The Eects of Convertible Debt on Debt Contracting

Transcription

1 The Eects of Convertible Debt on Debt Contracting Richard Carrizosa 1 Abstract Current reporting requirements for convertible debt assign the value of the conversion option to debt, and as a result distort both the rm's cost of debt and measures of debt and equity. I test whether these distortions aect two aspects of debt contracting: pricing and covenant selection. I analyze a large sample of private debt contracts and nd that after controlling for rm, issue, and macroeconomic risk factors, interest rates for borrowers with outstanding convertible debt are lower than those for borrowers with no convertible debt. The result is stronger for high credit risk rms, and rms that rely more heavily on convertible debt nancing. Also, the results hold after controlling for the higher growth options, lower cash interest payments, and lower relative seniority generally associated with convertible debt issuance. These ndings are consistent with creditors taking into account the equity characteristics of convertible debt when determining loan pricing. Lastly, I analyze the eects of post-loan-issuance increases in covenant slack caused by conversion, and nd that loans to rms with convertible debt at loan issuance have more nancial covenants, specically those covenants that rely on debt and equity measures. Overall, I nd that lenders do not rely solely on nancial reports, but rather incorporate additional information when measuring rm credit risk, and designing loan contracts for borrowers with convertible debt. address: rcarrizo@stern.nyu.edu (Richard Carrizosa) 1 New York University. This paper is based on a chapter of my thesis. I wish to thank my thesis advisors for their guidance and valuable comments: Daniel Cohen, William Greene, Baruch Lev (Chair), Stephen Ryan, and Rangarajan Sundaram. This paper has also beneted from discussions with Andre DeSouza, and Lucile Faurel. Preprint submitted to Elsevier January 13, 2010

2 1. Introduction Convertible debt is a hybrid nancing instrument that settles as either debt or equity at the option of the security holder. Current U.S. accounting rules require convertible debt to be reported entirely as a liability on the balance sheet, which fails to recognize the distinct equity characteristics of the instrument. As a result, balance sheet measures overstate rm protability, and understate rm solvency. I focus on solvency measurement and examine whether the failure of current accounting rules to reect the complex nature of convertible debt aects the decisions of creditors. Specically, I test whether the overstatement of leverage aects lenders' assessments of rm credit risk, and in turn the interest rates charged for new loans to rms with outstanding convertible debt at the time of loan issuance. Upon conversion, current accounting rules require the book value of convertible debt to be transferred to equity, and causes predictable post-loan-issuance increases in slack for nancial covenants that rely on measures of debt and equity. 2 I examine whether anticipated slack increases associated with convertible debt conversion cause loan covenant selection to dier across borrowers. Settlement of convertible debt as a liability or equity depends primarily on which outcome is most favorable for the security holder. In addition, the probability that a convertible settles in a particular state changes over time. Prior studies implement various methods to measure the distinct debt and equity characteristics of convertible debt, and report relatively large equity components (King [12], Barth et al. [3]). King [12] nds the average equity value makes up 16.7 percent of the book values of the convertible bonds in his sample. Incorporating equity components of convertible debt can cause measures of solvency to deviate substantially from measures derived from nancial reports. Creditors use liability and equity measurements to assess rm solvency. Structural mod- 2 Upon conversion, rms may choose to record the market value of convertible debt as equity. Recognition of the market value of the converted shares requires the rm to report a loss, as a result rms commonly record the book value of the convertible. 2

3 els rely on debt and equity to measure the rm default boundary and probability, and link leverage measurement to credit risk assessment (Merton [15]). The hybrid nature of convertible debt adds complexity to rm capital structure that is not reected in balance sheet measures of debt and equity. However, Carrizosa [7] nds that creditors incorporate information beyond nancial reports and recognize the separate debt and equity components of convertibles when assessing rm credit risk. Therefore, I expect lenders to recognize the equity characteristics of convertible debt when determining initial loan spreads for borrowers with outstanding convertible debt. Current accounting rules fail to reect the equity characteristics of convertible debt, which results in larger decreases in debt, and larger increases in equity upon conversion, than does a more dynamic approach that reports equity values that increase with conversion probability. Debt, debt to cash ow, debt to balance sheet, and net worth nancial covenants are among those most commonly used in loan contracts, and experience increases in slack due to conversion. Lenders may nd post-loan-issuance slack increases undesirable because they allow borrowers to reduce the value of lenders' claims by replacing 'equity-like' convertibles with equivalent amounts of straight debt. Both the conversion probability, and the associated increase in debt and equity related covenant slack can be anticipated at the inception of the loan contract. I expect lenders to anticipate conversion, and select covenants to minimize the eect of conversion on debt and equity related covenant slack. I test a large sample of newly issued loans and nd that after controlling for rm, issue, and macroeconomic risk factors, rms with outstanding convertible debt at the time of loan issuance pay less interest than do rms without convertible debt. For the full sample, initial loan spreads to rms with convertibles are 6 to 14 basis points lower than those to rms with no convertibles, with larger dierences for rms that rely more heavily on convertible debt. Additional tests indicate the initial interest dierences are greater among non-investment grade rms, and range from 16 to 22 basis points. The ndings persist after controlling for the higher growth options, lower cash interest payments, and lower relative seniority associated 3

