The Impact of Changes in Balance Sheet Covenant Protection on the Design of Private Loan Contracts Peter Demerjian* Goizeuta Business School Emory University 1300 Clifton St NE Atlanta GA 30322 October 2010 Abstract I examine the economic implications of the recent decline in use of balance sheet-based financial covenants in private debt contracts. I find the borrowers that experience a drop in balance sheet covenant protection to have both higher interest rates (35.3 basis points) and shorter maturities (7.4 months) for their loans. However, I find no association between the decline in balance sheet covenant use and the inclusion of either income statement covenants or restrictive covenants. I also find the costly effects of dropping balance sheet covenants are restricted to cases where the specific demand for balance sheet covenants is low, but the overall demand for contractual protection is high. * 404-727-2329. [email protected] 1 Introduction Recent years have seen a sharp decline in the use of debt covenants written on balance sheet values. As documented in Demerjian (2010a), over the period 1996 through 2007, balance sheet covenant use in private loans fell from 80% to 30%.1 There are a number of explanations for this change. It may be that the overall demand for protection in debt contracts is decreasing; that is, borrowers may have improved over this time period and thus lenders require less protection written into the contract. Or, it could be due to changes in the underlying economics of borrowers. For example, if borrowers’ value is comprised of fewer fixed assets and more growth opportunities, the balance sheet may provide less information than it did in the past. Another explanation is that the quality of information from the balance sheet is declining. Demerjian (2010a) argues that the conceptual shift to the “balance sheet approach” and the increased use of fair value accounting has rendered the balance sheet insufficiently reliable for contracting. Finally, borrowers may desire greater flexibility and opt to avoid covenants in contracts, even if this means increased strictness of other terms. Given the significant decrease in balance sheet covenant use, there are likely consequences to this change. In this study, I assess the economic impact of the decline, focusing on changes in the design of debt contracts. I distinguish between two broad motivations for the reduced use of balance sheet covenants: changes in firm-level demand for covenant protection, and changes in covenant-level demand. Firm-level demand refers to the overall demand for protection in the debt contract by the creditor. This demand will be increasing in the default risk of the borrower, and will drive the aggregate amount of protection that the loan provisions provide. Covenant-level demand refers to the demand for a specific provision, in this study 1 Balance sheet covenants include those written exclusively with balance sheet values: net worth, leverage, or current ratio. 1 balance sheet covenants. While not mutually exclusive, these two types of demand are distinct: changes in covenant-level demand can impact the use of a specific covenant even when firmlevel demand is constant (or moving in the opposite direction as covenant-level demand). This distinction leads to different predictions based on whether the decline in balance sheet covenant use is due to changes in firm-level or covenant-level demand. If firm-level demand drives the decline in balance sheet covenants, I do not anticipate any change in other loan provisions; the reduction in protection due to not having a balance sheet covenant is offset by the lower demand for protection. In contrast, if the decline is due to a decrease in covenant-level demand, and firmlevel demand has not changed (or has increased), I expect that creditors will require compensation for the reduction in covenant protection by tightening other loan provisions. I examine the relation between changes in balance sheet covenant protection and four other aspects of loan contract design: the interest spread, the maturity, the use of income statement covenants, and the use of general covenants. I collect 1,458 private loan observations from 729 borrowers; the sample is designed so that each borrower has a balance sheet covenant in their earlier loan (the Pre-Drop period) and no balance sheet covenant in their later loan (the Drop period). I then analyze changes in the loan contract design between the Pre-Drop and Drop periods. Observations are split based on the change in default risk, my proxy for changes in firm-level demand. If default risk decreased, I label the observations “Low Risk”; these are cases where the firm-level demand for protection declined, and I expect that any loosening of loan provisions (including removing balance sheet covenants) can be attributed to improvement in the quality of borrowers. If default risk increased, I term these borrowers “High Risk”; in these cases, firm-level demand for protection has increased, so the drop in balance sheet covenant use is due to decline in the covenant-level 2 demand for balance sheet covenants. I expect tightening of contract provisions in High Risk observations, but not in Low Risk. The empirical results bear out this prediction. Using a difference-in-differences research design, I find that dropping a balance sheet covenant in a period of increasing firm-level demand is associated with higher interest spreads and shorter maturities. I find no association between dropped balance sheet covenants and either income statement covenant inclusion or general covenant inclusion. In economic terms, the impact on loan spread appears substantial. For High Risk borrowers without a balance sheet covenant, interest spreads are 35.3 basis points higher. Based on the average loan size, this leads to an increase in interest expense of $2.25M, or approximately 5.4% of income. If balance sheet covenants are becoming less useful for contracting and hence not used in contracts (i.e. due either to changes in borrower characteristics or changes in the information provided by the balance sheet), this suggests the decline in their use is costly. If the decline is due to changing preferences in contracting (i.e. a shift in the preferred mix of contract provisions), this suggests the average amount that borrowers are willing to pay to avoid having a balance sheet covenant. For maturity, loans from the High Risk group have maturities that are 7.4 months, or about 15%, shorter when balance sheet covenants are dropped. Though the cost is more difficult to quantify, this similarly suggests borrowers are either forced into more frequent loan renegotiation (with all the attendant risks and costs) or are willing to renegotiate more frequently to avoid being checked by covenants. This research sheds light on the overall design of loan contracts by showing how the different terms interact. I find, consistent with Christensen and Nikolaev (2010), that balance sheet and income statement covenants are not substitutes but likely serve distinct purposes in 3 debt contracts. I also find that balance sheet covenants and loan maturity are substitutes. This is consistent with the covenants providing control rights to lenders in technical default, thus allowing borrowers longer loans when performance is sufficiently good. Finally, consistent with Bradley and Roberts (2004), I find that balance sheet covenants are priced in loan contracts. In the next section I present the background and develop hypotheses. Data and research design are presented in Section 3. Principal empirical results are presented in Section 4, with supplementary test results in Section 5. I conclude in Section 6. 2 Background and Hypothesis Development 2.1 The Balance Sheet in Debt Contracting The balance sheet plays an important role in debt contracting; collateral values are derived from balance sheet values, covenants are written on balance sheet values, and other measures are calculated at least in part from GAAP-reported balance sheet figures (e.g. regulatory capital (Beatty et al. 1995)). Holthausen and Watts (2001) discuss how the balance sheet measures the liquidation value of the borrower. Similarly, Watts (2003) argues that the balance sheet must provide a “verifiable lower bound” of asset values to be useful for contracting. Kothari et al. (2010) discuss how certain features of US GAAP, particularly asymmetric verifiability requirements and conditional conservatism, have evolved to keep accounting useful for contracting. Several recent studies have examined the specific role of balance sheet data in debt contracts. Demerjian (2010b) finds that poorly performing firms that are closer to liquidation are more likely to have net worth covenants. Christensen and Nikolaev (2010) find that balance sheet covenants (which they term “capital covenants”) are in place to protect lenders by requiring firms to maintain a minimum level of value. 4 I define balance sheet covenants as those measured exclusively with balance sheet values, a group that includes net worth, leverage (debt-to-assets or debt-to-equity), and current ratio. I do not include the debt-to-cash flow (generally measured as the ratio of debt to EBITDA) since its measurement is partially from the income statement. Demerjian (2010a) notes that the use of balance sheet covenants has declined sharply in recent years, being used in 80% of private loans in 1996 to just 30% by 2007. This trend is illustrated in Figure 1. 2.2 Firm-Level and Covenant-Level Demand The decline in use of balance sheet covenants can be explained by changes in two distinct but related types of demand. The first, which I term “firm-level demand”, is the demand for protection by the creditor based on the general risk that the borrower presents in contracting. By this definition, firm-level demand can be measured as the default risk of the borrower: creditors of borrowers with high levels of default risk will demand more protection, while low levels of default risk will be rewarded with a lower demand for protection. Under high firm-level demand, the lender will require tighter contract provisions (e.g. high interest rates, short maturities, restrictive provisions); if the default risk of the borrower improves, loan provisions will be loosened commensurately. Much of the theory on covenant use (as well as other contract provisions) focuses implicitly on firm-level demand. The Agency Theory of Covenants, proposed in Bradley and Roberts (2004), follows Smith and Warner (1979) in linking the use of covenants to moral hazard problems. The Signaling Theory of Covenants, discussed in Demiroglu and James (2010), examines how borrowers signal their quality to the debt market by agreeing to have covenants. Each of these theories, and subsequent empirical tests, takes a broad 5 view of the role of covenants in contracts: a lender faces certain information problems with a borrower (either moral hazard or adverse selection) and uses covenants to address the problems.2 Firm-level demand for covenants is not the only driver of covenant demand. The second, which I term “covenant-level demand”, is the demand for a specific covenant (rather than other provisions); in this study, I focus on the covenant-level demand for balance sheet covenants.3 There are a number of reasons why the demand specific to balance sheet covenants may decline. If the underlying economics of borrowers shift, this may cause balance sheet information to be less useful in contracting. For example, if a firm starts investing more in R&D and other unreported intangible assets, the balance sheet will provide relatively less information about the firm’s true value (Skinner 1993). Or, balance sheet information may become less useful for contracting; Demerjian (2010a) examines whether fair value accounting has diminished the contracting value of balance sheet information. There may also be structural changes in the debt market that leads to changes in demand for balance sheet covenants. For example, Drucker and Puri (2009) document the association between net worth covenants and loan sales on the secondary market. Finally, borrowers may prefer other provisions rather than balance sheet covenants. They may be willing to pay (e.g. have a higher interest rate or shorter maturity loans) to remove the financial covenant and enjoy greater flexibility in operating and financing decisions.4 Firm-level and covenant-level demand are not mutually exclusive. In fact, since balance sheet covenants make up part of the set of protective provisions of the debt contract, declines in 2 For example, Billett et al. (2007) examine how growth opportunities, a proxy for general agency problems, affect maturity and debt covenant intensity in bond contracting. 