically significant relationship between the frequency of bad outcomes and covenant violations for loans with tight covenants. While this result is consistent with the signaling story, another explanation is that financial condition of firms with tight covenants deteriorates less prior to a covenant violation and as a result covenant violations are less likely to be associated with bad outcomes. Nevertheless, when we include the change in the covenant variable from the initiation of the loan to the violation we find that loans with tight covenants are still associated with a lower likelihood of bad outcomes upon a covenant violation though at slightly lower levels of statistical significance. Moreover, not surprisingly, we find that the likelihood of a bad violation outcome is positively related to the deterioration in the financial condition of borrowers. Overall, these results are consistent with the argument that covenant violations are less costly when covenants are set tightly.

F. The stock market reaction to loan announcements

Our final empirical test involves examining the relationship between the stock price reaction to loan announcements and the tightness and intensity of the covenant structure in the loan agreement. We limit this analysis to loans in the current ratio and Debt/EBITDA samples so that we can measure covenant tightness. To identify loan announcements we searched Factiva new archives for a news report (including a company press release) for 30 days before and after the loan inception date. As the announcement date we use the earlier of the date of the press report or the loan inception date (although, since most of the press reports precede the inception date our results are similar if we use the announcement date). We were able to identify announcement dates for 415 of the loans in our sample or about one-third of current ratio and Debt/EBITDA covenant samples. This is similar to the proportion of the Dealscan loans with news announcements reported in a recent paper by Gonzalez, Houston and James (2006). For each announced loan we computed market adjusted announcement returns over a three-day window centered on the announcement date. We use three-day returns because we are uncertain as to the timing of the announcement and whether the press report is about a loan announcement made during trading the previous day. Market adjusted returns are simply the difference between the firm’s announcement day returns and the return on the CRSP value weighted index. Consistent with previous empirical studies of the announcement effects of bank loan agreements we find a statistically significant average three-day announcement return of 1.20 percent for the loans in our sample (The z statistic is 2.28).

If the selection of restrictive covenants conveys favorable private information regarding the prospects of the borrower we would expect to observe higher stock returns when loans with tight covenants are announced. To examine this issue we estimate a regression that relates announcement day returns to covenant tightness and controls for loan size and risk characteristics of the borrower. A loan is considered to have tight covenants if either a tight Debt/EBITDA or current ratio covenant was chosen. The results of this analysis are reported in Table 10. Consistent with the argument that tight covenants signal private information we find a positive and statistically significant relationship between announcement returns and the presence of tight covenants. The second specification in Table 10 also suggests that the relationship between the announcement returns and covenant tightness is positive and significant even after controlling for loan size and the risk characteristics of the borrower.

We also examine the relationship between announcement day returns, covenant tightness and the intensity of covenants. Including covenant intensity in the regression leads to a substantial decrease in the sample size because often times one or more of the components of the covenant index are not reported. We find a positive relationship between returns and the tightness of covenants and a negative and statistically significant relationship between returns and covenant intensity. These results suggest that while covenant tightness conveys positive private information selecting intensive covenants conveys negative information. This latter result may arise in part from the fact that covenant intensity is related to the credit risk characteristics of the borrower and because the private information conveyed by intensive covenants depends on the severity of information asymmetries.

To address these issues we interact the covenant intensity measure with our proxies for information asymmetries (as we did in Table 6). In Table 10 we report the results using analyst forecast errors as an information proxy although the results are similar if we use the dispersion in earning announcement returns as an information proxy. As shown in the last two columns of Table 10 we find a positive and significant relationship between announcement returns and covenant tightness. Moreover, while the coefficient on covenant intensity is negative and statistically significant we find the coefficient estimate on the interaction variable is positive and statistically significant at the 5 percent level. This finding is consistent with the results reported in Table 6. Specifically, the extent to which the choice of covenant intensity provides a signal of favorable private information concerning the quality of the borrower depends on the severity of information problems. In other words the signaling content of covenant intensity choice varies with the precision with which lenders can estimate credit risk based on observable firm characteristics.

V. Summary and Conclusions

An intuitively appealing explanation of the choice of tight or intensive covenants is that the choice conveys information about the borrower’s confidence in future performance. Testing the information content of covenant choice is challenging however because restrictive covenants are also used to control agency problems and are therefore likely to be correlated with observable credit risk characteristics. Testing the signaling hypothesis involves coming up with a methodology to control for differences in observable risk characteristics and identifying a reasonable proxy for the private information conveyed through covenant choice. In this paper, we test the signaling hypothesis by assuming that realized future performance is correlated with the borrowers private information at the time the loan in originated. We attempt to control for observable risk differences by including in a model of the determinants of covenant choices market and accounting based measures of credit risk. In the case of covenant tightness we also use a novel approach of examining covenant choice by clusters of firms formed based on similar levels of the financial variables on which the covenant is based.

Overall, we find evidence consistent with the signaling hypothesis. The most compelling, in our views, is the empirical evidence regarding the selection of tight covenants. In particular, consistent with a signaling story we find a positive relationship between the choice of tight covenants and improvements in future performance as measured by changes in the covenant variable. Moreover, we find that tight covenants are associated with incrementally lower borrowing costs for firms that select tight covenants. Finally, as further support of a signaling story we find that the stock price reaction to bank loan announcements is greater when the loan agreement contains tight covenants. Taken together this evidence suggests that the choice of tight covenants convey favorable private information concerning the future performance of the borrower.

The evidence concerning information content of intensive covenants is more complicated. Our results suggest that covenant intensity choice conveys favorable information when information asymmetries are the most severe. In particular, we find that the relationship between intensity and future performance depends on the severity of information problems. As information asymmetries increase the selection of intensive covenants is more frequently associated with improvements in performance. The stock returns associated with loan announcements is also consistent with this finding. Specifically, we find that the relationship between stock returns and the intensity of the covenants in the loan agreement are increasing in our proxy for information asymmetries.