4 with convertible debt issuance. In sum, the loan pricing results are consistent with lenders taking into account the equity characteristics of convertible debt when assessing rm credit risk and determining loan pricing. Results from loan covenant tests suggest loans to rms with convertible debt outstanding at the loan issuance date include more nancial covenants than do loans to rms with only straight debt. The larger number of nancial covenants is caused specically by the inclusion of a larger number of debt and equity based covenants. This result is consistent with lenders' inclusion of multiple complementary debt and equity covenants to mitigate the reduced eectiveness caused by post-loan-issuance conversion of convertible debt. My ndings contribute to research regarding investor assessments of the economic substance of hybrid securities. Specically, I provide evidence that lenders recognize the hybrid characteristics of convertible debt when designing loan contracts, which is consistent with the ndings of Carrizosa [7]. As part of current eorts to improve accounting for nancial instruments with characteristics of equity, the FASB and IASB have decided to maintain the classication of convertible debt entirely as a liability. Liability classication of convertibles prevents nancial reports from fully capturing the capital structure complexity caused by convertible debt. However, I nd that lenders do not rely solely on nancial reports, but rather incorporate additional information when measuring rm credit risk, and designing loan contracts for borrowers with convertible debt. 2. Background and Hypothesis Development 2.1. Accounting for Convertible Debt Current accounting rules for convertible debt introduced by Accounting Principles Board Opinion No. 14 (APB 14), specify that the entire proceeds from convertible issuance be recorded in the balance sheet as debt only, and any discount or premium be amortized over the life of the issue. 3 Upon conversion, rms can choose to record the additional equity at 3 Accounting for Convertible Debt and Debt Issued with Stock Purchase Warrants, APB Opinion No. 14(New York: AICPA, 1969). 4

5 either the market or book value. Recognition of the market value of the converted shares requires the rm to report a loss, consequently the book method is commonly chosen and results in a transfer of the book value of convertible debt to equity. Critics of APB 14 claim it understates the true nancing cost of the rm, fails to accurately report dilution to current shareholders, and overstates the eect of the convertible issue on future cash ows. The reported interest required by lenders under APB 14, is the interest for an identical non-convertible issue, minus the value of the conversion option. A rm that issues debt with a conversion option value that completely osets the required interest will appear to have zero nancing costs. In addition, the book method to account for conversion fails to capture the additional dilution to shareholders that occurs when the conversion value exceeds the book value. Lastly, conversion results in the issuance of shares, rather than a reduction of assets. Therefore, the conversion option introduces uncertainty into the estimation of the eect of the convertible on future cash ows, and complicates leverage measurement. International Accounting Standards (IAS) dier from U.S. GAAP by requiring rms to bifurcate the debt and equity components of convertible debt at issuance, and measure the debt component as the fair value of an otherwise identical debt issue without the conversion option. The dierence between the proceeds, and the fair value of the non-convertible issue is recognized as equity. Upon conversion, the rm reclassies the liability component as equity, and makes no adjustment to the equity component recognized at issuance. Prior studies use option-pricing methods to implement the IAS 32 bifurcation method, and nd the magnitudes of the measured equity components to be signicant (King [12], Barth et al. [3]). 4 Recently the FASB has begun to consider alternatives to APB 14 accounting for convertible debt. For certain convertible issues that may be settled partially or entirely in cash, 4 The bifurcation methods implemented in King [12] and Barth et al. [3] are consistent with IAS 32, however they are used to separate the fair value of convertible debt at a nancial reporting date that occurs beyond the issuance date (IAS 32 only bifurcates the issuance proceeds). 5

6 U.S. GAAP now requires the bifurcation of liability and equity components. 5 In addition, eorts to improve accounting guidelines for hybrid nancing instruments, including convertible debt, are currently part of the FASB/IASB project for nancial instruments with characteristics of equity (formerly the liabilities and equity project). However, the standards boards have recently decided against requiring bifurcation of the debt and equity components of convertibles. I examine the impact of APB 14 accounting (non-bifurcation) on debt pricing and covenant design, and provide evidence to regulators regarding the consequences associated with the classication of convertible debt entirely as a liability Cost of Debt Failure to account for the distinct equity characteristics of convertible debt aects two primary types of measures used to assess credit risk. First, failing to separately account for the equity value of the conversion option causes nancial reports to understate the true nancing cost associated with convertible issues. As a result, a rm will report a lower interest expense, and higher income, when it issues convertible debt than when it issues identical non-convertible debt. Though the lower interest expense makes a rm appear more protable, the rm also receives the actual economic benet of a lower required cash interest payments. From a credit risk perspective, either the perceived inated protability, or actual benet of lower cash interest payments can cause investors to view the rm as less risky. Second, the allocation of the conversion option value to debt produces inated leverage measures by overstating debt, and understating of equity. Should creditors rely on unadjusted measures of protability and leverage from nancial reports of rms with convertible debt, then I expect overstated leverage to have a more signicant impact on their credit risk assessments than does overstated protability. Therefore, to determine whether lenders take into consideration the equity characteristics of convertible debt when assessing borrower credit risk and determining initial loan spreads, I focus on the eect of 5 The bifurcation method is consistent with that required by IAS 32 (FASB Sta Position No. APB 14-1, Accounting for Convertible Debt Instruments That May Be Settled in Cash upon Conversion, May 9, 2008). 6