3 Covenant-level demand is more aptly referred to as “individual provision-level demand”; this refers to the case where one provision is preferred over another for some reason other than the overall risk of the borrower. I refer to this as “covenant-level” in this study due to my focus on balance sheet covenant use. 4 Beatty et al. (2002), examining how changes in GAAP are incorporated into debt contracts, find that borrowers are willing to pay higher interest spreads to maintain reporting flexibility. 6 firm-level demand can manifest as declines in demand for balance sheet covenants. The critical distinction is that covenant-level demand can change even when firm-level demand for protection remains constant. By this definition, changes in covenant level-demand can occur independent of changes in firm-level demand. That is, covenant-level demand can decrease even when firm-level demand for protection has not changed or has increased. This distinction is important, because the driver of the decline in balance sheet covenant use (firm-level versus covenant-level) affects the consequences. Consider a borrower with improved operating performance, leading to lower default risk and a decline in firm-level demand for protection. If removing a balance sheet covenant exactly offsets the decrease in firmlevel demand (i.e. that level of protection that the balance sheet covenant provided is equal to the decline in protection required by the creditor), the other provisions of the loan contract should not change.5 If removing a balance sheet covenant either over- or under-compensates the borrower for decreased risk, other provisions will be changed to balance the aggregate protection provided by the loan contract with the borrower’s risk. Since the level of protection that a balance sheet covenant provides varies from borrower to borrower, I do not anticipate a systematic tightening or loosening of other contract provisions in response to dropping a covenant due to a decline in firm-level demand. In contrast, a decline in covenant-specific demand can occur when firm-level demand has not changed. Consider a borrower whose overall default risk is constant but where the demand for balance sheet covenants has decreased to the point where a covenant is not used in a new loan contract. With the covenant excluded, the aggregate protection has declined; however, the demand for protection has not changed, meaning the remaining provisions are insufficient based 5 More precisely, the aggregate level of protection provided by the other loan provisions (e.g. maturity, interest rate, collateral requirements, restrictive covenants) will not change. It is possible, due to macroeconomic or other changes that the efficient mix of other provisions will change, though the aggregate protection will be the same. 7 on the creditor’s demand for protection. As such, another provision or provisions will be tightened to compensate for the dropped covenant. For example, the creditor may require a higher interest rate or add restrictive covenants. Hence, a covenant-specific decline in demand must be accompanied by tightening in other loan provisions, holding firm-level demand constant. The extent of the tightening will be a function of both how much protection the dropped covenant had provided as well as any change in firm-level demand. To summarize, the decline in use of balance sheet covenants can be attributed to two broad reasons: a decline in firm-level demand or a decline in covenant-level demand. I expect any systematic effect of declining balance sheet covenant inclusion to be driven by changes in covenant-level (but not firm-level) demand. That is, I anticipate a systematic tightening of loan provisions when a balance sheet covenant is dropped due to covenant-level demand, but no changes when the drop is related to firm-level demand. 2.3 Hypotheses There are a variety of debt contract provisions that serve to either limit the risk of the creditor entering a risky lending agreement or compensate the creditor for bearing risk. If the decline in balance sheet covenant use has affected the protection provided to the creditor in the debt contract, I expect to see this manifested in the structure of the loan contract. In this section, I describe four provisions—interest spread, maturity, income statement covenants, and general covenants—that either limit or compensate for risk, and present hypotheses on how these provisions may be affected by declining balance sheet covenant protection. The first provision is interest spread, or the pricing of the loan. Following the standard risk-return relation, if removing a balance sheet covenant reduces the aggregate protection in the 8 loan contract, the creditor may require more interest as compensation. There is considerable evidence that inclusion of provisions that reduce risk and agency problems are priced in debt contracts. Examples include performance pricing (Asquith et al. 2005), collateral (Chan and Kanatas 1985, Stulz and Johnson 1985), and restrictive covenants (Reisel 2006, Chava et al. 2010). Bradley and Roberts (2004) show that a variety of covenants, including financial covenants, are priced. This evidence suggests that a decline in covenant protection, holding other things equal, may result in a higher interest spread. H1: The decline in balance sheet covenant inclusion is associated with higher interest spreads on loans. The second provision that could be affected is loan maturity. Barclay and Smith (1995) and Childs et al. (2005) find that the maturity structure of debt is linked to agency problems. To the extent that balance sheet covenants limit agency issues, their exclusion may induce a shorter debt maturity. More directly, technical default serves as a trigger for renegotiation of private loans. Dichev and Skinner (2002) find that covenants serve as “trip wires”, allowing the creditor the option of action (including the shortening of maturity) when a covenant is violated. Exclusion of covenants may therefore encourage the creditor to offer shorter maturity loans, since he loses the ability to renegotiate in technical default: a loan with a long maturity and a financial covenant is de facto a shorter-term loan conditioned on borrower performance. It follows that the fewer covenants should result in shorter loan maturities: H2: The decline in balance sheet covenant inclusion is associated with shorter maturity loans. Another possible substitute for balance sheet covenants are income statement covenants. These covenants—including interest coverage, fixed charge coverage, and debt-to-earnings ratio—function similarly to balance sheet covenants. That is, the borrower must maintain a 9 threshold, and the loan enters technical default if the threshold is not maintained. It is not clear that covenants written on these different financial statements will readily serve as substitutes: Christensen and Nikolaev (2010) shows that income statement and balance sheet covenants provide different information in contracting and are used by different borrowers. This is consistent with Kothari et al. (2010), who note that the income statement captures short-term performance while the balance sheet captures the liquidation value of assets. However, some studies examining covenant use have found a single set of drivers for use of all financial covenants, suggesting the possibility they are substitutes (Bradley and Roberts 2004). Given the ambiguity of the existing evidence, I consider income statement covenants as substitutes for balance sheet covenants: H3: The decline in balance sheet covenant inclusion is associated with an increase in income statement covenant inclusion. Finally, balance sheet covenants and restrictive covenants may be substitutes. Restrictive, or negative, covenants directly limit actions that can transfer wealth from the borrower to the creditor. For example, Smith and Warner (1979) show that covenants may restrict dividend payments (to limit underinvestment) or debt issuance (to limit claim dilution). The moral hazard problems addressed with restrictive covenants can also be limited with balance sheet covenants. Following the example above, net worth covenants can limit dividend payments, and leverage covenants restrict the amount of new debt a borrower can assume. While there is little theory on the decision to use a restrictive covenant versus a financial covenant—it is likely that financial covenants allow more flexibility, but with a greater risk of measurement error—it is possible they serve as substitutes, particularly if balance sheet information is losing its contracting usefulness (Demerjian 2010a). This leads to the final prediction: 10 H4: The decline in balance sheet covenant inclusion is associated with an increase in restrictive covenant inclusion. 3 Sample and Research Design 3.1 Sample Private loan data is drawn from the LPC/Dealscan database. This database provides detailed data on loans, including interest rates, loan size, and maturity. Additionally, data on a variety of contract provisions, including covenants, is provided. The loans in this sample are drawn from 1996 (when Dealscan begins comprehensive coverage of the private debt market) to 2007. Accounting data is drawn from the quarterly Compustat Xpressfeed. Dealscan lacks common firm descriptors (e.g. cusip or gvkey) so observations are matched to Compustat by name and hand-checked to verify accuracy. Loans are matched to the accounting quarter most closely preceding the loan’s initiation date. All flow variables from Compustat (i.e. from the income statement or statement of cash flows) are annualized using the sum of the current and prior three quarters. Dealscan data is organized on two levels, facilities and packages. Facilities are individual loans, while loan packages are groups of facilities from the same lender. All the facilities in a loan package have the same set of covenants, so my unit of analysis is the package.6 The sample selection consists of four steps: 1. I first identify borrowers with at least two loan packages over the sample period. 