Appendix A: Calculation of Covenant Tightness

1. Current Ratio Tightness

In order to control for the direct effect of the loan on a borrower’s current ratio, we form current ratio clusters based on the first post-loan fiscal quarter end current ratio of borrowing firms. Though this introduces a forward looking bias to our analysis, it is quite plausible to assume that pre-loan negotiations on financial covenants are based on pro-forma balance sheets considering the effect of the loan on financial ratios. Note that all our results hold when we use last pre-deal quarterly current ratio or average quarterly current ratio. Second, as suggested by previous studies (e.g., Dichev and Skinner (2002), Chava and Roberts (2006)) on current ratio covenant, about 10 percent of the time firms are immediately in violation of their current ratio covenant. Without access to exact covenant definitions, it is difficult to know whether these violations are due to measurement error or they are actual violations. We took several steps to minimize the measurement error in our current ratio covenant sample. First, we eliminated all loans where the ratio of covenant to current ratio greater than or equal to 1.20. All of our results are very similar when we delete loans with the ratio of covenant to current ratio above 1.10 or 1.00. Second, inspection of a sub-sample of loan contracts on Edgar indicates that non-GAAP adjustments to current ratio covenants are very common in loan contracts when the borrower has a current ratio below 1. Also, immediate violations mentioned before are the most common among these loans. Therefore, we eliminate all such loans. As a final step, we make a correction to the tightness classification from the cluster analysis. Specifically, in the top cluster (i.e. cluster of firms with current ratio above 3.50), unlike in any other cluster, there is a big variation of borrower current ratios (i.e. ranging between 3.50 and 9.30). This variation creates unexpected problems when a cluster-classification is used: For example, a borrower with a current ratio of 3.50 and covenant choice of 2.50 is classified as choosing a loose covenant while another borrower with a current ratio of 9.00 and a covenant choice of 2.75 is classified as choosing a tight covenant. Obviously, it is difficult to argue that this is a fair classification. Therefore, only for this top cluster, we re-classify all covenant choices below 2/3 of existing current ratio as loose.[37] For example, regardless of the cluster median covenant, a borrower that has a current ratio of 4.50 is classified as choosing a loose covenant as long as he agrees to a covenant below 3.00.

2. Debt/EBITDA Tightness

As in the current ratio sample, in order to control for the direct effect of the loan on a borrower’s debt levels, we form Debt/EBITDA clusters based on a first post-loan fiscal quarter end measure. In order to clean the sample, we eliminated all loans where the ratio of borrower Debt/EBITDA to the covenant is above 1.20 at loan inception.[38] In addition, we eliminated loans where the borrower had Debt/EBITDA below zero or above 20, because of concerns about severe measurement error and the difficulty of assessing performance changes when the initial Debt/EBITDA is negative or very high. As a final step, we re-assess the tightness of the loans in the bottom cluster. The cluster analysis classifies any borrower in the bottom cluster (i.e. Debt/EBITDA between 0 and 2.00) that chooses a covenant below 4.00 as choosing a tight covenant. For example, a borrower which has a Debt/EBITDA of 0.25 and a covenant of 3.00 is classified as choosing a tight covenant by the cluster analysis. This, obviously, is not a very compelling classification. To correct for this problem, in the bottom cluster only, we re-classify all covenant choices where the ratio of Debt/EBITDA to the covenant below 2/3 as loose.

Appendix B:

Ordered Probit Estimates of Credit Ratings

For firms without credit ratings, we estimate credit rations using an ordered probit model proposed by Blume, Lim, and Mackinlay (1998). We start the analysis by using the entire universe of Compustat firms with historical long-term credit ratings (i.e. senior credit rating before 1998) in the 1994-2004 period. Blume et al use credit ratings from Warga file, but only because Compustat senior credit ratings are not available in the earlier parts of their sample period. Because S&P ratings are available from Compustat for our entire sampling period we use ratings on Compustat.

I. Variables

The accounting ratios used to estimate ratings are: pretax interest coverage, operating income to sales, long-term debt to assets, and total debt to assets.[39] Following, Blume, Lim, and MacKinlay (1998) we use three-year averages of these ratios. Because pretax interest coverage is highly skewed and negative coverage ratios are not economically meaningful, we winsorize annual interest coverage ratios at 0 and 100. To control for the non-linearities in the relationship between interest coverage ratio and credit ratings we calculate four variables based on the three year average coverage ratio:

| |C1it |C2it |C3it |C4it |

|Cit Є [0,5) |Cit |0 |0 |0 |

|Cit Є [5,10) |5 |Cit - 5 |0 |0 |

|Cit Є [10,20) |5 |5 |Cit - 10 |0 |

|Cit Є [20,100] |5 |5 |10 |Cit - 20 |

Cit is the pretax interest coverage of firm i at the of calendar year t. Cjit represents the jth component of the coverage ratio as defined in the table above, where j = 1, 2, 3, 4.

We used natural log of inflation adjusted market capitalization to control for the effect of firm size on credit ratings. Also, using CRSP daily stock files, we calculated beta and residual standard deviation of common stock returns from the Scholes-Williams market model. For each calendar year we included beta and residual standard deviation of companies with at least 100 daily stock returns. We use the value-weighted CRSP index as the market index. To control for the variation in the cross-sectional averages of standard deviation of residuals over time, each year we divided standard deviation of residuals for each firm by cross-sectional averages of that year.

II. Model

We estimate the model in the same way Blume et al do. Our data panels are organized by calendar, not fiscal, year. In our pooled probit analysis we assume that alphas change over time (first year's alpha is set equal to zero), while the slope coefficients remain constant over our sampling period. Assuming heteroskedasticity we model the error terms as an exponential function of market capitalization. We assume that the alpha for the residual variance equation is zero. We estimate the model parameters using Maximum Likelihood Estimation (MLE).

