Do Creditor Rights Increase Employment Risk? Evidence from Debt Covenants

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. Do Creditor Rights Increase Employment Risk? Evidence from Debt Covenants Antonio Falato and Nellie Liang 2012-42 NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Do Creditor Rights Increase Employment Risk? Evidence from Debt Covenants Antonio Falato Nellie Liang1 Federal Reserve Board Federal Reserve Board April 2012 1Corresponding Author: Antonio Falato, Division of Research and Statistics, Federal Reserve Board, Washington, DC 20551. Phone: (202) 452-2861. E-mail: antonio.falato@frb.gov. Special thanks to Mark Carey and Greg Nini for their help with Dealscan and for kindly sharing their Compustat-Dealscan key. We thank Michael Roberts and Sudheer Chava for helpful comments and suggestions. Nicholas Ryan, Richard Verlander, and Brandon Nedwek provided excellent research assistance. All remaining errors are ours. The views expressed in this paper are those of the authors and do not necessarily re(cid:135)ect the views of the Federal Reserve Board of Governors or the Federal Reserve System.

Abstract This paper studies whether (cid:133)nancial contracts exacerbate or mitigate agency con(cid:135)icts among stakeholders. We consider a speci(cid:133)c contractual provision, debt covenants, and examine how, by allocating control rights between shareholders and debtholders, debt covenants a⁄ect the employment relationship. We analyze the role of covenants in both public (bonds) and private (loans) debt contracts. For public debt covenants, we estimate dynamic employment equations and (cid:133)nd a signi(cid:133)cant negative e⁄ect of leverage on employment only for (cid:133)rms with relatively high covenant protection. For private debt covenants, we use a regression discontinuity design and document sizable job cuts following a covenant violation. Overall, these (cid:133)ndings suggest that creditor rights increase employment risk. As such, they complement the recent literature on (cid:133)nancial covenants by showing that covenants a⁄ect a broader set of operating decisions than previously recognized. Moreover, the results contribute to our understanding of the consequences of the allocation of control rights within the (cid:133)rm by identifying a speci(cid:133)c risk-shifting channel from debtholders to employees.

1 Introduction One fundamental contribution of modern corporate (cid:133)nance is the insight by Jensen and Meckling (1976) that (cid:133)rms are a complex nexus of contractual relations. Important aspects of this original insight have been developed. In particular, the literature has extensively studied con(cid:135)icts of interests between shareholders and managers and between shareholders and debtholders (see Stein (2003) for a comprehensive survey). It is now well understood that debtholders can potentially mitigate con(cid:135)icts of interests with shareholders through contractual features such as (cid:133)nancial covenants. Arecentgrowingempiricalliteratureshowsthatdebtcovenantsareindeede⁄ective,and thatthetransferofcontrolrightsthataccompaniescovenantviolationshasimportantconsequences for (cid:133)rm behavior. Chava and Roberts (2008), Roberts and Su(cid:133)(2007), and Nini, Smith and Su(cid:133) (2009b) show that following loan covenant violations, (cid:133)rms reduce investment, asset growth, and debt growth, and are more likely to cut dividends and to replace their CEO. Billet, King and Mauer (2008) show that creditors use bond covenants to mitigate con(cid:135)icts with stockholders over the exercise of growth options. Anotherimportantinsightofthenexusofcontractsviewhasreceivedrelativelylittleattention.1 There can also be a fundamental con(cid:135)ict of interest between creditors and other stakeholders stemming from the fact that each of these groups represents a priority claim on (cid:133)rm revenues. As Jensen and Meckling (1976) put it, (cid:147)(cid:133)rms incur obligations daily to suppliers, to employees, to di⁄erent classes of investors, etc. So long as the (cid:133)rm is prospering, the adjudication of claims is seldom a problem. When the (cid:133)rm has di¢ culty meeting some of its obligations, however, 1"Thereisinaveryrealsenseonlyamultitudeofcomplexrelationships(i.e.,contracts)betweenthelegal(cid:133)ction (the(cid:133)rm)andtheownersoflabor, materialandcapitalinputsandtheconsumersofoutput. The(cid:133)rm... isalegal (cid:133)ction,whichservesasafocusforacomplexprocessinwhichthecon(cid:135)ictingobjectivesofindividuals(someofwhom may (cid:147)represent(cid:148)other organizations) are brought into equilibrium within a framework of contractual relations. In this sense the (cid:147)behavior(cid:148)of the (cid:133)rm is like the behavior of a market, that is, the outcome of a complex equilibrium process." (Jensen and Meckling (1976)). 1