7 leverage overstatement. Structural models of default provide a direct link between measures of debt and equity, and default probability (Merton [15]). The Merton [15] default probability depends heavily on leverage, and has been shown to explain cross-sectional variation in both actual default probability, and bond yield spreads (Hillegeist et al. [11], Bharath and Shumway [4]). Balance sheet debt and equity overstate the default probability of rms with outstanding convertible debt by overstating the default threshold and understating the distance to default, with the degree of the distortion varying across rms and time depending on the conversion option value. For rms with outstanding convertible debt that seek additional nancing, I test whether such misrepresentations aect initial loan pricing. Evidence from prior research suggests creditors recognize the equity characteristics of convertible debt when assessing rm credit risk. Carrizosa [7] uses option pricing methods to measure the debt and equity components of convertible issues, and nds that creditors view the debt component similar to that of an identical non-convertible issue. In addition, credit rating agencies acknowledge the potential dierences between balance sheet classication of securities and their economic substance, and attempt to assess the debt and equity characteristics of hybrids when making rating determinations. The analysts formally assign convertible securities to categories based on their debt and equity characteristics, and recognize only certain fractions of the debt when determining ratings. 6 As a result, I expect lenders to incorporate the equity characteristics of convertible debt when determining the pricing of loans to rms with outstanding convertible debt at the time of loan issuance. Evidence suggests the magnitudes of equity components of convertible debt are signicant 6 The 'equity credit' concept for hybrids is laid out in Moody's ratings methodology document entitled Moody's Tool Kit: A Framework for Assessing Hybrid Securities, 1999 ( The analyst compares the hybrid features to those of common equity, and assigns the security to a category on the debt-equity continuum. Each category determines what fractions of the issue should be classied as debt and equity when incorporating the issue into rm nancial ratios. Convertible debt in particular requires periodic reclassication due to the sensitivity of the conversion probability to issuer operating performance and market conditions. 7

8 both at issuance and beyond (Barth et al. [3], King [12], Carrizosa [7]). Therefore, after controlling for other determinants of credit spreads, including book leverage, I expect that rms with outstanding convertible debt will receive lower interest rates on new loans than do rms with no convertible debt. Prior research nds that the eect of changes in leverage on debt prices is stronger for rms with larger leverage ratios (Collin-Dufresne et al. [8]). Therefore, I anticipate the eect of convertible debt on new loan pricing to be stronger for high credit risk rms. Lastly, I expect the magnitude of the reduction in interest for rms with convertibles to be proportional to the size of the equity component relative to total debt. This leads to the following hypotheses (in alternative form): H1a: Ceteris paribus, new loan interest rates are lower for rms with outstanding convertible debt at the time of loan issuance than for rms without convertibles. H1b: The relationship between new loan interest rates and convertible debt is stronger for rms with large amounts of outstanding convertible debt relative to total debt. H1c: The relationship between new loan interest rates and convertible debt is stronger for high credit risk rms. The equity characteristics of convertible debt are linked to the probability that the issue is converted to equity. Upon conversion, the rm avoids the contractually required future payments of interest and principal, and reduces the likelihood of default. Therefore, lower interest charged to borrowers with convertible debt can be interpreted as compensation to borrowers for the reduction in credit risk expected from conversion. Therefore, I predict that in connection with the lower interest rates, borrowers with outstanding convertibles will experience larger decreases in debt/equity ratios following loan issuance, compared to those of rms without convertibles. Current accounting rules prevent any adjustments to book value of debt or equity prior to conversion, and upon conversion the popular book-value method results in a transfer of the debt book value to equity. As a result, I expect postloan-issuance changes in debt/equity ratios to be particularly sensitive to the conversion of convertible debt. This leads to the next hypothesis (in alternative form): 8

9 H2: Post-loan-issuance changes in book debt-equity ratios for rms with outstanding convertible debt are more negative than those of rms rms without convertibles Covenant Design Debt and equity related covenants restrict the dilution of creditors' claims, and reduce bondholder-stockholder conicts of interest (Smith and Warner [16]). The conversion of convertible debt to equity causes predictable changes in nancial ratios, and increases slack for covenants commonly used in private debt contracts. Increases in slack due to conversion reduce the eectiveness of several common covenants including debt, debt to cash ow, debt to balance sheet, and net worth covenants. I examine whether the anticipation of the eect of conversion on debt and equity related covenant slack aects covenant selection and design. Dichev and Skinner [10] nd that covenants are generally set tightly at issuance, therefore I do not expect reductions in initial slack to mitigate the eect of post-loan-issuance conversion. Given a selection of various covenants, rms will choose those that are more informative of credit risk given borrower characteristics (Demerjian [9]). Assuming alternative covenants can act as substitutes, I expect loan contracts to include fewer debt and equity based covenants for rms with outstanding convertible debt. On the other hand, if other covenants serve as complements rather than substitutes, then loan contracts for rms with convertibles may show no signicant dierences in their use of debt and equity related covenants (or more if the complementary covenants are also debt or equity based). Lenders may reduce the slack increasing eect of conversion by designing the aected covenants to become more restrictive over time (commonly referred to as build-up). The use of build-up could also lead to an increase in debt and equity based covenants. If lenders accept lower interest payments from borrowers with outstanding convertible debt conditional upon eventual conversion and reduction in debt, they may actually prefer debt and equity related covenants with build-up provisions. Use of these covenants would allow lenders to obtain compensation for any anticipated reductions in debt that do not materialize. I test empirically the relationship between outstanding convertible debt at the time of loan 9