Ordered Probit Model Estimates of Credit Ratings for the Panel Data, 1994 – 2004

The estimates are for the ordered probit model parameters using a panel data sample of 15,269 observations from 1994 through 2004. To conserve space year fixed effects are omitted. The lower boundaries for rating category parameters are the estimates of the partition parameters for the rating categories. The variance parameter is the estimate of the coefficient associated with the market value of equity when the variance of the disturbances is modeled as function of the deflated market value of equity. The standard errors are calculated under the assumption that the disturbances are uncorrelated.

|  |  |  |  |  |

| | |Standard | |P-value |

|  |Coefficient |Error |t Value |Approx r > |t| |

| | | | | |

|Beta | | | | |

|Pretax Interest Coverage | | | | |

| Max (5, Coverage) |0.069 |0.005 |14.32 |0.000 |

| Max (0, Coverage - 5) |0.013 |0.004 |3.44 |0.001 |

| Max (0, Coverage - 10) |0.013 |0.002 |5.32 |0.000 |

| Max (0, Coverage - 20) |-0.002 |0.001 |-2.90 |0.004 |

|Operating Margin |0.052 |0.006 |9.02 |0.000 |

|LT Debt Leverage |-1.619 |0.073 |-22.32 |0.000 |

|Total Debt Leverage |0.749 |0.053 |14.11 |0.000 |

|Market Value |0.213 |0.008 |27.06 |0.000 |

|Market Model Beta |-0.150 |0.010 |-14.28 |0.000 |

|Standard Error |-1.017 |0.033 |-30.59 |0.000 |

| | | | | |

|Lower Boundary for | | | | |

|Rating Category | | | | |

|C |-2.113 |0.077 |-27.31 |0.000 |

|CC-CCC |-1.412 |0.057 |-24.90 |0.000 |

|B |0.052 |0.040 |1.32 |0.186 |

|BB |0.834 |0.050 |16.82 |0.000 |

|BBB |1.529 |0.067 |22.93 |0.000 |

|A |2.217 |0.089 |24.90 |0.000 |

|AA |2.675 |0.107 |24.90 |0.000 |

| | | | | |

|Variance Parameter | | | | |

|Market Value |-0.188 |0.009 |-21.37 |0.000 |

|  |  |  |  |  |

References

Beneish, Messod D., and Eric Press, Cost of Technical Violation of Accounting-Based Debt Covenants, The Accounting Review, 1987, volume 68.2., pages 233-257.

Besanko, David and Anjan V. Thakor, Collateral and Rationing: Sorting Equilibria in Monopolistic and Competitive Credit Markets, International Economic Review, 1987, volume 28.3., pages 671-689.

Billett, Matthew T., Tao-Hsien Dolly King, and David C. Mauer, Growth Opportunities and the Choice of Leverage, Debt Maturity, and Covenants, Journal of Finance, forthcoming.

Blume, Marshall E., Felix Lim, A. Craig Mackinlay, The Declining Credit Quality of U.S. Corporate Debt: Myth or Reality?, Journal of Finance, 1998, volume 53.4., pages 1389-1413.

Booth, James R., and Lena Chua Booth, Loan Collateral Decisions and Corporate Borrowing Costs, Journal of Money, Credit, and Banking, 2006, volume 38.1., pages 67-90.

Bradley, Michael, and Michael R. Roberts, The Structure and Pricing of Corporate Debt Covenants, Duke University Working Paper, 2004.

Brav, Alan, Roni Michaely, Michael Roberts, Rebecca Zarutskie, Evidence on the Tradeoff between Risk and Return for IPO and SEO Firms, Duke University Working Paper, 2005.

Carey, Mark, and Mark Hrycay, Credit Flow, Risk, and the Role of Private Debt in Capital Structure, Federal Reserve Working Paper, 1999.

Chan, Yuk-Shee, and George Kanatas, Asymmetric Valuations and the Role of Collateral in Loan Agreements, Journal of Finance, 1985, volume 17.1., pages 84-95.

Chava, Sudheer, and Michael R. Roberts, Is Financial Contracting Costly? An Empirical Analysis of Debt Covenants and Corporate Investment, University of Houston Working Paper, 2006.

Chen, Kevin C. W., and K. C. John Wei, Creditors’ Decision to Waive Violations of Accounting-Based Covenants, The Accounting Review, 1993, volume 68.2., pages 218-232.

Christie, A., On Cross-Sectional Analysis in Accounting Research, Journal of Accounting and Economics, 1987, volume 9, pages 231-59.

Dessein, Wouter, Information and Control in Ventures and Alliances, Journal of Finance, 2005, volume 60.5., 2513-2549.

Dichev, Illia D., and Douglas J. Skinner, Large-Sample Evidence on the Debt Covenant Hypothesis, Journal of Accounting Research, 2002, volume 40.4., pages 1091-1123.

Dierkens, N., Information Asymmetry and Equity Issues, 1991, Journal of Financial and Quantitative Analysis, volume 26, pages 181-199.

Drucker, Steven and Manju Puri, On Loan Sales, Loan Contracting, and Lending Relationships, Columbia University Working Paper, 2006.

Elton, Edwin J., Martin J. Gruber, and Mustafa N. Gultekin, Professional Expectations: Accuracy and Diagnosis of Errors, Journal of Financial and Quantitative Analysis, 1984, volume 19.4., pages 351-363.

Gerleanu, Nicolas, and Jeffrey Zwiebel, Design and Renegotiation of Debt Covenants, University of Pennsylvania Working Paper, 2005.

Gonzalez, Laura, Joel Houston and Christopher James, Dogs that Bark: What Firms Announce Loan Agreements, University of Florida Working Paper, 2006.

Guner, A. Burak, Loan Sales and the Cost of Corporate Borrowing, Review of Financial Studies, forthcoming.