the issue of the priority of those claims can pose serious problems.(cid:148)Since a complex web of multiplecontractsultimatelydeterminetheadjudicationof claims, theallocationof rightsbetween shareholders andcreditors likelyhas animpact oncontractual relations betweenthe (cid:133)rmandother stakeholders. However, we have virtually no empirical evidence on whether creditor rights a⁄ect other stakeholders. In this paper, we attempt to (cid:133)ll this gap in the literature by examining potential con(cid:135)icts of interests between creditors and employees. In particular, we study how a speci(cid:133)c contractual provision,debtcovenants,whichisusedtomitigatecon(cid:135)ictsbetweendebtholdersandshareholders, a⁄ects employees, an important class of stakeholders. As a (cid:133)rm(cid:146)s (cid:133)nancial condition deteriorates and (cid:133)rms have di¢ culty meeting their obligations with creditors, debt covenants that strengthen creditor rights could lead (cid:133)rms to take actions that adversely a⁄ect employees, such as layo⁄s, in order to quickly generate earnings, which could then be available to pay principal and interest on the debt. Our evidence lends strong support to this hypothesis. By strengthening creditor control rights, private (loans) and public (bond) debt covenants are associated with lower job security for workers. Our empirical setting addresses directly the issue of the endogeneity between covenants, leverage, and (cid:133)rm characteristics, such as performance and growth opportunities, an issue that had not been previously addressed in the literature on (cid:133)nancing and employment (see Ofek (1993), Sharpe (1994), and Hanka (1998)). Our study is the (cid:133)rst to document large sample evidence of a relation between (cid:133)nancial covenants and employment risk. Debtcovenantsprotectcreditorsoutsidebankruptcythroughtwochannels(cid:151)byde(cid:133)ningatransfer of control rights when a covenant is violated or by in(cid:135)uencing management actions even before a covenant is violated. The (cid:133)rst is an ex post channel. Loan covenants are tied to performance indicators, and violations lead to a transfer of control rights (Chava and Roberts (2008)). Nini et al (2009b) argue that such violations provide creditors with the same rights as a payment default, 2

includingtheabilitytoaccelerateanyoutstandingprincipal andtoterminateanyunusedrevolving credit facility. They (cid:133)nd that loans renegotiated after a covenant violation are more costly and have more restrictions on (cid:133)rms(cid:146)activities. Thus, management may take actions after violation in order to ensure continued access to credit on terms that are not too costly or restrictive. This contracting channel for covenants can increase employment risk if the transfer of control rights to creditors as a (cid:133)rm(cid:146)s (cid:133)nancial condition deteriorates leads management to cut employees in order to quickly increase earnings and cash (cid:135)ow. The second channel is an ex ante discipline mechanism. It is well established since Smith and Warner (1979) and con(cid:133)rmed by recent evidence in Nini, Smith, and Su(cid:133)(2009a) that creditors can use covenants to constrain managerial discretion. In particular, they can do so by writing into their debt contract covenants that limit the ability of managers to take actions that could have potentially adverse e⁄ects on creditors.2 By introducing explicit constraints on managers(cid:146) actions, debt covenants, like the amount of debt, may reduce operating (cid:135)exibility and, thus, may force managers to make the hard choices in order to avoid deterioration in (cid:133)nancial conditions. One such hard choice is to give up the (cid:147)quiet life.(cid:148)To the extent that employment relations are a non-pecuniary private bene(cid:133)t for managers who prefer to avoid costly con(cid:135)icts with unions and workers, covenants can lead to greater employment risk through an ex ante discipline channel by forcing otherwise reluctant management to confront employees and unions (see Bertrand and Mullainathan (2003) and Cronqvist et al. (2008)). Thus, the discipline channel also implies that (cid:133)nancial covenants may end up hurting employees(cid:146)job security. In the (cid:133)rst part of our analysis, we assess the contracting channel by exploring the link between violations of loan covenants and employment. Following Chava and Roberts (2008), we use 2In addition, debt covenants can be used to restrict stockholders(cid:146)ability to take actions that can expropriate bondholder wealth (Billet et al, 2007). 3