2. Of these multi-deal borrowers, I find those that, at some point in the sample period, have at least one balance sheet covenant. 6 I use the terms “loan package” and “deal” interchangeably. 11 3. Of those with a balance sheet covenant, I identify those borrowers who have a subsequent loan package with no balance sheet covenants; that is, I identify borrowers that had balance sheet covenants “dropped” from one deal to the next. 4. I remove borrowers with multiple drops; that is, borrowers who have a balance sheet covenant dropped and in a subsequent package have a balance sheet covenant again.7 For the sample, I retain the first package from a borrower where there was no balance sheet covenant (the Drop deals) and the package with a balance sheet covenant immediately preceding the Drop (the Pre-Drop deals). Sample selection is summarized in Table 1. The intersection of the Dealscan population with Compustat data consists of 9,000 deals from 3,087 individual borrowers; the sample selection yields 1,458 deals from 729 individual borrowers. Table 2 presents the distribution of sample observations by year. By design, each firm has a pair of observations, one in the Pre-Drop period and one in the Drop period. The mean (median) amount of time between the Pre-Drop and Drop deal is 2.7 (2.3) years. This is illustrated in the table, with the Pre-Drop deals more concentrated earlier in the sample period and Drop deals more likely later. 3.2 Research Design I start by defining DROP as an indicator with a value of one when a deal does not have a balance sheet covenant (net worth, leverage, or current ratio) and zero when it does. Based on the sample selection criteria, each borrower will have two observations in the sample: the earlier will have a value of zero for Drop, and the later will have a value of one. 7 The observations dropped in this step include 44 borrowers with two drops, five borrowers with three drops, and one borrower with four drops, for a total of 214 observations. Including these observations does not affect the inferences from the empirical tests. 12 I then examine the association between Drop and four variables related to debt contract structure. INTEREST SPREAD is the amount over an index rate of interest (generally LIBOR or Prime) that the borrower is charged for the loan. Higher interest spreads imply higher risk, so I expect a positive association between Drop and Interest Spread. MATURITY is the term of the loan from inception to the stated maturity date, measured in months. Long-term loans are generally considered riskier than shorter loans, so I predict a negative association between Drop and Maturity. IS COVENANT is an indicator variable with a value of one if the borrower has at least one income statement covenant (interest coverage, fixed charge coverage, debt-to-earnings) and zero otherwise. GENERAL COVENANT is an index of indicator variables related to direct restrictions on borrower actions. There are five restrictions in the index: asset sales, dividends, debt issuance, equity issuance, and insurance proceeds sweeps. If either income statement or general covenants serve to substitute for balance sheet covenants, there will be a positive association between Drop and both IS Covenant and General Covenant. As discussed in Section 2.2, the nature of the decline in balance sheet covenant use (firmlevel demand versus covenant-level demand) has important implications for the predictions; specifically, I expect the hypothesized results to hold only for covenant-level decreases in demand. As such, I control for the nature of the decline in the research design. Directly measuring changes in covenant-level demand is difficult. It can be affected by a number of things, including changes in the underlying economics of borrowers and the move to balance sheet accounting (Christensen and Nikolaev 2010; Demerjian 2010a). Measuring a change in firm-level demand is relatively easier. As a proxy for firm-level demand, I use the expected default risk (EDF), based on Merton (1974), which incorporates leverage and the volatility of borrower asset value in an options-style model. Exploiting the structure of the sample, I measure 13 the change in EDF from the Pre-Drop to the Drop period on a borrower-specific basis. If the default risk declines from the Pre-Drop to the Drop period, I attribute the dropped covenant to lower firm-level demand.8 If default risk increases from the Pre-Drop to the Drop period, I expect the covenant was dropped due to a covenant-level factor. I define HIGH RISK as an indicator with a value of one if EDF increases from the Pre-Drop to the Drop period, and a value of zero otherwise. Using Drop and High Risk, I employ a difference-in-difference design in the main empirical tests. These regression-based tests have the structure: , Γ ! " Loan Parameters include Interest Spread, Maturity, IS Covenant, and General Covenant. The coefficient of interest in each regression is : this captures the association of the loan parameter with dropping the balance sheet covenant when the firm-level demand for protection is high. I predict a positive coefficient on when the Loan Parameter is Interest Spread, Is Covenant, or General Covenant, and a negative coefficient when it is Maturity. The tests include a variety of controls associated with the different loan parameters, including borrower and loan characteristics. Borrower characteristics include FIRM SIZE (the natural logarithm of total assets), ROA (EBITDA scaled by average total assets), LEVERAGE (total long-term debt scaled by total assets), RATED (an indicator if the borrower has an S&P Senior Debt Rating), ASSET MATURITY (the age of assets), and ASSET TANGIBILITY (the relative amount of tangible to intangible assets). Loan characteristics include LOAN SIZE (the natural logarithm of the loan amount), REVOLVER (an indicator if the loan package has a 8 A decline in firm-level demand can coincide with a decline in covenant-level demand. My assumption is that firmlevel changes dominate the effects of covenant-level changes; in other words, I expect that contracts are designed with aggregate demand for protection having first-order importance, and the specific mix of provisions having second-order importance. 14 revolving line of credit), INSTITUTIONAL TRANCHE (an indicator if the loan has a Term Loan B tranche or higher), PERFORMANCE PRICING (an indicator if the loan has a performance pricing provision), LENDERS (the number of lenders in the loan syndicate), and COLLATERAL (an indicator if the loan requires collateral). I provide detailed variable definitions in the Appendix. Summary statistics on the test and control variables are presented in Table 3, and simple correlations are shown in Table 4. Table 5 presents differences in test and control variables based on the period of the observation. The Drop column presents the mean and median level of the variables for sample observations without balance sheet covenants, while the Pre-Drop column presents similar statistics for observations preceding the drop. The next two columns present the difference and the p-value of a test of significance (t-test for means, Wilcoxon signed-rank test for medians). This descriptive evidence shows that Interest Spread is significantly higher in the Drop period (mean 27.643, median 42.5), consistent with prediction. However, Maturities after the drop are longer (mean 5.241, median 12), and both IS Covenants (mean -0.114) and General Covenants (mean -0.221) are lower in the Drop period, inconsistent with prediction. Though descriptive, this data indicates it is important to control for the driver of the decline in covenant protection. The data also shows, though Drop and Pre-Drop observations are matched by borrower, there are many significant differences in the control variables. This suggests that features of the borrowers and their loans are changing over time, and must be controlled in the tests to make clear inferences. 4 Empirical Results 4.1 Univariate Results 15 The analysis begins with univariate difference-in-differences tests on the four loan parameters. The results are shown in Table 6. The first set of columns, titled “High Risk”, presents data for the 282 pairs of observations (i.e. 282 Pre-Drop observations and 282 Drop observations) where EDF increased (High Risk = 1). Within this High Risk group, I further sort the firms between the Drop and Pre-Drop periods, and measure the mean values of the loan parameters. The second set of columns presents similar statistics for the 385 “Low Risk” observation-pairs (High Risk = 0). The first parameter tested, in the top rows, is Interest Spread. For High Risk borrowers, the average Interest Spread in the Drop period is 237.478, significantly higher than the Pre-Drop average of 168.229 (difference 69.249, t-statistic 5.73). In contrast, the average difference in the Low Risk observations (-5.49, t-statistic -0.65) is not significantly different from zero. The far right hand columns present the difference-in-differences; in this case, a significant figure of 74.739 basis points (t-statistic 5.06). By contrasting the High Risk and Low Risk settings, this test captures general trends in Interest Spread; essentially, the Low Risk observations serve as a benchmark to evaluate the High Risk. The results support H1: borrowers that have a balance sheet covenant dropped when there is high firm-level demand for covenant protection have significantly higher interest spreads than when demand is low. The send parameter tested is Maturity. For the High Risk firms, the difference between the Drop and Pre-Drop average maturity is close to zero (-0.989, t-statistic -0.61), suggesting that dropping a balance sheet covenant is associated with less than a one month decline in loan maturity. In contrast, there is a significant increase of almost ten months in maturity when the balance sheet covenant is dropped in the Low Risk setting (9.919, t-statistic 7.41).9 The difference between the two groups is -10.908, suggesting that borrowers with high firm-level risk 9 The results for Interest Spread and Maturity for the Low Risk group are consistent with an average improvement in borrower quality; even with the removal of balance sheet covenants, loans are significantly longer while interest spreads are constant. 16 that drop a balance sheet covenants have loans almost one year shorter those who drop in the low risk setting. This is consistent with H2. The results in the next section show a significant decline in income statement covenants for both High and Low Risk firms. The difference-in-differences shows a difference between the groups of -8%, which is mildly significant (t-statistic -1.72) but with the opposite of predicted sign. Similarly, General Covenants decline for all firms, though the decline is larger (though not significantly so) for High Risk firms. In total, the univariate results support H1 and H2, but do not support H3 and H4. 