10 issuance, and the use of covenants most aected by post-issuance decreases in debt (and increases in equity). 3. Research Design 3.1. Loan Spreads To test whether interest rates for loans to rms with outstanding convertible debt are lower than those to rms with no convertible debt (H1a, b, and c), I run the following OLS regression: Spread ijt = β 0 + β 1 CDF D10 it + β 2 ROA it + β 3 BOOKLEV it + β 4 RET ST D it + β 5 SP R it [EQ1] +β 6 INT P DF D it + β 7 LMBA it + β 8 RND it + β 9 REV ijt + β 10 P P ijt + β 11 MAT ijt +β 12 MCS t + β 13 SLOP E t + ε ijt The dependent variable, Spread, is the initial interest rate spread over LIBOR. The primary variable of interest is CDFD10, an indicator that equals one when the the book value of convertible debt as a fraction of total debt exceeds 10 percent. To test H1b, I include additional indicator variables that equal one when the book value of convertible debt as a fraction of total debt exceeds 25 percent (CDFD25) and 50 percent (CDFD50). In addition to rm-specic risk, lenders take into account both issue-specic risk and macroeconomic conditions when determining interest rates. I include the following rm-specic control variables to test H1a, b, and c (measured prior to loan issuance date): 10

11 ROA: BOOKLEV: Net income / total assets Book leverage measured as nancial debt / total capitalization, where total capitalization is nancial debt plus preferred stock and book value of equity RETSTD: Annualized stock return volatility measured using daily returns for the year prior to loan issue LSIZE: SPR: Log of total assets S&P long-term debt credit rating (numeric scale from AAA=1 to D=22) INTPDFD: Ratio of cash interest paid divided by total debt from the most recent annual report prior to loan issuance LMBA: Log of market to book value of assets, where market value of assets is market value of equity plus preferred stock plus nancial debt minus deferred taxes and investment tax credit RND: Indicator equal to 1 when the borrower reports positive research and development expenses, 0 otherwise I include the following variables to control for issue-specic risk and macroeconomic conditions that aect interest rates charged by lenders: PROC: REV: PP: MAT: MCS: Issue proceeds / total assets Indicator equal to 1 for revolving debt, 0 otherwise Indicator equal to 1 for performance pricing, 0 otherwise Maturity of issue (months) Moody's credit spread, the dierence between yields on BAA and AAA rated corporate bonds SLOPE: Slope of yield curve (10yr-1yr treasury rate) Each regression also includes controls for industry (indicators for Fama-French 10 indus- 11

12 try denition), and loan use of proceeds. 7 I adjust estimated standard errors for clustering at the rm and year level for all regressions. H1a, b, and c each predict β 1 < 0. I replace CDFD10 with indicator variables that correspond to higher fractions of convertible debt to total debt (CDFD25 and CDFD50) to test H1b. To test H1c, I partition the loan spread sample into investment grade (long-term S&P credit rating at least BBB), and non-investment grade observations. To ensure that lower interest rates charged to borrowers with convertible debt is due to lenders' recognition of the equity characteristic of the convertible, I control for potentially confounding factors associated with convertible issuance. First, issuers of convertible debt have higher growth options than their non-convertible debt issuing counterparts (Lewis et al. [13]). To rule out high growth options as a cause of lower interest rates, I include two variables found by prior research to be associated with growth options : 1) the market-to-book assets ratio (Adam and Goyal [1]), and 2) an indicator variable for rms with R&D expenses (Long et al. [14]). In addition, reduced cash interest associated with convertible issues can also lead to lower initial loan spreads. Therefore, I include the ratio of cash interest to total debt as an additional control variable. Lastly, the seniority of convertible debt is typically lower than that of private debt. Therefore, I test the robustness of my results to a possible alternative explanation that lower interest rates to rms with convertibles is due to the association between convertible debt and the lower overall seniority of existing debt. I use both parametric and non-parametric univariate tests to examine changes in debt/equity ratios following loan issuance (H2). I calculate ratios using book value of nancial debt to equity, and measure changes over the four, eight, and twelve quarter periods from the rst reported quarter after the loan issue. I also examine changes in nancial debt and equity separately to determine which causes changes in debt/equity ratios. 7 To control for the intended use of loan proceeds I include indicator variables for the ve following categories: general corporate purposes, renancing, acquisition, capital expenditures, and all others. 12