a panel dataset of publicly traded (cid:133)rms (4,934 (cid:133)rm-year observations) from 1994 to 2007 which is constructed by merging data on private debt issues from Loan Pricing Corporation(cid:146)s (LPC) Dealscan with the Compustat database. We also (cid:133)nd that covenant violations for loans occur frequently (Dichev and Skinner (2002)), which allows us to address identi(cid:133)cation and use a regression discontinuitydesignmethodologytoestimatethee⁄ectofloancovenantviolationsonemployment. We document sizable job cuts following loan covenant violations. Our results show that employment drops in response to a covenant violation by approximately 8% to 12% per year, a drop which is about twice as large as the median employment drop in our sample of 5.2%. This (cid:133)nding is robust to examining only the subsample of (cid:133)rms that are "close" to the covenant threshold, using a speci(cid:133)cation that includes only employment reductions, and complementing Compustat employment data with hand-collected information on 1,708 layo⁄announcements. Moreover, the result is robust to the inclusion of several control variables, including smooth functions of the distance to the covenant threshold, (cid:133)rm and year (cid:133)xed e⁄ects, proxies for (cid:133)rm growth opportunities (Tobin(cid:146)s Q), measures of capital structure (leverage) and (cid:133)nancial health (Altman(cid:146)s z-score), and proxies for earnings manipulation (abnormal accruals). Thus, consistent with the contracting hypothesis, the transfer of control rights accompanying a covenant violation leads to a signi(cid:133)cant decline in employment, as creditors(cid:146)intervention leads managers to cut employees, which would reduce expenses and raise earnings, and reduce the costs of a new renegotiated loan. In the second part of our analysis, we assess the discipline channel by considering the e⁄ect of public debt covenants on employment risk. Our dataset is constructed by merging data on public debt issues from the Fixed Investment Securities Database (FISD) with the Compustat database. FISD reports the incidence of more than 50 di⁄erent types of covenants for debt issues by non(cid:133)nancial (cid:133)rms. Using these data on individual debt issues, we construct a (cid:133)rm(cid:146)s history of covenants by tracking the (cid:133)rm(cid:146)s FISD debt issues through time and adjusting for conversions 4

and retirements at maturity which allows us to construct a (cid:133)rm-level index of covenant protection. Our measure of covenant protection, which is based on Billett, King, and Mauer (2007), groups covenants into (cid:133)fteen major categories and constructs covenant indicator variables for a (cid:133)rm(cid:146)s outstanding debt issues, which are then summed to compute an index of covenant protection. We address identi(cid:133)cation by using a Generalized Method of Moments procedure to estimate dynamic employment equations with panel data for a large sample of publicly traded (cid:133)rms (1,918 (cid:133)rms and 11,324(cid:133)rm-yearobservations)from1990to2007. Weestimatedynamicemploymentequationsthat in addition to standard determinants of employment, include (cid:133)nancial variables such as leverage and cash (cid:135)ow. The employment equations we estimate are standard in the labor literature (see Nickell (1986) for a complete survey) and the GMM approach we employ has been recently used in the literature on (cid:133)nancial constraints and investment (see, for example, Bond and Meghir (2004) and Brown, Fazzari, and Petersen (2008)). The main advantage of this approach is that it allows us to derive estimates of (cid:133)nancing variables controlling for expected future pro(cid:133)tability. We (cid:133)nd that leverage has a negative e⁄ect and cash (cid:135)ow has a positive e⁄ect only for (cid:133)rms with relatively high covenant protection. In particular, our coe¢ cient estimates imply that, for (cid:133)rms with relatively high covenant protection, a one standard deviation increase in leverage leads to a drop in employment of 5.6%, a drop which is about as large as the median employment drop in our sample of 5.2%. By contrast, for (cid:133)rms with relatively low covenant protection, the point estimates for leverage and cash (cid:135)ow are statistically insigni(cid:133)cant. These results are robust to measuring covenants with only restrictions on payout and (cid:133)nancial decisions, and controlling for (cid:133)rm growth opportunities (Tobin(cid:146)s Q) and debt maturity. Moreover, they are stronger among (cid:133)rms with relatively low cash holdings and low free cash (cid:135)ow, and for (cid:133)rms with simpler debt structures (measured as in Davydenko and Strebulaev (2004) by the Her(cid:133)ndhal index - a measure 5