Heckman, James, Sample Selection Bias as a Specification Error, Econometrica, 1979, volume 47.1., pages 153-162. 

Jensen, Michael C. and Meckling, William H., Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure, Journal of Financial Economics, 1976, volume 3.4., pages 305-360.

Jimenez, Gabriel, Vicente Salas, and Jesus Saurina, Determinants of Collateral, Journal of Financial Economics, forthcoming.

Krishnaswami, S. and V. Subramaniam, Information Asymmetry, Valuation, and the Corporate Spin-off Decision, Journal of Financial Economics, 1999, volume 53, pages 73-112.

Lee, Lung-Fei, Unionism and Wage Rates: A Simultaneous Equations Model with Qualitative and Limited Dependent Variables, International Economic Review, 1978, volume 19.2., pages 415-433. 

Leftwich, Richard, Accounting Information in Private Markets: Evidence from Private Lending Agreements, The Accounting Review, 1983, volume 58, pages 23-42.

Malitz, I., On Financial Contracting: The Determinants of Bond Covenants, Financial Management, 1986, volume 15, pages 18-25.

Moerman, Regina Wittenberg, The Impact of Information Asymmetry on Debt Pricing and Maturity, University of Chicago Working Paper, 2005.

Myers, Stewart C., Determinants of Corporate Borrowing, Journal of Financial Economics, 1977, volume 5.2., pages 147-175.

Nash, Robert C., Jeffry M. Netter, and Annette B. Poulsen, Determinants of Contractual Relations between Shareholders and Bondholders: Investment Opportunities and Restrictive Covenants, Journal of Corporate Finance, 2003, volume 9, pages 201-232.

Norton, Edward, Hua Wang, and Chunrong Ai, Computing Interaction Effects and Standard Errors in Logit and Probit Models, 2004, STATA Journal, volume 4, pages 103–116.

Rajan, Raghuram, and Andrew Winton, Covenants and Collateral as Incentives to Monitor, Journal of Finance, 1995, volume 50.4., pages 1113-1144.

Smith, Clifford W., A Perspective on Accounting-Based Covenant Violations, The Accounting Review, 1993, volume 68.2., pages 289-303.

Smith, Clifford W., and Jerold B. Warner, On Financial Contracting: An Analysis of Bond Covenants, Journal of Financial Economics, 1979, volume 7, pages 117-161.

Sufi, Amir, Bank Lines of Credit in Corporate Finance: An Empirical Analysis, University of Chicago Working Paper, 2006.

Sweeney, Amy P., Debt-Covenant Violations and Managers’ Accounting Responses, Journal of Accounting and Economics, 1994, Volume 17, 281-308.

Zimmerman, C., An Approach to Writing Loan Agreement Covenants, Journal of Commercial Bank Lending, 1975, volume 58, pages 2-17.

Table 1

Summary Statistics of Bank Loans 1995-2001

The table below presents borrower and loan characteristics for a sample of dollar denominated loans of non-financial U.S. firms with publicly traded common stocks. The period of analysis is 1995-2001 and the source of loan information is Loan Pricing Corporation’s (LPC) Dealscan database. Loans whose maturity (at loan inception) is less than one year are excluded from the analysis. Borrower financials are as of the last pre-loan fiscal year and are obtained from Compustat. Information on borrower common stocks is from CRSP. Analyst forecast errors are calculated by using analysts’ earnings forecasts in I/B/E/S. Market capitalization equals to the number of shares outstanding [Data 25] times fiscal year and closing stock price [Data 199]. The data in brackets are annual Compustat data item numbers. Another measure of borrower size is Total assets which equals [Data 6]. Estimated credit rating is an estimate of borrowing firm credit quality from the ordered probit model of Blume, Lim, and MacKinlay (1998). Details of the estimation are in Appendix A. Z-score is an alternative proxy we use to measure borrower creditworthiness and is defined as in Sufi (2006). EBITDA / Sales is an operating performance measure that is equal to [Data 13] / [Data 12]. Standard deviation of EBITDA / Sales is the five-year pre-loan volatility of [Data 13] / [Data 12]. Pre-loan leverage of borrowers is Total Debt / Assets, equals [Data 9 + Data 34 + Data 104] / [Data 6]. A measure of borrower fixed assets is PPE / Assets, which is measured by dividing [Data 30] to [Data 6]. A ratio of leverage to operating profits is Total Debt / EBITDA, which equals [Data 9 + Data 34 + Data 104] / [Data 13]. A proxy for borrower investment opportunities and potential risk shifting activities is (Capital expenditures + R&D) / Sales, which is equal to [Data 128 + Data 46] / Data 12. Liquidity is measured by Current ratio, which is the ratio of current assets [Data 4] to current liabilities [Data 5]. Age is the number of years between the loan inception date and the first date borrower stocks appear on CRSP files. Analyst forecast error is calculated by taking the absolute value of the difference between mean consensus analysts forecast of borrower earnings regarding the last fiscal year before loan inception and realized borrower earnings, and dividing this difference by borrower common stock price at the beginning of the month of analyst forecasts. We use analyst forecasts issued one month before the fiscal year end of interest to calculate forecast errors. Standard deviation of earnings announcement returns is the volatility of abnormal stock returns on borrower common stock (-1, +1) trading days around quarterly earnings announcement dates as reported by Compustat. The market index used to calculate the abnormal returns is value-weighted CRSP index. We calculate this volatility measure only for borrowers with more than three pre-loan quarterly earnings announcement dates. All-in-drawn spread is calculated and reported by Dealscan as the total borrowing cost of the drawn portion of a loan over and above LIBOR. Maturity is the number of months between loan inception and expiration date. Loan amount / Total assets is the maximum amount available to the borrower scaled by borrower’s total assets. Covenant intensity index equals the sum of six covenant indicators (collateral, dividend restriction, more than 2 financial covenants, asset sales sweep, equity issuance sweep, and debt issuance sweep) and hence ranges between 0 and 6. When the indicator for one of the six covenants is missing the index is set equal to missing and the loan is excluded from the analysis. Number of lenders indicates the number of lenders in the loan syndicate. Syndicated loan equals 1 if loan is syndicated (includes more than one lender); 0 otherwise. Secured is an indicator variable which equals 1 if the loan is backed by collateral. Sweep is an indicator variable which equals 1 if the loan includes a(n) asset sales, debt issuance, equity issuance, or excess cash flow sweep. Performance pricing equals 1 if loan pricing is tied to borrower performance; and 0 otherwise. The summary statistics are reported for three classes of loans: overall loan sample, sample of loans with a Debt/EBITDA covenant, sample of loans with a current ratio covenant. While summary statistics for the overall sample include loans with coverage and liquidity covenants the statistical tests are based on comparing borrowers/loans in the coverage and liquidity samples to firms without these covenants in their loan agreements. We provide loan level summary statistics. We conduct two-tailed mean and median comparison tests to investigate whether there is any difference between borrower and loan characteristics of covenant samples and non-covenant benchmarks. a, b, and c, indicate significance at the 1%, 5%, and 10% levels, respectively.