13 3.2. Covenant Design Prior studies nd that borrower, issue, and macroeconomic factors determine covenant structure for both private and public debt (Bradley and Roberts [6], Billett et al. [5]). To examine the relationship between convertible debt and covenant selection, I estimate the following Poisson regression model: COV NUM it = β 0 + β 1 CDF D10 it + β 2 ROA it + β 3 BOOKLEV it + β 4 RET ST D it + β 5 SP R it [EQ2] +β 6 INT P DF D it + β 7 LMBA it + β 8 RND it + β 9 REV AT it + β 10 T ERMAT it +β 11 P P it + β 12 MAT it + β 13 MCS t + β 14 SLOP E t + ε it The dependent variable COVNUM corresponds to one of three covenant count variables: the number of nancial covenants (FCOVN), the number of nancial covenants that rely on debt or net worth ratios (DEFCOVN), or other nancial covenants (NDEFCOVN, where FCOVN=DEFCOVN+NDEFCOVN). The main independent variable of interest is again the convertible indicator, CDFD10. Loan covenants are specied for a package of facilities that may include both revolvers and term loans. Therefore, to control for the relative sizes of the revolver and term portions if the loan package, I include the following control variables: REVAT: TERMAT: revolver issue proceeds / total assets term loan issue proceeds / total assets 4. Data and Descriptive Statistics I obtain loan data from Securities Database Corporation (SDC) that spans Loans are issued in packages, which consist of one or more facilities or tranches. The most common facilities are revolving and term loans, with separate interest rates assigned to each facility. Covenants are common to all facilities within a package, and SDC covenant coverage spans Table 1 outlines the sample selection criteria imposed for the loan spread sample. Private rms make up a signicant portion of the loans from SDC, and do not have nancial information available on COMPUSTAT. I exclude nancial and utility rms to abstract from 13

14 any potential eect of regulation on the relationship between convertible debt characteristics and initial loan spreads. Convertible debt is only available from COMPUSTAT on an annual basis. To ensure an accurate estimate of the amount of outstanding convertible debt at the time of loan issuance, I restrict the sample to observations that follow fourth quarter reports, and to those which I can verify whether convertibles are outstanding using public bond data from Mergent FISD. 8 After further limiting the sample to observations with available S&P long-term debt ratings and equity market data from CRSP, the resulting sample consists of 3,231 loan packages that contain 4,511 facilities. Table 2 presents summary statistics for the loan spread sample. Nearly 20 percent of the facilities are issued to rms with convertible debt that makes up at least 10 percent of total debt. The larger spreads, lower ROA, and higher return volatility suggest convertible rms are generally more risky than borrowers with no convertible debt outstanding. In addition, the lower long-term debt credit ratings underscore the importance of controlling for dierences in rm-specic credit risk in the analysis. As expected, rms with convertibles benet from lower cash interest payments, and have higher growth options as indicated by their larger market-to-book asset ratios. Loans to rms with convertible debt are typically larger, more likely to include performance pricing, and have longer maturities than those to rms without convertibles. Table 3 reports correlation coecients for initial loan spreads and rm-specic risk characteristics. The positive correlation between the convertible indicator CDFD10 and loan spreads reects the higher risk associated with rms with convertibles. In addition, the strong positive correlations between leverage and return volatility, and credit spreads are consistent with predictions from structural models. Both lower cash interest and higher growth options are associated with lower loan spreads, therefore controlling for these factors 8 For each rm observation, I identify the amount of debt outstanding for each public debt issue just prior to loan issuance using the Mergent FISD amount outstanding history le. I use the sum of any identied convertible debt issues as the total convertible debt outstanding. Otherwise, if I am unable to identify any convertible issues, but am able to identify non-convertible issues, then I assume there is no convertible debt outstanding at the time of loan issuance. 14

15 is necessary to identify the specic eect of convertible debt on leverage measurement and credit risk assessments. 5. Results 5.1. Loan Spreads Results from regressions of initial loan spreads on convertible indicators and controls in Table 4 show a strong negative relationship between convertible debt and initial loan spreads. Column 1 includes control variables for rm-specic and issue-specic risk, and excludes controls for growth options and cash interest payments. The negative and signicant coecient on CDFD10 suggests interest rates to rms with outstanding convertible debt are on average 7.2 basis points less than rates to rms with no convertibles. In addition, the signs of the control variables are consistent with theory and prior research. Specically, initial loan spreads are positively associated with leverage and return volatility (Merton [15]), and negatively associated with protability, revolving debt, and performance pricing (Asquith et al. [2]). To test whether the lower interest to rms with convertibles is caused by higher growth options and lower cash interest payments, I include control variables INTPDFD, LMBA, and RND in Table 4 Column 2. The result that rms with convertibles pay less interest is attenuated, but still negative and signicant at the 5 percent level (coecient increases from -7.2 to -6.2). The signs of the coecients on the additional control variables are consistent with expectations, with the exception of RND (RND is negatively correlated with loan spreads, though the coecient in the full regression is positive). The results of columns 1 and 2 support H1a, and suggest the lower initial interest rates for loans to rms with convertibles are not caused by the rm's higher growth options, or lower cash interest payments. Columns 3 through 6 of Table 4 test whether larger amounts of convertible debt outstanding at loan issuance are associated with lower initial loans spreads (H1b). The negative and 15