of dissimilarity of face value - of public bonds outstanding). Overall, we interpret these results to be consistent with the discipline hypothesis that creditor rights increase employment risk by strengthening the disciplinary role of debt. In summary, we present empirical evidence that debt covenants, by strengthening creditor rights, lead to signi(cid:133)cant employment risk. To the best of our knowledge, this is the (cid:133)rst direct evidence consistent with the important implication of Jensen and Meckling (1976) that there are con(cid:135)icts of interest between creditors and other stakeholders with priority claims. In particular, credit contracts that mitigate one set of con(cid:135)icts, those between debtholders and shareholders, can have spillover e⁄ects on parties that are not directly subject to credit contracts, in this case employees. Our (cid:133)ndings have several implications for the literature. First, weexpandpreviousevidenceonthereale⁄ectsof(cid:133)nancialcontracting. Previousresearch has focused mostly on the e⁄ect of covenants on investment and (cid:133)nancial decisions (Chava and Roberts(2008), andNini, Smith, andSu(cid:133)(2009a)), andmorerecentlyCEOturnover(Nini, Smith, andSu(cid:133), (2009b)). Ourresultthatcovenantviolationsincreaseemploymentriskiscomplementary to these previous studies. Since creditors and employees have directly competing claims to a (cid:133)rm(cid:146)s internal cash (cid:135)ows, our analysis o⁄ers a new direct test of the contracting channel. In addition, our results suggest that the transfer of control rights matters also for key operating decisions. Second, we document evidence that debt covenants work through both an ex post contracting channel and an ex ante discipline channel. Thus, our evidence that there is a link between bond covenants and employment even without a covenant violation or debt default establishes that debt covenants, muchlikeJensen(1986)classicalhypothesisabouttheamountofdebt, actasanexante disciplining mechanism on management by reducing operating (cid:135)exibility. In addition, the (cid:133)nding that loan covenant violations, which lead to an ex-post transfer of control rights, lead managers to cut employees, is also consistent with discipline from either direct bank intervention, or an indirect 6

need to reduce expenses and raise earnings, and reduce the costs of a new renegotiated loan. Third, we document that (cid:133)nancial contracting has real e⁄ects, speci(cid:133)cally on employment. Previous studies have documented indirect real costs of bankruptcy, such as lost customers and employee relationships (Titman and Opler (1994)). Our study shows that some of these real e⁄ects are operative even before debt default or bankruptcy, which suggests that debtholders do not wait until a (cid:133)rm enters distress to exercise in(cid:135)uence. Finally, we provide evidence that covenants may be an important mechanism through which leverage can increase employment risk. Previous research suggests that employment risk is greater in more highly levered (cid:133)rms or industries: For example, Sharpe (1994) (cid:133)nds that employment in high debt (cid:133)rms more closely tracks the business cycle (see also Ofek (1993) and Hanka (1998), and Kaplan (1989), Muscarella and Vetsuypens (1990) and, more recently, Davis, Haltiwanger, Jarmin, Lerner, and Miranda (2008) for evidence from leveraged buyouts). However, these papers do not directly test a mechanism, and thus cannot show that debt acts directly as a disciplining channel because the observed correlation between debt and employment reductions could instead be caused by the strong association of debt with poor historical performance or with low growth opportunities. Our evidence is more direct since we test speci(cid:133)c channels through which loan covenant violations lead to employment cuts. 2 Loan Covenant Analysis In this section we study the consequences for employment of violations of covenants in private debt contracts (loans). Our analysis in this section follows closely Chava and Roberts (2008) and their insight that the "tightness" of loan covenants(cid:151)i.e., the distance between the covenant threshold and the actual accounting measure(cid:151)can be used to estimate the causal e⁄ect of (cid:133)nancing within 7