4.2 Multivariate Tests In this section I expand the analysis to include control variables associated with Interest Spread, Maturity, IS Covenants, and General Covenants. In each test, the variable of interest is the interaction between Drop and High Risk. Each regression includes control variables related to the dependent variable. The regressions also include control for industry fixed effects, and standard errors are clustered at the firm level. Results are presented in Table 7. The first column shows results for Interest Spread. Since Interest Spread is a continuous variable, I use ordinary least squares regression. Neither the coefficient on Drop nor High Risk is significantly different from zero. The coefficient on the interaction is positive as predicted and significant: this suggests that loans dropping a balance sheet covenant in a period of increased firm-level demand have, on average, interest spreads 35.3 basis points higher than when the drop is in a period of decreased firm-level demand. This result is consistent with the univariate test, which showed a difference of 74.7 basis points between groups; the attenuation in the size of the coefficient can be attributed to the inclusion of control variables. The control variables show that 17 large firms and firms with higher earnings have lower interest spreads, while borrowers with higher leverage have higher interest spreads. The second column presents similar results for Maturity. As with Interest Spread, I use OLS regression.10 In this case, both the Drop and High Risk have positive and significant coefficients. However, the coefficient on the interaction term is negative and significant, suggesting that dropping a balance sheet covenant when demand is high results in loans approximately 7.4 months shorter than it would be with low firm-level demand. The dependent variable in the third column is a dichotomous variable for use of an income statement covenant, so I use probit regression. The results, consistent with the univariate tests, provide little evidence of income statement covenants substituting for a dropped balance sheet covenant. Specifically, the coefficient on the interaction term is negative but insignificant.11 The fourth column shows regression results for the General Covenant index. Since this variable features a count index of restrictive provisions, I use a negative binomial regression. The results on the interaction are, consistent with the univariate analysis, not significantly different from zero. In total the results support the findings in Table 6: dropping a balance sheet covenant in a period of increasing firm-level demand is associated with higher interest spreads and shorter maturities, but is uncorrelated with use of other types of covenants. 4.3 Economic Impact on Borrowers 10 Maturity is a continuous variable. However, the observed distribution is censored, as maturity cannot be negative: it is not sensible in this context that a loan would mature before it was issued. To examine whether this censoring will impact the inferences from using OLS, I do two things. First, I examine the fitted values of maturity following the OLS regression. I find that only one observation’s predicted value (of 1,248 observations) is negative, suggesting that imposing a lower bound would make only a small difference. Second, I rerun the regression using the tobit model. The results and interpretation are substantively identical. Hence, for ease of discussion and interpretation, I report the OLS regression results in Table 7. 11 As an additional check, I use the Ai and Norton (2003) adjustment to measure the coefficient and z-statistic. The result (-0.012, z-statistic -0.33) yields similar inferences as those reported in Table 7. 18 The regression results suggest two changes to the debt contract when a balance sheet covenant is dropped despite high firm-level demand. First, the average interest spread is 35.3 basis points higher. Second, loans are on average 7.4 months shorter. In this section I evaluate the economic significance of these results for the subsample of affected borrowers. The 282 loans where a balance sheet covenant was dropped in a period of increased firmlevel demand have an average size of $636M. Hence, the interest spread results imply (holding other things equal), an increase in interest expense of $2.25M ($636*0.00353). The average interest expense and net income before discontinued operations and extraordinary items for these firms are $63.4M and $41.1M; the economic impact of the increased interest spread is 3.5% and 5.4% of these figures respectively. If balance sheet covenants are being removed from contracts due to their lack of effectiveness (i.e. either due to changes in borrowers that make the balance sheet less informative for that firm, or changes in accounting standards that make the balance sheet less informative in general), this amount is the real cost borne by equity holders of the borrower. If the borrowers are choosing not to have balance sheet covenants, (i.e. to avoid restrictions on their actions and maintain flexibility, as in Beatty et al. (2002)), this places an approximate price on this flexibility. The impact of decreased maturity is more difficult to quantify. The average maturity for borrowers that dropped a balance sheet covenant with high firm-level demand is 41.8 months. The regression coefficient of -7.4 suggests loans would have maturities of 49.2 months with a balance sheet covenant, holding other things equal. Shorter loans require more frequent renegotiation. The difference in average maturities between the two groups implies that the firms with the shorter loans must get new loans 3.7% more frequently (on an annual basis) than the 19 group with longer loans.12 As such, these borrowers will incur the associated fixed loan fees (upfront fees) with similarly greater frequency. The average upfront fee for these firms, based on data from Dealscan, is 64 basis points. The 3.7% more frequent renegotiation rate implies the short-maturity borrowers will incur additional upfront fees once every 27 years (1 / 0.037 = 27). Dividing 64 basis points by 27 years, shorter maturities lead to an annual cost in added upfront fees of 2.4 basis points, or about $153,000 per year. Given the size of the borrowers and the loans, this amount is not economically significant. However, there are other costs to short maturities that likely do impose real costs on the borrower. Diamond (1991, 1993) and Childs et al. (2005) argue that more frequent renegotiation of debt exposes the borrower to liquidity risk: for example, the lender may decide not to extend further credit to the borrower. Beneish and Press (1993) show that borrowers subject to technical default receive more stringent loan terms in renegotiated loans. In either case, shorter maturities allow the creditor the option to limit their own risk, hence transferring risk to the borrower. While the costs are difficult to quantify, there is considerable research to suggest they are economically meaningful. 5 Additional Tests 5.1 Instrumental Variables Estimation It is commonly held that loan interest spread and maturity are jointly determined (Wittenberg-Moerman 2009). Consistent with this, I find a negative relation between Maturity 12 Loans without a balance sheet covenant have a maturity of 41.8 months, while those with the covenant have 49.2 months. This means loans without balance sheet covenants are about 15% shorter (1 – (41.8 / 49.2)), so renegotiation would take place 15% sooner for loans without covenants. Since the average loan is 4.1 years long, renegotiation is 3.7% more frequent (15% / 4.1 years) on an annual basis. 20 and Interest Spread in the first two regressions in Table 7. To address the potential simultaneity in determination of Interest Spread and Maturity, I use instrumental variable regressions. I start by selecting instruments; these should be associated with the endogenous variable they are serving as instruments for, but uncorrelated with the error term in the underlying structural equation. In the Interest Spread regression, I instrument Maturity using eight variables. I start with two variables used in the main Maturity regression (Asset Maturity and Asset Tangibility) that are related to the asset structure of the borrower. I add another along similar lines, CAPITAL EXPENDITURES, which captures the borrower’s rate of replenishing their fixed assets. I add three indicator variables related to the stated loan purpose: RESTRUCTURING, WORKING CAPITAL and CORPORATE PURPOSES. The purpose of the loan may dictate the timing of its maturity (for example, based on the expect duration of restructuring, or the operating cycle of the firm), but there is no reason to expect the purpose should affect price terms. Finally, I add two instruments related to the growth opportunities of the borrower: MARKET-TO-BOOK and R&D INTENSITY. Market-to-book ratio is a general measure of investment opportunities, while R&D Intensity specifically captures innovations of the firm. I expect borrowers with high levels of either of these variables will have shorter loans; the creditor will require a shorter maturity as a means to monitor the status of these harder-tovalue (relative to fixed assets) opportunities. Following Wittenberg-Moerman (2009), I use EBITDA and Leverage to instrument Interest Spread in the Maturity regressions. I use two-stage least squares (2SLS) to estimate the Interest Spread and Maturity regressions separately. Though potentially less efficient than other methods (e.g. three-stage least squares) that utilize more information by jointly estimating the equations, 2SLS only requires 21 that the instruments be exogenous for the specific equation in which they are used.13 Regression results are presented in Table 8. The first two columns present the 1st and 2nd stage results for Interest Spread. The first stage reports coefficients of all exogenous variables (including all instruments) regressed on the endogenous variable (in this case, Maturity). I run a series of specification tests on the first stage results. First, since the equation is overidentified (there are more instruments than endogenous variables), I calculate the Sargan-Basmann χ2 to confirm the instruments are conditionally exogenous. The statistic has a p-value of 0.3072, so the null hypothesis that the instruments are exogenous cannot be rejected. I next calculate the DurbinWu-Hausman statistic to determine if Maturity is in fact endogenous with Interest Spread. This statistic, which has an F-distribution, has a p-value of 0.0918, consistent with the variables being endogenous. Finally, I examine the strength of the instruments using the Partial R2 and Partial FStatistic of the instruments. The Partial R2 is 3.6%, and the F-Statistic is statistically significant (p-value 0.0003). However, as noted in Larcker and Rusticus (2010), an F-statistic at this level suggests that the instruments may be weak.14 The 2nd stage results are shown in the second column. The coefficients on the main variables are similar to the OLS results: Drop is insignificant, High Risk is negative and weakly significant, and the interaction is positive and significant. The coefficient is higher in this regression, 46.220 versus 35.296. The control variables have similar signs and magnitudes as the OLS coefficients. Notably, the coefficient on Maturity is positive (opposite of OLS) and not significantly different from zero. This is not surprising for two reasons. First, the test of endogeneity is only weakly significant, suggesting that 2SLS represents only a small 13 Three-stage least squares requires the instruments to be exogenous in all the equations that are being jointly estimated (Larcker and Rusticus 2010). 