16 signicant coecient on CDFD25 in column 3 suggests initial loan spreads for loans to rms with convertible debt amounting to at least 25 percent of total debt are approximately 13.2 basis points lower than those to rms without convertibles. Including controls for growth options and cash interest payments attenuates the result, though the coecient on CDFD25 remains negative and signicant at the 1 percent level (increases from to -11.9). The results in columns 4 and 5 suggest the eect of convertible debt on interest rates is even greater for rms with at least 50 percent of convertible debt. Overall, the results in columns 3 through 6 support H1b and suggest larger fractions of convertible debt are associated with lower initial loan spreads. Table 5 presents regression results for tests of the relationship between convertible debt and initial loan spreads across high and low credit risk rm observations (H1c) (each regression includes controls for growth options and cash interest payments). Result for noninvestment grade observations (non-investment grade long-term debt rating at time of loan issuance, columns 1 through 3) show that rms with convertible debt pay signicantly less interest for new loans than their counterparts that have no convertible debt at the loan issuance date. A comparison of the results in columns 1 through 3 of Table 5 to those in Table 4 suggest the eect of convertible debt on initial loan spreads is stronger for noninvestment grade rms (for example, the coecient on CDFD10 is -6.2 for the full sample, and for non-investment grade observations). Consistent with the results from Table 4, the interest-reducing eect of convertible debt is stronger for rms with larger fractions of convertible debt. In contrast, the insignicant coecients on CDFD10, CDFD25, and CDFD50 in columns 4 through 6 suggest that convertible debt has no signicant impact on interest rates for investment grade observations. Further, the explanatory power of leverage is larger for non-investment grade observations. Together, the results suggest that the interest-reducing eect of convertible debt is stronger specically when leverage measurement has a greater impact on credit risk assessment. In sum, the results of Table 5 support H1c, and suggest the eect of convertible debt on initial loan spreads is stronger for high 16

17 credit risk rms. In terms of claim seniority, a large majority of loans are issued senior secured, while the seniority of convertible debt varies but is not often secured. The convertible debt indicator variables correspond to the fractions of total debt that consists of convertible debt, and may be negatively associated with the overall seniority of the existing claims of the rm at the loan issuance date. The seniority of existing claims aects loan pricing by determining how much existing debt ranks pari passu with the new loan issued. To test whether the results in Tables 4 and 5 are caused by an association between the presence of convertible debt and the lower overall seniority of existing debt issues, I examine a sample that consists solely of observations for rms with no secured debt, and only senior convertibles at the loan issuance date. By excluding observations for rms with secured debt, I ensure that on average the new loan is more senior than all existing rm debt. In addition, by restricting the sample to senior convertibles, I eliminate the possibility that convertible debt is associated with more junior claims. The restricted sample consists of loan 2,150 facilities. The results in Table 6 conrm previous results, and suggest that after controlling for the relative seniority of convertible issues, loans to rms with convertible debt have lower interest rates than do those to rms without convertible debt. If the lower interest charged to rms with convertibles is consideration for the equity component (or conversion probability), then book debt-to-equity ratios (BDE equals FD/SE, where FD is nancial debt, and SE is shareholders' equity) for these rms should decrease relative to ratios for other rms following loan issuance. Changes in BDE are measured using log dierences, and are trimmed at the 2nd and 98th percentiles. Table 7 shows that for rms without convertibles, the BDE decreases over the three year period following issuance (median change in BDE for non-convertible observasions is after four quarters, and after twelve quarters). Consistent with H2, the decrease in BDE is twice as large for rms with at least 10 percent of convertible debt prior to loan issuance (median change in BDE for convertible observations is after four quarters, and after twelve quarters). 17

18 Parametric and non-parametric univariate tests show the dierences between changes in BDE ratios for non-convertible and convertible rms are signicant. Decreases in book debt and/or increases in book equity can cause BDE to decrease. However, if decreases in BDE are caused by conversion, then I expect book debt to decrease, and equity to increase for convertible observations relative to non-convertible observations. I examine the changes in book debt and equity separately, and nd that decreases in BDE are generally caused by increases in both debt and equity, with relatively larger equity increases. However, the results also suggest that rms with convertible debt have relatively smaller increases in debt, and relatively larger increases in equity (median twelve quarter change in FD (SE) for non-convertible observations is.104 (.222) compared to.054 (.286) for convertible observations). The results in Table 7 suggest post-loan-issuance conversion of convertible debt results in predictable changes in book debt-to-equity ratios. In sum, the evidence suggests that rms with convertible debt outstanding pay lower interest on new loans than do rms without convertibles. This result is stronger for rms with larger fractions of convertibles, and non-investment grade long-term debt credit ratings. Also, the result is robust to controls for the higher growth options, the lower cash interest, and the lower relative seniority associated with convertible issuance. Overall, the ndings are consistent with rms recognizing the equity characteristics of convertible debt when measuring leverage and assessing rm credit risk Covenant Design The covenant design sample consists of the subsample of unique packages in the loan spread sample that also have available covenant information. 9 The sample consists of 548 unique loan packages, over 20 percent of which are issued to rms with at least 10 percent convertible debt. The summary statistics for issue and rm characteristics in Table 8 suggest loans to rms with convertibles have more nancial covenants. Further inspection shows that 9 SDC loan covenant coverage spans the period from 1995 to 2004, and only those observations for which the appropriate loan documentation is available. 18