a regression discontinuity design setting. 2.1 Motivation Conditional on the transfer of control rights, creditors can take a number of actions that a⁄ect employment. An important aspect of the contracting channel is that employment may be a⁄ected directly by creditors intervening in operating decisions. For example, creditor interventions may take the form of "advising(cid:148)management to reduce headcount and operating expenses after a covenant violation. Thefollowingquotefromthe(cid:133)rstquarter10-Q(cid:133)lingofInterpharmHoldingsin2008exempli(cid:133)es such a situation: Subsequently, on January 28, 2008, Wells Fargo informed the Company that it wouldconsiderprovidingtheCompanywithcreditavailabilityontheconditionthatthe Company (i) develops and implements a new operating plan focused on increasing the amountofeligiblecollateralandreducingcostsand(ii)developanalternative(cid:133)nancing arrangement. Further, onFebruary5, 2008, the CompanyandWells Fargoenteredinto the Forbearance Agreement... In connection with its negotiation of the Forbearance Agreement, the Company completed a restructuring of its operations on January 25, 2008 and submitted a new operating plan to Wells Fargo, which the Company believes will result in positive cash (cid:135)ow and net pro(cid:133)ts, and includes...reducing payroll and headcount by approximately 20%. Another example of a similar quote is from the annual 10-K (cid:133)ling of Meade Instruments Corp. in 2008: We are working with our lender on a potential amendment to our agreement to 8

cure this technical default. There can be no guarantee that such amendments may be obtained as of February 29, 2008. Our restructuring plans include implementation of headcount reductions, corporate overhead and manufacturing costs. Finally, another similar quote from the second quarter 10-Q (cid:133)ling of Advanced Materials Inc. in 2004: The Company is in the process of attempting to cure its line of credit and term loan violations. Management has implemented a plan to reduce expenses and improve sales. Selling, general and administrative expenses for the (cid:133)rst quarter of (cid:133)scal 2004 and2003were$397,000and$499,000, respectively, adecreaseof $102,000or20%. This decrease was due primarily to a reduction in the number of employees as the Company continues to improve individual productivity. 2.2 Data and Sample Selection 2.2.1 Loan Data Our loan information comes from a July 2008 extract of Loan Pricing Corporation(cid:146)s (LPC) Dealscan database. The data consist of dollar-denominated private loans made by commercial banks and nonbank (e.g., investment bank, insurance companies, and pension funds) lenders to U.S. corporations during the period 1981 to 2007. The basic unit of observation in Dealscan is a loan, also referred to as a facility or a tranche. Loans are often grouped together into deals or packages. Most of the loans used in this study are senior secured claims, features common to commercial loans. We use the data to gather information on restrictive covenants. Because information on covenants is fairly limited prior to 1994, we focus our attention on the sample of loans with start dates between 1994 and 2007. Additionally, we require that each loan 9

contain a covenant restricting the current ratio, or the net worth or tangible net worth (which we group together as net worth loans) to lie above a certain threshold. We focus on these covenants for two reasons, as elaborated by Chava and Roberts (2008) and Dichev and Skinner (2002). First, they appear relatively frequently in the Dealscan database (Table I in Chava and Roberts (2008) shows that covenants restricting the current ratio or net worth are found in 9,294 loans (6,386 packages) with a combined face value of over a trillion dollars). Second, and most importantly, the accounting measures used for these two covenants are standardized and unambiguous. 2.2.2 Sample Construction OursampleconstructionstrategyfollowscloselyChavaandRoberts(2008)andDichevandSkinner (2002). Thus, in this section we summarize the main parts of our sample construction strategy, detail the few parts where it di⁄ers from these papers, and refer to Chava and Roberts (2008) for further details. We start with the annual merged CRSP-Compustat database, excluding (cid:133)nancial (cid:133)rms (SIC codes 6000-6999). While Chava and Roberts (2008) use quarterly data, we use annual data because (cid:133)rms do not report employment at the quarterly frequency. We acknowledge that this data limitation is likely to make our assessment of when the covenant violation occurs more noisy. For brevity, we refer to this subset as the Compustat sample. All variables are de(cid:133)ned in Appendix A. Data from Compustat are merged with loan information from Dealscan by matching company names and loanorigination dates fromDealscanto companynames and corresponding active dates in the CRSP historical header (cid:133)le.3 We then draw our sample containing (cid:133)rm-year observations 3Special thanks to Mark Carey and Greg Nini for their help with Dealscan and for kindly sharing their Compustat-Dealscan key. 10