14 Stock et al. (2002), suggests a minimum F-value of 15.09 when using five instruments. With a Partial F-Statistic of 3.762 and eight instruments, these instruments jointly fall below the indicated threshold. 22 improvement over OLS. Second, as noted above, the instruments for Maturity are potentially weak. Hence, the results of this 2SLS regression must be interpreted with caution. The second two columns present results for Maturity. As before, 1st and 2nd stage results and specification tests are presented. The test of overidentifying conditions again fails to reject null, indicating the two instruments are exogenous to Maturity. The Durbin-Wu-Hausman Test rejects the null of no endogeneity. Finally, the instruments appear to be strong: the Partial R2 is 0.169, and the Partial F-Statistic of 68.395 is above the threshold set in Stock et al. (2002). This suggests the results in the 2nd stage are efficient, and represent a significant improvement over OLS. In those results, the coefficient on the interaction term is -6.609; this is lower than the OLS coefficient in absolute terms, but still statistically significant. The signs and magnitudes of other variables are similar between 2SLS and OLS, including Interest Spread. In total, the 2SLS results do not contradict the OLS results. 5.2 Endogenous DROP The previous tests treat Drop as a strictly exogenous variable. This may be true, at least in part; Demerjian (2010a) examines how changing accounting standards, which are arguably exogenous to either borrower or loan features, have affected the inclusion of balance sheet covenants. More likely, though, is that the decision to drop a balance sheet covenant will have some determinants in common with other loan parameters. In this section, I examine the impact on Interest Spread and Maturity when Drop is not exogenous. I start by modeling the determinants of the decision to Drop. While I expect this variable to share determinants with both Interest Spread and Maturity, I do not model these as being simultaneously determined; specifically, Drop should impact Interest Spread and Maturity, but 23 Interest Spread and Maturity do not determine Drop. Since Drop is a dichotomous variable, I use a probit regression with determinants from the Interest Spread and Maturity regressions. I add three other controls likely related to inclusion of balance sheet covenants. COLLATERAL is an indicator with a value of one if the loan is secured. Since Net Worth covenants measure the value of the borrower in the event of liquidation, these may be complementary with collateral requirements. I also include EDF and CHANGE IN EDF. Balance sheet covenants are generally used by riskier, poorer performing firms, so measures of financial distress may predict their inclusion. I also include control variables for year and industry. Probit regression results for the selection equation are presented in the first column of Table 9. The regression yields a fitted value for Drop between zero and one. I code fitted values with a breakpoint of 0.5: observations less than one-half receive a value of zero for PREDICTED DROP, and fitted values of one-half or more receive a value of one. The model accurately classifies 73.4% of the observations. I then use Predicted Drop (both on its own and interacted with High Risk) in the 2SLS regressions. I do not report first stage results and specification tests, which are substantively similar to those reported in Table 8. The results are consistent when using Predicted Drop rather than Drop. The coefficient on the interaction in the Interest Spread regression is positive and significant, suggesting borrowers with high covenant demand who drop the a balance sheet covenant (based on the prediction model) have an interest spread 51.3 basis points higher. Similarly, the maturity is 5.9 months shorter absent a balance sheet covenant. 5.3 Other Sensitivity Tests 24 I run a variety of additional specifications of the main OLS regressions. Summary results (coefficients and t-statistics for the Interest Spread and Maturity regressions) are shown in Table 10, with each test described below. Ranking HIGH RISK In the main tests, the High Risk is defined as a dichotomous variable, with an increase in EDF indicating high firm-level demand for covenants and a decrease indicating low. This approach assumes a constant effect regardless of the size of the change; i.e. creditors will respond equally to a small increase in default risk as to a larger increase. To allow for variation in effects for different size changes, I sort observations into quartiles based on the Change in EDF. I then interact an indicator for each quartile with Drop. I expect to see coefficients increasing (in absolute terms) across the quartiles. Summary results for this piece-wise regression are presented in Table 10, Panel A. The first (most negative) quartile serves as the reference category, so I report coefficients and tstatistics for the second through fourth interaction terms. In the Interest Spread results, the coefficients increase monotonically, being negative in the 2nd quartile, close to zero in the 3rd, and positive in the 4th. The difference between the 2nd and 4th quartile coefficients is statistically significant (based on a Chow Test, reported in the last column). In the Maturity results, the same monotonic pattern holds, with a near-zero change in the 2nd quartile and significant negative decreases in the 3rd and 4th. As with Interest Spread, a test of equality of the 2nd and 4th quartile coefficients shows a significant difference. These results support the main findings using the dichotomous measurement of High Risk. 25 Specific Covenants The main tests group all balance sheet covenants together. To assure the results are not due to one specific covenant, I run the main OLS regressions using only those cases where a specific covenant was dropped. The first three columns in Table 10, Panel B, show summary regression results where borrowers dropped a Net Worth (NW), Leverage (LEV) and Current Ratio (CR) covenant respectively. For Net Worth and Leverage, the coefficients are significant and in the predicted direction, consistent with the main findings. For Current Ratio, the magnitude of the coefficients is similar as for other covenants, but only significant in the Maturity regression. This is likely due to low statistical power, as only about 200 observations involve the current ratio. Rating It common to include the borrower’s Debt Rating as an explanatory variable in regressions of loan parameters. In the main sample, only 42% of the observations have an S&P rating. As a robustness check, I rerun the main regressions sorting observations based on rated status. Summary results, shown in Table 10, Panel B, suggest that the predicted effects are stronger in the unrated subsample than in the rated subsample; however, a Chow test (untabulated) indicates no statistical difference between either of the pairs of coefficients. 6 Conclusion I examine the impact of the sharp decline in use of balance sheet covenants over the period 1996 to 2007. Focusing on borrowers where a balance sheet covenant was dropped even though firm-level demand for contractual protection was high, I find the loss of covenant 26 protection is costly. Specifically, the average firm has an interest spread 35.3 basis points higher and a maturity 7.4 months shorter. I do not find any association between this drop and the use of income statement covenants or restrictive covenants, suggesting these different provisions serve distinct needs. The increase in interest spread costs the average sample borrower $2.25M in additional interest each year, or 5.4% of net income before discontinued operations and extraordinary items. There are a number of forces that lead to lower use of balance sheet covenants even when firm-level demand for covenants is high. These include shifts in the underlying economics of borrowers, shifts in the contracting usefulness of balance sheet information, and changes in the preferences of borrowers for reporting and operating flexibility. While the current study focuses generally on the effects of reduced covenant protection, it does not attempt to distinguish between these different motivations. To the extent that the implications on contract design differ based on these drivers, this is potentially a valuable avenue for future research. 27 Appendix Variable Definitions Variable Name Test Variables Drop Interest Spread Maturity IS Covenant General Covenant High Risk Predicted Drop Borrower Characteristics Firm Size ROA Leverage Rated Asset Maturity Asset Tangibility EDF Change in EDF Capital Expenditures Market-to-Book R&D Intensity Loan Characteristics Loan Size Revolver Institutional Tranche Definition Indicator with a value of one if deal has at least one balance sheet covenant (Net Worth, Leverage, or Current Ratio) All-In Drawn Loan Spread Loan Maturity Date – Loan Inception Date (in months) Indicator with a value of one if the deal has an income statement covenant (Interest Coverage, Fixed Charge Coverage, Debt-to-Earnings) An index with a value of one for restrictions on: dividend payment, asset sales, debt issuance, equity issuance, and use of insurance proceeds An indicator with a value of one if EDF increased from the Pre-Drop to Drop period Based on the fitted value from a regression of Drop on a set of controls; an indicator with a value of one if the fitted value is greater than or equal to 0.5, and zero if the fitted value is less than 0.5 Source Dealscan Dealscan Dealscan Dealscan Dealscan Dealscan / Compustat / CRSP Dealscan / Compustat / CRSP The natural logarithm of total assets: ln(ATQ) EBITDA scaled by average total assets: OIBDPQ / ATQ Total debt scaled by total asset: (DLCQ+DLTTQ) / ATQ An indicator with a value of one if the borrower has an S&P Senior Unsecured Debt Rating (SPLTICRM) The weighted average of receivable age and fixed asset age: (ACTQ/(ACTQ+PPENTQ)* (PPENTQ/COGSQ) ) + (PPENTQ / (ACTQ+ PPENTQ)*(PPENTQ/DPY)) Ratio of tangible assets: (PPENTQ + INVTQ) / ATQ Expected Default Frequency, based on Merton (1974), calculated based on Hillegeist et al. (2004) The annual change in EDF Capital expenditures scaled by total assets: CAPXY / ATQ The market value of equity over the book value of equity: CSHOQ*PRCCQ / CEQQ Ratio of research and development expenditures to total assets: XRDQ / ATQ Compustat ln(loan amount) An indicator with a value of one if some portion of the deal is a revolving line of credit An indicator with a value of one if the loan has Dealscan Compustat Compustat Compustat Compustat Compustat Compustat / CRSP Compustat / CRSP Compustat Compustat Compustat Dealscan Dealscan 28 Performance Pricing Lenders Collateral Restructuring Working Capital Corporate Purposes a term loan tranche labeled ‘B’ or higher An indicator with a value of one if the loan includes a performance pricing provision The number of syndicate members An indicator with a value of one if the loan requires collateral An indicator with a value of one if the stated purpose of the loan is restructuring An indicator with a value of one if the stated purpose of the loan is working capital An indicator with a value of one if the stated purpose of the loan is corporate reasons Dealscan Dealscan Dealscan Dealscan Dealscan Dealscan Notes: All Compustat variables are from the Xpressfeed quarterly data. All income statement variables are annualized by summing the current and prior three quarterly observations. Compustat variables are winsorized at the top and bottom 1% of observations. 