|  |Overall |Debt/EBITDA covenant |Current ratio covenant |

|  |N=11,660 |N=3,739 |N=956 |

|  |Mean |Median |

|  |Tight | |Loose |Tight | |Loose |

| |N=347 | |N=212 |N=421 | |N=535 |

|  |Mean |  |Median |  |Mean |

|  |Sample N=238 | |Sample N= 1,126 | |Sample N= 1,387 |

|  |Mean |  |Median |

|  |N |Mean |  |

|  |

|  | |Age | |Standard deviation of | |Analyst forecast error |

| | | | |quarterly earning | | |

| | | | |announcement returns | | |

|  | |Coef. |  |Std Err | |Coef. |

|Pseudo R-squared | |0.156 | |0.159 | |0.151 |

|  |

|  | |Age | |Standard deviation of | |Analyst forecast error |

| | | | |quarterly earning | | |

| | | | |announcement returns | | |

|  | |Coef. |  |Std Err | |Coef. |

|Pseudo R-squared | |0.271 | |0.296 | |0.253 |

|  |

|  | |Age | |Standard deviation of | |Analyst forecast error |

| | | | |quarterly earning | | |

| | | | |announcement returns | | |

|  | |Coef. |  |Std. Err. | |Coef. |

|Number of observations | |3081 | |3024 | |2657 |

|Pseudo R-squared | |0.127 | |0.128 | |0.142 |

|  |

|  | |Age | |Standard deviation of | |Analyst forecast error |

| | | | |quarterly earning | | |

| | | | |announcement returns | | |

|  | |Coef. |  |Std. Err. | |Coef. |

|Number of observations | |3369 | |3305 | |2879 |

|Pseudo R-squared | |0.125 | |0.126 | |0.139 |

|  |

|  | |Age | |Standard deviation of | |Analyst forecast error |

| | | | |quarterly earning | | |

| | | | |announcement returns | | |

|  | |Coef. |  |Std. Err. | |Coef. |

|Number of observations | |2968 | |2910 | |2560 |

|Pseudo R-squared | |0.128 | |0.129 | |0.14 |

|  |  |  |

|  |Tight | |Loose |Tight | |Loose |

|  |Coef. |  |Std. Err. | |Coe|  |

| | | | | |f. | |

|Adjusted R-squared |0.464 | |0.423 |0.601 | |0.675 |

|  |

|  |Overall Sample |

|  |Intense | |Not Intense |

|  |Coef. |  |Std. Err. | |Coef. |  |Std. Err. |

|  |  | | | | | |  |

|Inverse Mill’s Ratio |-0.458 |b |0.223 | |0.809 |b |0.395 |

|  |  | | | | | | |

|Loan characteristics |  | | | | | | |

| Maturity (months) |0.000 | |0.001 | |-0.014 |b |0.004 |

| Deal amount / Total assets |-0.007 | |0.019 | |0.024 | |0.211 |

| Dummy: Syndicated loan |0.132 | |0.101 | |-- | |-- |

|  |  | | | | | | |

|Borrower credit risk characteristics: |  | | | | | | |

| Log (Market capitalization) |-0.073 |a |0.009 | |-0.062 |c |0.038 |

| EBITDA / Sales |-0.430 |a |0.083 | |-0.422 | |0.327 |

| Total debt / Total assets |0.311 |a |0.041 | |-0.314 | |0.306 |

| Power, Plant, and Equipment / Total assets |-0.003 | |0.191 | |-0.501 | |1.170 |

| (Capital expenditures + R&D) / Sales |0.225 |b |0.093 | |0.012 | |0.473 |

| Volatility of EBITDA / Sales |0.038 | |0.269 | |1.751 |c |0.906 |

|  |  | | | | | | |

|Number of observations |1229 | |226 |

|Adjusted R-squared |0.297 | |0.499 |

|  |  |  |  |  |  |  |  |

Table 8 Cont’d

|Panel B. Intense includes BR Index = 5, 6 and not intense includes BR Index = 0, 1 |

|  |Overall Sample |

| |Intense | |Not Intense |

|  |Coef. |  |Std. Err. | |Coef. |  |Std. Err. |

|  |  | | | | | |  |

|Inverse Mill’s Ratio |-0.612 |a |0.167 | |0.486 |c |0.277 |

|  | | | | | | | |

|Loan characteristics | | | | | | | |

| Maturity (months) |0.000 | |0.001 | |-0.013 |a |0.002 |

| Deal amount / Total assets |-0.020 | |0.019 | |0.082 | |0.124 |

| Dummy: Syndicated loan |0.120 | |0.100 | |-0.376 |c |0.208 |

|  | | | | | | | |

|Borrower credit risk characteristics: | | | | | | | |

| Log (Market capitalization) |-0.069 |a |0.008 | |-0.162 |a |0.034 |

| EBITDA / Sales |-0.402 |a |0.082 | |-0.280 | |0.308 |

| Total debt / Total assets |0.276 |a |0.039 | |0.045 | |0.189 |

| Power, Plant, and Equipment / Total assets |0.016 | |0.191 | |-0.407 | |0.520 |