19 the larger number of nancial covenants to rms with convertibles is primarily due to a larger number of covenants that are most aected by post-loan-issuance conversion of convertible debt, specically those that rely on measures of debt and net worth. Table 9 presents results from Poisson regressions of nancial covenant counts, and - nancial covenant subgroups, on variables that reect rm, issue, and macroeconomic risk. I rst examine the relationship between the total number of nancial covenants included in loan contracts, and outstanding convertible debt at contract inception. The positive and signicant coecient on CDFD10 in column 1 suggests that loan contracts to rms with convertible debt include more nancial covenants. Next, I examine separately the association between outstanding convertible debt and the inclusion of debt and equity related nancial covenants. Results in column 2 suggest loans to rms with convertibles include more nancial covenants that are aected by post-loan issuance conversion. Combined with the nding that convertible debt does not aect the number of other nancial covenants (covenants that do not rely on debt or equity measures), the results in Table 9 suggest the increase in nancial covenants is mostly due to an increase in the inclusion of debt and equity related covenants. 10 Additional tests (unreported) do not suggest loans to rms with convertible debt have a higher probability of including a build-up provision for their debt and equity related nancial covenants. Overall, the results suggest lenders include additional debt and equity related covenants in loan contracts to borrowers with outstanding convertible debt. A possible explanation for the increased usage of debt and equity related covenants is that combining multiple aected covenants mitigates the complications associated with post-issuance covenant slack increases caused by conversion. 10 Other nancial covenants include those based on coverage, protability, liquidity, and capital expenditure measures. 19

20 6. Conclusion The failure of current accounting rules to recognize the inherent equity characteristics of convertible debt distorts measures of protability and solvency. I focus on the eect of convertible debt accounting on leverage measurement, and examine whether the overstatement of leverage aects debt pricing. In addition, I test whether increases in covenant slack due to post-loan-issuance conversions of convertible debt aect covenant selection. Results from tests of initial loan spreads suggest that after controlling for rm, issue, and macroeconomic risk factors, rms with outstanding convertible debt at the time of loan issuance pay 6 to 22 basis points less interest on new loans than do rms without convertible debt. Moreover, interest rates are lower for rms that rely heavily on convertible debt, and for those with non-investment grade long-term debt ratings. The results hold after controlling for the higher growth options, lower cash interest payments, and lower relative seniority generally associated with convertible debt issuance. In sum, the loan pricing ndings are consistent with lenders taking into account the equity characteristics of convertible debt when assessing rm credit risk and determining loan pricing. In addition, results from tests of loan covenants suggest loans to rms with convertible debt outstanding at the loan issuance date include more debt and equity related nancial covenants than do loans to rms with only straight debt. This nding is consistent with lenders' use of multiple complementary debt and equity based covenants to minimize the eect post-loan-issuance slack increases caused by the conversion of convertible debt. Current eorts by the FASB and IASB aim to improve accounting for nancial instruments with characteristics of equity, including convertible debt. Recently, the FASB and IASB have decided to maintain the current classication of convertible debt entirely as a liability. I provide evidence that liability classication of convertibles produces balance sheet measures of leverage that overstate rm credit risk. However, the evidence also shows that creditors do not rely solely on balance sheet classication to assess the economic substance of convertible debt. Rather, they adjust balance sheet measures of leverage to incorporate 20

21 the distinct debt and equity characteristics of convertible debt when assessing rm credit risk, and determining loan interest rates. 21

22 References [1] Adam, T. and Goyal, V. K. (2008). The investment opportunity set and its proxy variables. Journal of Financial Research, 31(1):4163. [2] Asquith, P., Beatty, A., and Weber, J. (2005). Performance pricing in bank debt contracts. Journal of Accounting and Economics, 40: [3] Barth, M. E., Landsman, W. R., and Rendleman, Richard J., J. (1998). Option pricingbased bond value estimates and a fundamental components approach to account for corporate debt. The Accounting Review, 73(1): [4] Bharath, S. T. and Shumway, T. (2008). Forecasting default with the merton distance to default model. Rev. Financ. Stud., 21(3): [5] Billett, M. T., King, T.-H. D., and Mauer, D. C. (2007). Growth opportunities and the choice of leverage, debt maturity, and covenants. The Journal of Finance, 62(2): [6] Bradley, M. and Roberts, M. R. (2004). The structure and pricing of corporate debt covenants. Working Paper. [7] Carrizosa, R. D. (2010). Market participants' evaluation of the economic substance of convertible debt. Working Paper, pages 144. [8] Collin-Dufresne, P., Goldstein, R. S., and Martin, J. S. (2001). The determinants of credit spread changes. The Journal of Finance, 56: [9] Demerjian, P. R. (2007). Financial ratios and credit risk: The selection of nancial ratio covenants in debt contracts. SSRN elibrary. [10] Dichev, I. D. and Skinner, D. J. (2002). Large-sample evidence on the debt covenant hypothesis. Journal of Accounting Research, 40(4):