29 References Ai, C., Norton, E., 2003. Interaction terms in logit and probit models. Economics Letters 80: 123-129. Asquith, P., Beatty, A., Weber, J., 2005. Performance pricing in bank debt contracts. Journal of Accounting and Economics 40: 101-128. Barclay, M., Smith, C., 1995. The maturity structure of corporate debt. Journal of Finance 50(2): 609-631. Beatty, A., Chamberlain, S., Magliolo, J., 1995. Managing financial reports of commercial banks: The influence of taxes, regulatory capital, and earnings. Journal of Accounting Research 33(2): 231-261. Beatty, A., Ramesh, K., Weber, J., 2002. The importance of accounting changes in debt contracts: The cost of flexibility in covenant calculations. Journal of Accounting and Economics 33: 205-227. Beneish, M., Press, E., 1993. Costs of technical violation of accounting-based debt covenants. The Accounting Review 68(2): 233-257. Billett, M., King, T., Mauer, D., 2007. Growth opportunities and the choice of leverage, debt maturity and covenants. The Journal of Finance 62(2): 697-730. Bradley, M., Roberts, M., 2004. The structure and pricing of corporate debt covenants. Working Paper, Duke University. Chan, Y., Kanatas, G., 1985. Asymmetric valuations and the role of collateral in loan agreements. Journal of Money, Credit, and Banking 17(1): 84-95. Chava, S., Kumar, P., Warga, A., 2009. Managerial agency and bond covenants. Review of Financial Studies 23(3): 1120-1148. Childs, P., Mauer, D., Ott, S., 2005. Interactions of corporate financing and investment decisions: The effects of agency conflicts. Journal of Financial Economics 76: 667-690. Christensen, H., Nikolaev, V., 2010. Capital versus performance covenants in debt contracts. Working Paper, University of Chicago. Demerjian, P. 2010a. Accounting standards and debt covenants: Has the “balance sheet approach” damaged the balance sheet? Working Paper, Emory University. Demerjian, P. 2010b. Financial covenants, credit risk, and the resolution of uncertainty. Working Paper, Emory University. 30 Demiroglu, C., James, C., 2010. The information content of bank loan covenants. The Review of Financial Studies 23(10): 3700-3737. Diamond, D., 1991. Debt maturity structure and liquidity risk. Quarterly Journal of Economics 106(3): 709-737. Diamond, D., 1993. Seniority and maturity of debt contracts. Journal of Financial Economics 33: 341-368. Dichev, I., Skinner, D., 2002. Large-sample evidence on the debt covenant hypothesis. Journal of Accounting Research 40(4): 1091-1123. Drucker, S., Puri, M., 2009. On loan sales, loan contracting, and lending relationships. The Review of Financial Studies 22(7): 2835-2872. Hillegeist, S., E. Keating, D. Cram, and K. Lundstedt. 2004. Assessing the probability of bankruptcy. Review of Accounting Studies 9: 5-34. Holthausen, R., Watts, R., 2001. The relevance of the value-relevance literature for financial accounting standard setting. Journal of Accounting and Economics 31: 3-75. Kothari, S., Ramanna, K., Skinner, D., 2010. Implications for GAAP from an analysis of positive research in accounting. Working Paper. MIT. Larcker, D., Rusticus, T., 2010. On the use of instrumental variables in accounting research. Journal of Accounting and Economics 49: 186-205. Merton, R. 1974. On the pricing of corporate debt: The risk structure of interest rates. The Journal of Finance XXIX(2): 449-470. Reisel, N., 2006. On the value of restrictive covenants: An empirical investigation of public bond issues. Working Paper, Southern Methodist University. Skinner, D., 1993. The investment opportunity set and accounting procedure choice. Journal of Accounting and Economics 16: 407-445. Smith, C., Warner, J., 1979. On financial contracting. Journal of Financial Economics 7: 117161. Stock, J., Wright, J., Yogo, M., 2002. A survey of weak instruments and weak identification in generalized method of moments. Journal of Business & Economics Statistics 20: 518-529. Stulz, R., Johnson, H., 1985. An analysis of secured debt. Journal of Financial Economics 14: 501-521. 31 Watts, R., 2003. Conservatism in accounting Part I: Explanations and implications. Accounting Horizons 17: 207-221. Wittenberg-Moreman, R., 2009. The impact of information asymmetry on debt pricing and maturity. Working Paper, University of Chicago. 32 Table 1 Sample Selection Starting - borrowers with < 2 loans - borrowers with no BS covenant - borrowers with no dropped BS covenant - borrowers with multiple drops Final Sample Deals Borrowers 9,000 - 1,028 7,972 -3,542 4,430 -2,758 1,672 -214 1,458 3,087 -1,028 2,059 -474 1,585 -806 779 -50 729 Notes to Table 1: This table presents the sample selection process. I start with the intersection of LPC/Dealscan and Compustat for 1996 through 2007, a total of 9,000 loan packages (“Deals”) to 3,087 borrowers. The first step removes borrowers with a single loan over the sample period. The second step removes borrowers that never had a balance sheet covenant (net worth, leverage, or current ratio) in a deal over the sample period. The third step removes borrowers who never “dropped” a balance sheet covenant; that is, borrowers who have balance sheet covenants in all their deals. The final step removes observations of borrowers with multiple drops; that is, borrowers who have a balance sheet covenant dropped at some point during the sample period, and later have a loan package including a balance sheet covenant. The final sample consists of 729 borrowers and 1,458 loan packages. 33 Table 2 Distribution of Sample Deals Year 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Total Pre-Drop 65 71 57 78 71 101 99 81 63 34 8 1 729 Drop 2 16 29 27 47 60 78 92 142 84 96 56 729 Notes to Table 2: This table tracks the number of sample observations by year. PRE-DROP observations are deals with a balance sheet covenant (net worth, leverage, or current ratio). DROP observations are deals with no balance sheet covenant. 34 Table 3 Descriptive Statistics Test Variables Drop Interest Spread Maturity IS Covenant General Covenant High Risk Mean 0.500 194.223 42.498 0.769 2.259 0.423 Borrower Characteristics Mean Firm Size ROA Leverage Rated Asset Maturity Asset Tangibility 6.474 0.137 0.280 0.420 5.891 0.456 Loan Characteristics Mean Loan Size Revolver Institutional Tranche Performance Pricing Lenders Collateral 5.291 0.942 0.140 0.722 8.441 0.595 Standard Deviation 0.500 136.223 19.403 0.422 1.872 0.494 Standard Deviation 1.679 0.103 0.202 0.494 19.376 0.243 Standard Deviation 1.727 0.234 0.347 0.448 8.990 0.491 1st quartile 0.000 87.500 30.000 1.000 1.000 0.000 1st quartile 5.391 0.085 0.126 0.000 1.463 0.263 1st quartile 4.151 1.000 0.000 0.000 2.000 0.000 Median 0.500 175.000 37.120 1.000 1.000 0.000 Median 6.461 0.133 0.265 0.000 2.615 0.449 Median 5.317 1.000 0.000 1.000 6.000 1.000 3rd quartile 1.000 265.625 60.000 1.000 5.000 1.000 3rd quartile 7.535 0.188 0.395 1.000 6.035 0.650 3rd quartile 6.477 1.000 0.000 1.000 11.000 1.000 Notes to Table 3: This table presents descriptive statistics on test variables, borrower characteristics, and loan characteristics. DROP is an indicator with a value of one if the observation has no balance sheet covenants (net worth, leverage, or current ratio) and zero otherwise. INTEREST SPREAD is the interest rate spread above the index (usually LIBOR or Prime) charged to the borrower. MATURITY is the stated duration of the loan in months. IS COVENANT is an indicator for inclusion of an income statement covenant (interest coverage, fixed charge coverage, or debt-to-earnings). GENERAL COVENANT is an index of restrictive contract provisions, including restrictions on dividend payments, debt issuance, equity issuance, asset sales, and use of insurance proceeds. HIGH RISK is an indicator with a value of one if EDF increases when the balance sheet covenant is dropped, and zero otherwise. FIRM SIZE is the natural logarithm of total assets. ROA is EBITDA scaled by average total assets. LEVERAGE is total debt scaled by total assets. RATED is an indicator if the borrower has an S&P Senior Debt Rating. ASSET MATURITY is the weighted average of receivable age (accounts receivable / cost of goods sold) and fixed asset age (PP&E / depreciation). ASSET TANGIBILITY is inventory plus PP&E scaled by total assets. LOAN SIZE is the natural logarithm of the amount of the loan. REVOLVER is an indicator for if the loan package contains a revolving line of credit component. INSTITUTIONAL TRANCHE is an indicator for if loan has a tranche titled “Term Loan B” or higher. PERFORMANCE PRICING is an indicator for if the loan has a performance pricing provision. LENDERS is the number of syndicate members. COLLATERAL is an indicator if the loan requires collateral. Detailed variable definitions are provided in the Appendix. All borrower characteristics other than Firm Size and Rated are winsorized at the top and bottom 1% of observations. 