| (Capital expenditures + R&D) / Sales |0.214 |b |0.093 | |0.151 | |0.335 |

| Volatility of EBITDA / Sales |-0.041 | |0.269 | |2.351 |a |0.826 |

|  | | | | | | | |

|Number of observations |1229 | |612 |

|Adjusted R-squared |0.304 | |0.550 |

|  |  |  |  |  |  |  |  |

Table 9

Restrictiveness of Covenants, Covenant Violations, and Outcome of Violations

The table below presents probit regressions that explain the relationship between restrictiveness of loan covenants at loan inception and the probability of subsequent covenant violations as well as violation outcomes. Panel A presents probit regressions (marginal effects reported) that explain the probability of covenant violations for a sample of loans that include a current ratio or a debt / ebitda covenant (or both). The dependent variables in Panel A are indicator variables that equal to 1 if there is a covenant violation during the period from loan inception to years 1 and 3. Panel B presents probit regressions (marginal effects reported) that explain the relationship between ex-ante restrictiveness of covenants and the outcome (e.g. lender response) of covenant violations. The dependent variable for the probits in Panel B is an indicator variable that equals 1 if the borrower gets delisted from CRSP due to poor performance, declares bankruptcy, or terminates the loan because of its inability to comply with loan covenants. Covenant violations for each loan are hand collected from quarterly borrower filings (10-Qs and 10-Ks) on Edgar. First, using Compustat we measured quarterly covenant slack for each of the loans in our current ratio and debt/ebitda samples for the first three years after loan inception and identified “potential” covenant violators. Second, we tracked the condition of each of these “potential” violators on Edgar from loan inception until the minimum of loan expiration (e.g. maturity or termination), borrower delisting, and the end of 12th quarter after loan inception. Violations in this table are violations reported in SEC filings, not the violations identified by our slack measure. Violation outcome information in Panel B is jointly collected from the SEC filings of covenant violators and CRSP delisting files. Covenant intensity equals the sum of six covenant indicators (collateral, dividend restriction, more than 2 financial covenants, asset sales sweep, equity issuance sweep, and debt issuance sweep) and hence ranges between 0 and 6. When the indicator for one of the six covenants is missing the index is set equal to missing and the loan is excluded from the analysis. Robust standard errors are reported. Letters a, b, and c indicate significance at 1%, 5%, and 10% levels, respectively.

|Panel A. Relationship Between Covenant Violation and Restrictiveness of Covenants |

|Dependent variable |Violation year 1 | |Violation year 3 |

|  |Coef. |  |Std. Err. | |Coef. |  |Std. Err. |

|Number of observations |712 | |712 | |712 | |712 |

|Pseudo R-squared |0.072 | |0.110 | |0.053 | |0.095 |

|  |

|Dependent variable = 1 if company is delisted, bankruptcy or violations of covenant not cured; 0 otherwise. |

|  |Coef. |  |Std. Err. | |Coef. |  |Std. Err. |

|  | | | | | | |  |

|Dummy: Violation |0.387 |a |0.105 | |0.376 |a |0.106 |

|Dummy: Tight covenant |0.009 | |0.017 | |0.006 | |0.017 |

|Covenant intensity index | | | | |0.007 | |0.004 |

|Tight covenant * Violation |-0.049 |a |0.013 | |-0.047 |a |0.013 |

|Log (market capitalization) |-0.027 |a |0.005 | |-0.027 |a |0.005 |

|EBITDA / Sales |-0.070 | |-0.064 | |-0.078 | |0.064 |

|Total debt/ Total assets |0.135 |b |0.027 | |0.121 |a |0.026 |

|PPE / Total assets |0.143 | |0.990 | |0.145 | |0.143 |

|(Capital expenditures + R&D) / Sales |-0.010 | |0.053 | |0.001 | |0.053 |

| | | | | | | | |

|Number of observations |712 | |712 |

|Pseudo R-squared |0.223 | |0.229 |

|  |  |  |  |  |  |  |  |

Table 10

Announcement Returns and Covenant Structure: Weighted Least Squares Regressions

The table below presents weighted least squares regressions that explain the relationship between restrictiveness (i.e. tightness and intensity) of loan covenants and loan announcement returns using a sample of bank loan announcements by non-financial U.S. public firms during the 1995-2001 period. We focus only on loans with a current ratio or a debt / ebitda covenant (or both). The source of loan information is Loan Pricing Corporation’s (LPC) Dealscan database. Loans with maturity (at loan inception) less than a year are excluded from the analysis. Loan announcement dates are hand searched at Factiva news archives within one-month of loan inception date. Announcement returns are calculated at the (-1, +1) trading day window centered on loan announcement date by using the CAR approach using the value-weighted CRSP index as the market index. Tight is an indicator variable that equals to 1 if the loan has a tight current ratio or debt / ebitda covenant; 0 otherwise. Covenant intensity equals the sum of six covenant indicators (collateral, dividend restriction, more than 2 financial covenants, asset sales sweep, equity issuance sweep, and debt issuance sweep) and hence ranges between 0 and 6. When the indicator for one of the six covenants is missing the index is set equal to missing and the loan is excluded from the analysis. Analyst forecast error is calculated by taking the absolute value of the difference between mean consensus analysts forecast of borrower earnings regarding the fiscal year end before loan inception and realized borrower earnings, and dividing this difference by borrower common stock price at the beginning of the month of analyst forecasts. We use analyst forecasts issued one month before the fiscal year end of interest to calculate forecast errors. Each regression includes an intercept, industry, year, loan type, and loan purpose fixed effects (unreported). In the estimation process each observation is weighted by the inverse of residual stock return volatility. Robust standard errors are reported. Letters a, b, and c indicate significance at 1%, 5%, and 10% levels, respectively.

| |Dependent variable: (-1, +1) Market-adjusted return around the loan announcement |

| |Coef. | |Std. Err. | |Coef. | |Std. Err. |

|Adjusted R-squared |0.031 | |0.058 | |0.162 | |0.066 |

| | | | | | | | | | | | | | | | |

-----------------------

[1] Another reason tightness of covenants may be related to subsequent improvements in operating performance is that tight covenants effect a manager’s financial reporting decisions (See Dichev and Skinner (2002)).