23 [11] Hillegeist, S., Keating, E., Cram, D., and Lundstedt, K. (2004). Assessing the probability of bankruptcy. Review of Accounting Studies, 9:534(30). [12] King, R. D. (1984). The eect of convertible bond equity values on dilution and leverage. The Accounting Review, 59(3): [13] Lewis, C. M., Rogalski, R. J., and Seward, J. K. (1999). Is convertible debt a substitute for straight debt or for common equity? Financial Management, 28(3):527. [14] Long, M. S., Wald, J. K., and Zhang, J. (2002). A cross-sectional analysis of rm growth options. Working Paper. [15] Merton, R. C. (1974). On the pricing of corporate debt: The risk structure of interest rates. Journal of Finance, 29(2): [16] Smith, C. J. and Warner, J. B. (1979). On nancial contracting : An analysis of bond covenants. Journal of Financial Economics, 7(2):

24 Table 1: Initial Loan Spread Sample Selection Panel A: Sample Selection Number of Selection Criteria Facilities 1 Loan facilities from SDC, Jan Nov ,089 2 Excluding amendments to previous loans 42,166 3 COMPUSTAT data for quarter prior to loan issuance 10,099 4 Excluding nancial and utility rms 9,605 5 Exclude observations with less than 10% debt-to-total capital ratio or negative shareholders' equity 8,468 6 Veried convertible debt status at time of loan issuance 5,457 7 Long-term debt S&P credit rating prior to loan issuance 4,548 8 Return volatility from CRSP 4,511 Panel B: Sample Summary Loan Packages 3,231 Loan Facilities 4,511 Revolvers 3,442 Term Loans 1,069 Sources: Securities Data Corp. (SDC), COMPUSTAT, Mergent FISD, CRSP. 24

25 Table 2: Firm and Loan Characteristics, Loan Spread Sample Total (N=4511) No/Low Convertible (N=3782) Convertible (N=729) Variable mean sd median mean sd median mean sd median Di t pval SPREAD LSIZE ROA BOOKLEV RETSTD SPR INTPDFD LMBA RND PROC REV PP MAT Description: Firm and issue characteristics for the full loan spread sample, and no/low convertible (convertible debt makes up less than 10 percent of total debt) and convertible (convertible debt makes up at least 10 percent of total debt) subsamples. The loan sample spans the period from January 1990 to November Variable Denitions: SPREAD is the initial interest rate spread over LIBOR. LSIZE is the log of total assets. ROA is the ratio of net income to total assets. BOOKLEV is book leverage measured as the ratio of nancial debt to total capitalization, where total capitalization is nancial debt plus preferred stock plus book value of equity. RETSTD is the annualized stock return volatility measured using daily returns for the year prior to loan issue. SPR is the S&P long-term debt credit rating (numeric scale from AAA=1 to D=22). INTPDFD is the ratio of cash interest paid to total debt from most recent annual report prior to loan issuance. LMBA is the log of the ratio of market to book value of assets, (market value of equity + preferred stock + nancial debt - deferred taxes and investment tax credit) / total assets). RND is an indicator variable equal to 1 when the borrower reports positive research and development expenses, 0 otherwise. PROC is the ratio of issue proceeds to total assets. REV is an indicator variable equal to 1 for revolving debt, 0 otherwise. PP is an indicator variable equal to 1 for performance pricing, 0 otherwise. MAT is the maturity of issue (months). 25

26 Table 3: Correlation Matrix, Loan Spread Sample SPREAD CDFD10 LSIZE ROA BOOKLEV RETSTD SPR INTPDFD LMBA RND SPREAD CDFD LSIZE ROA BOOKLEV RETSTD SPR INTPDFD LMBA RND Description: Correlation matrix for initial loan spreads and rm characteristic (Pearson correlation coecients below the diagonal, Spearman coecients above) Variable Denitions: SPREAD is the initial interest rate spread over LIBOR. CDFD10 is an indicator variable that equals 1 when the ratio of convertible debt to total debt exceeds 10 percent, and 0 otherwise. LSIZE is the log of total assets. ROA is the ratio of net income to total assets. BOOKLEV is book leverage measured as the ratio of nancial debt to total capitalization, where total capitalization is nancial debt plus preferred stock plus book value of equity. RETSTD is the annualized stock return volatility measured using daily returns for the year prior to loan issue. SPR is the S&P long-term debt credit rating (numeric scale from AAA=1 to D=22). INTPDFD is the ratio of cash interest paid to total debt from most recent annual report prior to loan issuance. LMBA is the log of the ratio of market to book value of assets, (market value of equity + preferred stock + nancial debt - deferred taxes and investment tax credit) / total assets). RND is an indicator variable equal to 1 when the borrower reports positive research and development expenses, 0 otherwise. 26