35 Table 4 Correlation Matrix -0.07 -0.06 0.11 -0.03 -0.01 0.03 -0.07 -0.02 0.01 -0.01 -0.01 0.08 0.01 0.16 0.67 0.28 0.08 0.79 -0.06 0.01 0.20 0.68 -0.43 -0.14 -0.03 0.07 0.06 0.19 0.05 -0.02 0.19 0.19 -0.18 0.31 0.34 0.24 0.16 -0.08 0.22 -0.01 0.20 0.15 0.24 0.08 0.52 -0.07 0.11 0.13 0.49 -0.20 -0.04 -0.01 0.03 -0.06 0.00 0.03 0.08 0.08 0.22 0.08 0.16 0.05 -0.45 0.25 0.08 0.08 -0.07 0.79 0.21 0.13 0.51 0.06 0.12 -0.07 -0.18 0.04 0.12 0.02 -0.02 -0.06 0.06 -0.08 -0.07 -0.16 -0.02 0.08 0.13 0.29 0.32 0.10 0.36 0.01 0.00 -0.01 0.23 0.11 0.07 -0.02 0.10 -0.11 -0.12 -0.33 0.28 0.37 0.19 -0.01 0.21 0.20 -0.04 0.13 -0.05 0.01 0.37 0.20 -0.02 0.64 0.25 -0.14 0.05 0.01 0.20 0.11 0.12 -0.02 -0.02 0.01 0.09 -0.05 0.06 0.11 0.34 0.81 -0.36 -0.11 0.20 0.13 -0.01 -0.02 0.09 0.26 0.31 -0.09 Collateral 0.11 -0.06 0.03 0.21 0.01 0.01 0.00 0.08 0.03 0.33 0.25 0.11 0.32 0.02 0.03 0.06 -0.06 -0.02 0.00 0.06 -0.08 0.05 0.07 Lenders 0.30 -0.02 -0.09 0.18 0.06 -0.08 -0.03 -0.05 0.07 0.12 0.10 0.37 0.13 0.08 0.05 -0.17 0.05 -0.08 0.00 -0.03 0.64 -0.02 0.28 Perf. Price 0.26 0.26 -0.03 0.08 0.21 0.02 0.05 0.04 0.02 0.28 0.05 0.32 0.28 0.27 0.06 0.02 0.26 0.00 0.04 0.21 0.13 0.14 -0.10 Inst. Tranche -0.10 -0.00 0.22 0.08 -0.49 -0.39 0.25 -0.20 -0.06 -0.01 -0.49 -0.12 0.33 -0.29 -0.40 0.62 -0.10 -0.35 0.18 0.20 0.05 -0.04 0.05 Revolver 0.08 -0.41 0.06 -0.11 -0.06 -0.07 Loan Size 0.00 0.10 -0.03 -0.02 0.11 Asset Tang. Firm Size -0.06 0.18 0.25 0.25 Asset Mat. High Risk -0.14 -0.07 0.26 Rated Gen Cov 0.14 -0.12 Leverage IS Cov 0.10 ROA Maturity 0.08 0.14 -0.14 -0.07 0.00 0.08 -0.10 0.00 0.05 -0.05 -0.05 0.04 -0.07 0.13 -0.12 -0.02 0.03 Int. Spread Drop Drop Int. Spread Maturity IS Cov Gen Cov High Risk Firm Size ROA Leverage Rated Asset Mat. Asset Tang. Loan Size Revolver Inst. Tranche Perf. Price Lenders Collateral -0.06 -0.23 0.18 0.07 0.12 0.02 0.52 0.15 0.14 0.41 0.02 0.06 0.63 0.06 0.17 0.20 0.03 0.51 0.06 0.08 0.32 0.08 -0.43 -0.18 0.15 -0.20 -0.00 -0.05 -0.36 -0.01 0.26 -0.09 -0.21 -0.32 Notes to Table 4: This table presents simple correlations. Pearson correlations are in the upper triangle, with Spearman rank correlations in the lower. DROP is an indicator with a value of one if the observation has no balance sheet covenants (net worth, leverage, or current ratio) and zero otherwise. INTEREST SPREAD is the interest rate spread above the index (usually LIBOR or Prime) charged to the borrower. MATURITY is the stated duration of the loan in months. IS COVENANT is an indicator for inclusion of an income statement covenant (interest coverage, fixed charge coverage, or debt-to-earnings). GENERAL COVENANT is an index of restrictive contract provisions, including restrictions on dividend payments, debt issuance, equity issuance, asset sales, and use of insurance proceeds. HIGH RISK is an indicator with a value of one if EDF increases when the balance sheet covenant is dropped, and zero otherwise. FIRM SIZE is the natural logarithm of total assets. ROA is EBITDA scaled by average total assets. LEVERAGE is total debt scaled by total assets. RATED is an indicator if the borrower has an S&P Senior Debt Rating. ASSET MATURITY is the weighted average of receivable age (accounts receivable / cost of goods sold) and fixed asset age (PP&E / depreciation). ASSET TANGIBILITY is inventory plus PP&E scaled by total assets. LOAN SIZE is natural logarithm of the amount of the loan. REVOLVER is an indicator for if the loan package contains a revolving line of credit component. INSTITUTIONAL TRANCHE is an indicator for if loan has a tranche titled “Term Loan B” or higher. PERFORMANCE PRICING is an indicator for if the loan has a performance pricing provision. LENDERS is the number of syndicate members. COLLATERAL is an indicator if the loan requires collateral. Detailed variable definitions are provided in the Appendix. All borrower characteristics other than Firm Size and Rated are winsorized at the top and bottom 1% of observations. Statistically significant correlations (<10% level) are in boldface. 36 Table 5 Descriptive Analysis Test Variables Interest Spread Maturity Income Statement Covenant General Covenant Index mean median mean median mean median mean median Borrower Characteristics Firm Size ROA Leverage Rated Asset Maturity Asset Tangibility mean median mean median mean median mean median mean median mean median Loan Characteristics Loan Size Revolver Institutional Tranche Performance Pricing Lenders Collateral mean median mean median mean median mean median mean median mean median Drop 208.044 192.500 45.119 48.000 0.712 1.000 2.148 1.000 Drop 6.605 6.571 0.127 0.125 0.284 0.262 0.443 0.000 6.239 2.432 0.310 0.233 Drop 5.374 5.416 0.926 1.000 0.184 0.000 0.669 1.000 7.925 6.000 0.608 1.000 Pre-Drop 180.401 150.000 39.877 36.000 0.826 1.000 2.369 1.000 Pre-Drop 6.344 6.306 0.148 0.142 0.275 0.268 0.396 0.000 5.538 2.852 0.323 0.252 Pre-Drop 5.208 5.298 0.957 1.000 0.096 0.000 0.774 1.000 8.957 6.000 0.583 1.000 Difference 27.643 42.500 5.241 12.000 -0.114 0.000 -0.221 0.000 Difference 0.262 0.265 -0.020 -0.017 0.009 -0.006 0.047 0.000 0.701 -0.420 -0.013 -0.019 Difference 0.166 0.118 -0.032 0.000 0.088 0.000 -0.104 0.000 -1.033 0.000 0.025 0.000 p-value 0.000 0.004 <0.0001 <0.0001 <0.0001 <0.0001 0.024 0.005 p-value 0.0030 0.0020 0.0050 0.0005 0.3879 0.9305 0.0713 0.0713 0.5076 0.0923 0.3028 0.1728 p-value 0.0662 0.0934 0.0101 0.0102 <0.0001 <0.0001 <0.0001 <0.0001 0.0282 0.4244 0.3372 0.3370 Notes to Table 5: This table presents differences in test and control variables based on inclusion of balance sheet covenants. DROP observations are those without a balance sheet covenant. PRE-DROP observations are those from the period immediately preceding the balance sheet covenant drop. DIFFERENCE is the average difference between the Drop and Pre-Drop periods, with a p-value of the significance based on a t-test (for means) or Wilcoxon test (for medians). INTEREST SPREAD is the interest rate spread above the index (usually LIBOR or Prime) charged to the borrower. MATURITY is the stated duration of the loan in months. IS COVENANT is an indicator for inclusion of an income statement covenant (interest coverage, fixed charge coverage, or debt-to-earnings). GENERAL COVENANT is an index of restrictive contract provisions, including restrictions on dividend payments, debt issuance, equity issuance, asset sales, and use of insurance proceeds. HIGH RISK is an indicator with a value of one if EDF increases when the balance sheet covenant is dropped, and zero otherwise. FIRM SIZE is the natural logarithm of total assets. ROA is EBITDA scaled by average total assets. LEVERAGE is total debt scaled by total assets. RATED is an indicator if the borrower has an S&P Senior Debt Rating. ASSET MATURITY is the weighted average of receivable age (accounts receivable / cost of goods sold) and fixed asset age (PP&E / depreciation). ASSET TANGIBILITY is inventory plus PP&E scaled by total assets. LOAN SIZE is natural logarithm of the amount of the loan. REVOLVER is an indicator for if the loan package contains a revolving line of credit component. INSTITUTIONAL TRANCHE is an indicator for if loan has a tranche titled “Term Loan B” or higher. PERFORMANCE PRICING is an indicator for if the loan has a performance pricing provision. LENDERS is the number of syndicate members. COLLATERAL is an indicator if the loan requires collateral. Detailed variable definitions are provided in the Appendix. All borrower characteristics other than Firm Size and Rated are winsorized at the top and bottom 1% of observations. Statistically significant differences (<10% level) are in boldface. 37 Table 6 Univariate Difference-in-Differences Analysis Interest Spread Obs Mean Drop 282 237.478 High Risk Pre-Drop Difference 282 168.229 69.249*** t-statistic 5.73 p-value <0.0001 Drop 385 174.344 Low Risk Pre-Drop Difference 385 179.834 -5.49 t-statistic -0.65 p-value 0.5187 Difference-in-Differences (High Risk – Low Risk) predicted sign + Diff-in-Diff 74.739*** t-statistic 5.06 p-value <0.0001 Low Risk Pre-Drop Difference 385 38.367 9.919*** t-statistic 7.41 p-value <0.0001 Difference-in-Differences (High Risk – Low Risk) predicted sign Diff-in-Diff -10.908*** t-statistic -5.20 p-value <0.0001 Low Risk Pre-Drop Difference 385 0.821 -0.086*** t-statistic -2.88 p-value 0.0040 Difference-in-Differences (High Risk – Low Risk) predicted sign + Diff-in-Diff -0.080† t-statistic -1.72 p-value 0.0846 Low Risk Pre-Drop Difference 385 2.132 -0.093 t-statistic -0.71 p-value 0.4756 Difference-in-Differences (High Risk – Low Risk) predicted sign + Diff-in-Diff -0.311 t-statistic -1.49 p-value 0.1350 Maturity Obs Mean Drop 282 41.809 High Risk Pre-Drop Difference 282 42.798 -0.989 t-statistic -0.61 p-value 0.5404 Drop 385 48.286 IS Covenant Obs Mean Drop 282 0.674 High Risk Pre-Drop Difference 282 0.840 -0.166*** t-statistic -4.67 p-value <0.0001 Drop 385 0.735 General Covenant Obs Mean Drop 282 2.305 High Risk Pre-Drop Difference 282 2.709 -0.404** t-statistic -2.49 p-value 0.0127 Drop 385 2.039 Notes to Table 6: This table presents difference-in-difference analysis on four loan parameters. DROP observations are those without a balance sheet covenant. PRE-DROP observations are those from the period immediately preceding the balance sheet covenant drop. HIGH RISK are those observations where default risk (EDF) increased from the Pre-Drop to the Drop period. LOW RISK are those observations where default risk stayed the same or decreased from the Pre-Drop to the Drop period. INTEREST SPREAD is the interest rate spread above the index (usually LIBOR or Prime) charged to the borrower. MATURITY is the stated duration of the loan in months. IS COVENANT is an indicator for inclusion of an income statement covenant (interest coverage, fixed charge coverage, or debt-to-earnings). GENERAL COVENANT is an index of restrictive contract provisions, including restrictions on dividend payments, debt issuance, equity issuance, asset sales, and use of insurance proceeds. The table sorts firms into High Risk and Low Risk groups, and measures the difference between the Drop and Pre-Drop observations within these groups. The difference-in-differences captures the Drop – Pre-Drop difference between the High Risk and Low Risk groups. ***, **, and * indiciate statistical significance at the 1%, 5% and 10% level of significance. † indicates statistical significance in the opposite of predicted direction at the 10% level. 38 Table 7 Multivariate Regression Analysis Dependent Variable = Drop High Risk Drop * High Risk Maturity Interest Spread 3.138 (0.40) -11.633 (-1.37) 35.296*** (2.94) -0.576*** (-3.11) Interest Spread IS Covenant Firm Size ROA Leverage Rated 0.693 (0.08) -27.695*** (-7.35) -319.780*** (-9.87) 150.984*** (8.73) 0.556 (0.07) Asset Maturity Asset Tangibility Loan Size Revolver Institutional Tranche Performance Pricing Lenders Maturity 8.390*** (6.90) 3.437*** (2.59) -7.388*** (-3.91) -0.021*** (-4.70) 7.374*** (6.11) -3.906*** (-6.78) -0.260 (-0.21) 1.210 (1.33) 0.250 (0.08) 4.435*** (8.79) -11.875*** (-3.32) -61.738*** (-4.39) 96.222*** (9.96) -21.427*** (-2.69) 0.645 (1.48) 14.619*** (9.51) 526.265*** (9.91) 1,122 OLS 0.47 45.187*** (5.66) 1,248 OLS 0.28 Collateral Constant Observations Regression Model Adjusted R2 Concordance IS Covenant -0.292** (-2.21) -0.155 (-1.01) -0.028 (-0.13) General Covenant -0.086 (-1.45) 0.191*** (3.13) -0.090 (-1.04) -0.362*** (-5.43) 1.968*** (3.56) 0.970*** (3.13) -0.229 (-1.52) -0.061** (-2.06) 0.249 (0.99) 0.537*** (4.38) -0.096 (-1.61) 0.131** (2.14) -0.098 (-0.46) 0.574*** (2.88) 1.174*** (9.65) 0.034*** (3.28) 0.034 (0.26) 0.677 (0.83) 1,088 Probit 0.087*** (3.24) -0.111 (-1.04) 0.477*** (7.88) 0.418*** (6.92) 0.001 (0.47) 0.450*** (8.07) -0.216 (-0.49) 1,122 Negative Binomial 0.83 Notes to Table 7: This table presents multivariate regression results. There are four regressions for the four different dependent variables. Reported results include coefficient estimates and t-statistics (for OLS) or z-statistics (for Probit and Negative Binomial). INTEREST SPREAD is the interest rate spread above the index (usually LIBOR or Prime) charged to the borrower. MATURITY is the stated duration of the loan in months. IS COVENANT is an indicator for inclusion of an income statement covenant (interest coverage, fixed charge coverage, or debt-to-earnings). GENERAL COVENANT is an index of restrictive contract provisions, including restrictions on dividend payments, debt issuance, equity issuance, asset sales, and use of insurance proceeds. DROP is an indicator with a value of one if the observation has no balance sheet covenants (net worth, leverage, or current ratio) and zero otherwise. HIGH RISK is an indicator with a value of one if EDF increases when the balance sheet covenant is dropped, and zero otherwise. FIRM SIZE is the natural logarithm of total assets. ROA is EBITDA scaled by average total assets. LEVERAGE is total debt scaled by total assets. RATED is an indicator if the borrower has an S&P Senior Debt 39 Rating. ASSET MATURITY is the weighted average of receivable age (accounts receivable / cost of goods sold) and fixed asset age (PP&E / depreciation). ASSET TANGIBILITY is inventory plus PP&E scaled by total assets. LOAN SIZE is the natural logarithm of the amount of the loan. REVOLVER is an indicator for if the loan package contains a revolving line of credit component. INSTITUTIONAL TRANCHE is an indicator for if loan has a tranche titled “Term Loan B” or higher. PERFORMANCE PRICING is an indicator for if the loan has a performance pricing provision. LENDERS is the number of syndicate members. COLLATERAL is an indicator if the loan requires collateral. Concordance is the percentage of observations correctly classified in the probit model. Detailed variable definitions are provided in the Appendix. All borrower characteristics other than Firm Size and Rated are winsorized at the top and bottom 1% of observations. Each regression includes industry fixed effects, and standard errors are clustered by borrower. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels respectively. 40 Table 8 Instrumental Variable Regressions Drop High Risk Drop * High Risk Maturity Interest Spread 1st Stage 2nd Stage 9.105*** -13.666 (7.84) (-1.13) 2.464* -16.694* (1.67) (-1.93) -7.434*** 46.220*** (-3.86) (3.42) 1.193 (1.13) Maturity 1st Stage 2nd Stage 1.178 8.557*** (0.19) (7.53) -11.648 2.678* (-1.39) (1.89) 35.688*** -6.609*** (3.01) (-3.32) Interest Spread IS Covenant Firm Size ROA Leverage Rated Asset Maturity Asset Tangibility Loan Size Revolver Institutional Tranche Performance Pricing Lenders Purpose: Restructuring Purpose: Working Capital Purpose: Corporate Market-to-Book R&D Intensity Capital Expenditures Constant Observations R2 Sargan-Basmann statistic (χ2, p-value) Durbin-Wu-Hausman statistic (F, p-value) Partial R2 Partial F-statistic (F, p-value) 5.405*** -9.732 (3.37) (0.86) -2.650*** -24.562*** (-3.50) (-4.28) 14.841*** -370.985*** (2.47) (-8.89) -7.085** 162.320*** (-2.15) (6.19) -0.615 0.306 (-0.41) (0.03) 1.929 (1.57) 3.925 (1.00) 3.830*** -16.117*** (5.64) (-2.61) 0.829 -80.419*** (0.22) (-2.81) 13.495*** 70.284*** (7.73) (3.66) 5.317*** -34.726*** (3.56) (-2.82) 0.003 0.540 (0.04) (1.10) 9.014*** (4.54) 6.787*** (3.63) 5.496*** (2.77) -0.270* (-1.83) -11.835 (-1.50) -10.267 (-1.07) 18.552** 496.857*** (2.28) (7.28) 1,046 1,046 0.33 0.47 8.297, 0.3072 2.851, 0.0918 0.036 3.762, 0.0003 -12.751 (-1.35) -23.504*** (-5.10) -347.349*** (-8.74) 159.901*** (6.30) 0.720 (0.09) 7.236 (0.86) 28.900 (1.23) -15.998*** (-4.07) 95.425*** (7.63) -0.044*** (-3.75) 7.289*** (4.91) -4.417*** (-5.73) 0.582 (0.40) 1.155 (0.96) 1.258 (0.35) 3.862*** (5.91) 17.938*** (8.39) 420.689*** 47.710*** (7.26) (5.07) 1,098 1,098 0.46 0.32 0.004, 0.9481 3.824, 0.0510 0.169 68.395, <0.0001 Notes to Table 8: This table presents instrumental variable regressions for Interest Spread and Maturity, using two-stage least squares. Reported results include coefficient estimates and t-statistics (for 1st stage regressions) or z-statistics (for 2nd stage regressions). INTEREST SPREAD is the interest rate spread above the index (usually LIBOR or Prime) charged to the borrower. MATURITY is the stated duration of the loan in months. DROP is an indicator with a value of one if the observation has no 41 balance sheet covenants (net worth, leverage, or current ratio) and zero otherwise. HIGH RISK is an indicator with a value of one if EDF increases when the balance sheet covenant is dropped, and zero otherwise. IS COVENANT is an indicator for inclusion of an income statement covenant (interest coverage, fixed charge coverage, or debt-to-earnings). FIRM SIZE is the natural logarithm of total assets. ROA is EBITDA scaled by average total assets. LEVERAGE is total debt scaled by total assets. RATED is an indicator if the borrower has an S&P Senior Secured Debt Rating. ASSET MATURITY is the weighted average of receivable age (accounts receivable / cost of goods sold) and fixed asset age (PP&E / depreciation). ASSET TANGIBILITY is inventory plus PP&E scaled by total assets. LOAN SIZE is the natural logarithm of the amount of the loan. REVOLVER is an indicator for if the loan package contains a revolving line of credit component. INSTITUTIONAL TRANCHE is an indicator for if loan has a tranche titled “Term Loan B” or higher. PERFORMANCE PRICING is an indicator for if the loan has a performance pricing provision. LENDERS is the number of syndicate members. COLLATERAL is an indicator if the loan requires collateral. PURPOSE: RESTRUCTURING is an indicator for if the loan’s purpose is restructuring or reorganization. PURPOSE: WORKING CAPITAL is an indicator for if the loan’s purpose is working capital. PURPOSE: CORPORATE is an indicator for if the loan’s purpose is related to general corporate operations. MARKET-TO-BOOK is the ratio of the market value of equity to the book value of equity. R&D INTENSITY is the ratio of research and development expenditures to sales. CAPITAL EXPENDITURES is capital expenditures scaled by total assets. There are four specification tests for the 1st stage regressions. The Sargan-Basmann statistic tests for overidentifying restrictions: a significant statistic indicates the instruments may not be exogenous. The Dubin-Wu-Hausman statistic tests whether the presumed endogenous variable is in fact endogenous: a significant coefficient indicates endogeneity. The Partial R2 and Partial F-statistic test the joint significance of the set of instruments. Detailed variable definitions are provided in the Appendix. All borrower characteristics other than Firm Size and Rated are winsorized at the top and bottom 1% of observations. Each regression includes industry fixed effects, and standard errors are clustered by borrower. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels respectively. 42 Table 9 Endogenous DROP 1st Stage Selection Equation Dependent Variable = Drop Predicted Drop High Risk 0.337*** (3.26) Predicted Drop * High Risk Maturity Interest Spread Income Statement Covenant Firm Size ROA Leverage Rated Asset Maturity Asset Tangibility Loan Size Revolver Institutional Tranche Performance Pricing Lenders Collateral EDF Change in EDF Constant Observations Concordance R2 -0.021 (-0.35) -1.007* (-1.87) 0.212 (0.74) 0.080 (0.62) -0.172* (-1.91) -0.274 (-0.91) 0.086 (1.56) 0.200 (0.92) 0.392*** (2.67) -0.640*** (-5.44) -0.010 (-1.42) -0.070 (-0.63) 1.029** (2.37) -0.263 (-0.73) -1.653* (-1.76) 1,098 0.73 2nd Stage IV Regressions Dependent Variable = Interest Spread Maturity -12.124 8.681*** (-0.95) (6.85) -18.837** 2.243 (-2.22) (1.48) 51.332*** -5.885*** (3.31) (-2.63) 1.194 (1.140 -0.047*** (-3.96) -11.136 7.445*** (-1.01) (4.98) -25.168*** -4.656*** (-4.31) (-5.90) -366.078*** (-8.83) 164.504*** (6.43) 2.354 0.311 (0.26) (0.21) 1.321 (1.11) 2.310 (0.63) -16.261*** 3.918*** (-2.69) (6.01) -80.897*** (-2.85) 67.703*** 17.785*** (3.54) (8.29) -32.628** (-2.53) 0.602 (1.23) 498.507*** (7.26) 1,046 48.848*** (5.21) 1,098 0.48 0.31 Notes to Table 9: This table presents instrumental variable regression results for Interest Spread and Maturity when Drop is replaced by a predicted value based on various borrower and loan characteristics. Reported results include coefficient estimates and z-statistics. INTEREST SPREAD is the interest rate spread above the index (usually LIBOR or Prime) charged to the borrower. MATURITY is the stated duration of the loan in months. DROP is an indicator with a value of one if the observation has no balance sheet covenants (net worth, leverage, or current ratio) and zero otherwise. HIGH RISK is an indicator with a value of one if EDF increases when the balance sheet covenant is dropped, and zero otherwise. IS COVENANT is an indicator for inclusion of an income statement covenant (interest coverage, fixed charge coverage, or debt-to-earnings). FIRM SIZE is the natural logarithm of total assets. ROA is EBITDA scaled by average total assets. LEVERAGE is total debt scaled by total assets. RATED is an indicator if the borrower has an S&P Senior Debt Rating. ASSET MATURITY is the weighted average of 43 receivable age (accounts receivable / cost of goods sold) and fixed asset age (PP&E / depreciation). ASSET TANGIBILITY is inventory plus PP&E scaled by total assets. LOAN SIZE is the natural logarithm of the amount of the loan. REVOLVER is an indicator for if the loan package contains a revolving line of credit component. INSTITUTIONAL TRANCHE is an indicator for if loan has a tranche titled “Term Loan B” or higher. PERFORMANCE PRICING is an indicator for if the loan has a performance pricing provision. LENDERS is the number of syndicate members. COLLATERAL is an indicator if the loan requires collateral. EDF is the likelihood of default based on the Merton (1974) model. CHANGE IN EDF is the annual change in EDF. PREDICTED DROP is the based on the 1st stage regression of Drop on borrower and loan variables; predicted values less than 0.5 receive a value of zero for Predicted Drop, while predicted value greater than or equal to 0.5 receive a value of one. Concordance is the percentage of observations correctly classified in the probit model. All borrower characteristics other than Firm Size and Rated are winsorized at the top and bottom 1% of observations. Each regression includes industry fixed effects, and standard errors are clustered by borrower. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels respectively. 44 Table 10 Sensitivity Tests Panel A: Piece-Wise Regressions based on Quartile of Change in EDF Quartile 2 3 Interest Spread Coefficient -5.619 -21.562 (N = 1,122) t-statistic (-1.74) (-0.40) Maturity Coefficient 0.516 -6.876 (N = 1,248) t-statistic (0.23) (-3.02) Panel B: Other Sensitivity Tests Specification: predicted sign Coefficient Interest + t-statistic Spread Observations Coefficient t-statistic Maturity Observations 4 50.730 (2.79) -7.699 (-3.17) Chow Test (4 – 2) F: 20.67 p: <0.0001 F: 10.82 p: 0.0011 1 2 3 4 5 NW LEV CR Rated Unrated 34.237 (2.54) 836 -5.708 (-2.71) 938 27.272 (1.73) 337 -7.539 (-2.07) 362 29.339 (0.77) 187 -7.188 (-1.82) 201 23.181 (1.23) 492 -6.153 (-2.17) 544 42.041 (2.87) 630 -7.959 (-3.61) 704 Notes to Table 10: This table presents summary results for sensitivity tests. Panel A presents results for piece-wise regressions based on the quartile of Change in EDF. The coefficient presented is on the interaction of Drop with the quartile of Change in EDF (with quartile 1 serving as the reference category). Panel B presents results for various sensitivity tests. Specifications 1, 2, and 3 run the OLS regressions for Interest Spread and Maturity, but use only observations that drop a net worth, leverage, or current ratio covenant respectively. Specifications 4 and 5 test subsamples that are rated (based on having an S&P Senior Debt Rating) and unrated respectively. Significant coefficients are in boldface. 45 Figure 1 46
© Copyright 2026 Paperzz