[2] See for example, Jensen and Meckling (1976), Smith and Warner (1979), Myers (1977), and Smith (1993).

[3] See for example, Booth and Booth (2006), Bradley and Roberts (2004), Malitz (1986), and Nash, Netter and Poulsen (2003). Investigating the choice of bank loan covenants, Bradley and Roberts (2004) find that growth firms get more restrictive covenants. However, Nash, Netter and Poulsen (2003) find that public debt contracts of high growth firms are less likely to include restrictions on dividends and debt issuance.

[4] Jimenez, Salas and Saurina (2006) make this argument in the context of the choice of securing a loan.

[5] A commercial lending text describes the criteria for setting target thresholds as follows: “The thresholds are set based on the company’s historical and projected performance, as well as the banker’s determination of key performance levels required to provide adequate protection”. Principles of Loan Structure 3-6 Edge Development Group 2006 available at .

[6] For example, asset sweeps require that a portion of the proceeds form asset sales to be used to pay down the loan and collateral requirements provide the lender title to the assets conditional on default.

[7] A large majority of the loan data in LPC are from SEC filings (13-Ds, 14-Ds, 13-Es, 10-Ks, 10-Qs, 8-Ks and Registration Statements (e.g. S-series filings)). According to Carey and Hrycay (1999), from 1995 onward, Dealscan contains the “large majority” of sizable commercial loans in the US. Over 90 percent of all Dealscan loans in this period are syndicated (i.e. underwritten and financed by a group of banks, insurance companies, and other financing entities).

[8] We hand matched Dealscan to CRSP and Compustat. Dealscan does not report standard borrower identifiers such as a CUSIP number. Therefore, we identify CRSP records of Dealscan borrowers by using ticker, name, and industry matching. In order to verify the accuracy of our Dealscan-CRSP matches we compare borrower sales and locations (i.e. state) from Dealscan to those in the CRSP-Compustat Merged Database. A more detailed description of our linking procedure is available upon request.

[9] Coverage covenants include at least one of the covenants below: interest coverage, fixed charge coverage, debt / EBITDA, senior debt / EBITDA, and debt service coverage. The liquidity covenants include current ratio or quick ratio covenants. The leverage covenant group consists of leverage, senior leverage, debt / equity, debt / tangible net worth, and loan / value covenants. The net worth covenant group includes loans with net worth and tangible net worth covenants. Capital expenditure covenants that restrict capital expenditures (typically limiting expenditures based on operating cash flows). A number of financial covenants contain a trend, also called a “build up”, which changes the covenant threshold to make the covenant more (or less) restrictive over time. For example, a leverage covenant might stipulate a maximum leverage of 50 percent at loan inception, decreasing to 40 percent at the end of first year, and 30 percent at the end of second year. Dealscan reports the initial covenant level as well as the highest (or lowest) covenant thresholds over the life of the loan. Detailed covenant schedules are available only for a limited number of loans with Tearsheets

[10] These frequencies are quite similar to those reported by Sufi (2006). He finds that 49 percent of the loans in Dealscan include Debt/EBITDA covenants and 46 percent include net worth covenants.

[11] For more detailed discussion of these covenants, see Bradley and Roberts (2004).

[12] See for example Smith and Warner (1979)

[13] According to Dichev and Skinner (2002) net worth covenants also have standard definitions. We do not examine net worth covenants for two reasons. First, the restrictiveness of a net worth covenant is jointly determined by where the base covenant, net income build up, equity issuance build up, and other build ups are set at. Because restrictiveness of the base covenant is typically inversely related to the restrictiveness associated with the buildups, it is difficult to measure the “net” tightness of a net worth covenant. Second, it is hard to come up with an appropriate operating performance benchmark for the net worth covenant. In particular, major changes in net worth are often associated with a borrower’s pay out and equity issuance decisions which are not necessarily correlated with improvements (or deteriorations) in a borrower’s operating performance.

[14] The borrower’s choice is further restricted by the fact that current ratio covenants are not continuous but are typically clustered at a few discrete numbers. The most common current ratio covenant clusters below 1.50 are 1.00 and 1.25.

[15] Appendix A contains a detailed description of how the current ratio and Debt/EBITDA covenants were defined as well as a discussion of how we address obvious measurement problems that arose.

[16] Note, however, that our Debt/EBITDA covenant tightness proxy may suffer from a measurement error problem to the extent that the covenant definitions in Tearsheets are incomplete. Any measurement error in this case, however, is likely to create attenuation bias on our empirical results, which makes it more difficult draw strong statistical inferences.

[17] The response of lenders to covenant violations varies. Frequently, covenant violations involve waivers or a renegotiating the terms of the loan agreement and not a demand for repayment or termination of the loan agreement. Nevertheless the results of previous empirical studies suggest that covenant violations are on average costly (See for example, Chen and Wei (1993) Beneish and Press (1993) and Chava and Roberts (2006) and Sufi (2006)).

[18] Jimenez et al (2006) follow a similar approach. An alternative approach would be to estimate a model of the determinants of future performance with covenant tightness and intensity as explanatory variables. When we use this approach we find a positive relationship between future performance and covenant tightness.

[19] Borrowers with CRSP delisting codes equal to 400-499 or 550-599 are classified as poor performers. Including firms with delisting codes equal to 500-549 does not affect any of our results.

[20] Also, note that Table 1 provides summary statistics for all loans with Debt/EBITDA covenants. As discussed above, in our empirical analysis, we use a subset of these loans with Tearsheets information. As Dichev and Skinner (2002) note loans with Tearsheets are larger “bellwether” loans taken out by larger borrowers.

[21] The ordered probit model assumes that credit ratings are a function of borrower’s interest coverage ratio, leverage, operating margin, market value of equity as well as beta and residual volatility of stock returns. We estimate the model using senior S&P credit ratings and panel data from 1994 to 2004 as reported in Compustat. Details concerning the estimation are in Appendix B.

[22] This also makes borrowers with current ratio covenants more vulnerable to liquidity shocks and leads them to hold significantly more cash than other borrowers. Consistent with this view, Sufi (2006) finds that firms with lower cash flows rely more heavily on cash and hence hold more cash out of their cash flows.

[23] Analysts’ earnings forecasts are obtained from IBES, which reports monthly summary statistics of analysts’ earnings forecasts for each firm with coverage.

[24] Elton et al (1984) investigate the determinants of forecast errors for a wide cross-section of firms and find that there are three components of these errors: economy-wide factors, industry-wide factors, and firm-specific factors. They show that 84 percent of the forecast errors during the month preceding fiscal year ends may be attributed to firm-specific factors, which implies that forecast errors at this time is a particularly good proxy for the level of asymmetric information about a firm’s cash flows.

[25] The quarterly earnings announcement dates are obtained from Compustat. We use value-weighted CRSP market index to measure market adjusted abnormal returns at (-1, +1) trading days centered on quarterly earnings announcements during the five years preceding loan inception. We then calculate the volatility of these abnormal returns for firms with at least four quarterly earnings announcement returns.

[26] Because the “true” incorporation dates of our borrowers are not available in electronic form, we use the number of months between loan inception and the first CRSP listing date of a borrower as a proxy for age. All of the results reported in this paper are very similar when we replace CRSP listing date with the first year in Compustat that the firm’s share price is available.

[27] Sufi (2006) argues that firms with high cash flow volatility may prefer to avoid covenants based on cash flows. Our finding that, conditional on the inclusion of a cash flow covenant, firms with high cash flow volatility choose less restrictive cash flow covenants seems to support and extend his findings.

[28] As a robustness check, in both univariate and multivariate performance tests we include only loans that were active as of the period of interest. For most loans, the effective maturity or termination dates are available from deal and facility remark columns in Dealscan. All our results are robust to this alternative methodology.

[29] We report industry adjusted changes to control for industry wide changes in performance. Our results are similar if we use unadjusted changes.

[30] The model of covenant intensity choice is estimated using a Poisson regression because covenant intensity is a non-negative count variable. Our results are similar using OLS regressions.

[31] The results are similar to the ones reported in Table 6 for performance measured over one, two and three year horizons.

[32] Jimenez et al (2006) make the argument concerning collateral as a signal of quality.

[33] To save space we do not report the estimates of the first stage probit model. The estimated relationships between tightness (intensity) and loan, borrower risk characteristics and our information proxies are similar to those reported in Tables 5 and 6. When estimating the first step probit model we do not include the future performance measures.

[34] To account for some of these changes for we obtained information on covenant adjustment schedules from Tearsheets and SEC filings (which often contain as an appendix the loan agreement). Since our Debt/EBITDA sample consists of loans with Tearsheets information, we have quarterly covenant schedules for all the loans in the Debt/EBITDA sample. Only about 10 percent of the loans in the current ratio sample have covenant schedules that change over time. Dealscan does not provide a detailed schedule of current ratio covenants. However, it does report whether there is an increasing, decreasing, or fluctuating trend in the covenant as well as the highest (or lowest) value that the covenant can take over the life of the loan. If we can not find a detailed covenant schedules from the sources in the Tearsheets or SEC 10-K filings and Dealscan indicates the current ratio covenant changes over the life of the loan, we linearly interpolate the covenant thresholds over the projected life of the loan. Chava and Roberts (2006) use this approach as well.

[35] In order to use a uniform sample of loans in alternative specifications and better evaluate the marginal effect of variables included in each specification, we restrict the covenant violation analysis reported in Table 9 only to loans with a non-missing covenant intensity index. The results are very similar when we include all loans.

[36] It is important to note that the marginal effects and standard errors of the interacted variables in Panel B of Table 9 are only suggestive. Because of the non-linear nature of the estimation procedure, it is not possible to compute point estimates for the interaction terms when using probit. We get very similar results when we compute corrected marginal effects and standard errors as suggested by Norton, Wang, and Ai (2004). We do not report these estimates for brevity.

[37] The results remain the same for a wide range of adjustment levels. Also, imposing the same adjustment to other cluster does not lead to any re-classifications.

[38] Note that Debt/EBITDA covenant is a maximum ratio covenant, unlike current ratio covenant which is a minimum ratio covenant. Therefore, borrowers in the Debt/EBITDA sample are in violation of the covenant when their Debt/EBITDA exceeds the covenant specified by their loan contract.

[39] The pretax interest coverage is defined as the ratio of [operating income after depreciation (178) + interest expense (15)] to [interest expense (15)], where the numbers in parentheses are Compustat annual data item numbers. The ratio of operating income to sales is defined as [operating income before depreciation (13)] to [Net sales (12)]. The ratio of long-term debt to assets is defined as [long-term debt (6)] to [assets (6)]. The ratio of total debt to assets is defined as [long-term debt (6) + debt in current liabilities (34) + average short-term borrowings (104)] to [assets (6)].

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download