in which (cid:133)rms are bound by either a current ratio or a net worth covenant during the period 1994 to 2007. Since our focus does not discriminate between these two covenants, we consider them together in our regression analysis. Since covenants generally apply to all loans in a package, we de(cid:133)ne the time period over which the (cid:133)rm is bound by the covenant as starting with the earliest loan start date in the package and ending with the latest maturity date. In e⁄ect, we assume that the (cid:133)rm is bound by the covenant for the longest possible life of all loans in the package. We also require our employment measure and the covenant(cid:146)s corresponding accounting measure to be non-missing. It is not infrequent for our extract of Dealscan to have some missing information on the covenant threshold, especially in the case of net worth covenants. We are able to partially mitigate this issue and manually recover some missing covenant information by looking at the package notes provided by Dealscan (package_comments).4 Overall, thisprocessresultsin4,986(cid:133)rm-yearobservations. Thus, ourunit of observation is a (cid:133)rm-year, each of which either is or is not in violation of a particular covenant. Ourkeyvariableof interest, employment, is fromCompustatandis the(logof) total numberof employees. In our empirical analysis, we include a number of variables that have been previously employed in the literature on loan covenants. In particular, we include (cid:133)rm size, pro(cid:133)tability, market-to-bookassetratio, leverage, debtmaturity, andAltman(cid:146)sZ-score(see, forexample, Chava and Roberts (2008)). Each variable is measured at the (cid:133)scal year-end prior to the year in which employment is measured. Since our sample selection is not random, obvious sample selection concerns might arise. Table 1 compares the characteristics of other (non(cid:133)nancial) (cid:133)rms in Compustat to those in our sample. Our sample contains relatively larger (cid:133)rms (in terms of sales) and with higher cash (cid:135)ows and 4We thank Mark Carey for suggesting to pursue this route. 11

leverage ratios relative to the Compustat population, Our sample is similar to Chava and Roberts (2008)althoughdirectcomparisonissomewhatimpededbythefactthattheyreportresults(Table II) for the net worth and current ratio samples separately. 2.2.3 Loan Covenant Violations A (cid:133)rm is in violation of a covenant if the value of its accounting variable breaches the covenant threshold, i.e., when either the current ratio or the net worth falls below the corresponding threshold. While conceptually straightforward, the measurement of the covenant threshold, and consequently the covenant violation, poses several challenges, such as the possibility of multiple overlapping deals, and, importantly, the fact that covenant thresholds can change over the life of the contract. We deal with these measurement issues following Chava and Roberts (2008) (see their Appendix B for details). In Table 1 we report Bind, the frequency of occurrence of covenant violations in our sample: 16% of the (cid:133)rm-year observations are classi(cid:133)ed as in violation. This (cid:133)gure is broadly in line with Dichev and Skinner (2002) and Chava and Roberts (2008), which is reassuring since we follow closely their data construction criteria. 2.3 Empirical Speci(cid:133)cation and Estimation Approach Our empirical speci(cid:133)cation follows the approach of Chava and Roberts (2008). In particular, we consider covenant violations as the treatment and non-violations as the control and adopt a regressiondiscontinuitydesignapproach. Wecandososincethetreatmente⁄ectisadiscontinuous function of the distance between the underlying accounting variable and the covenant threshold. 12

Speci(cid:133)cally, our treatment variable, Bind , is de(cid:133)ned as it 1 z z0 < 0 Bind = 8 it (cid:0) it it > > < 0 otherwise > > : where i and t index (cid:133)rm and year observations, z is the observed current ratio (or net worth), it and z0 is the corresponding threshold speci(cid:133)ed by the covenant. it Our baseline empirical model for this section is Emp = (cid:11)+(cid:11) Emp +(cid:11) Bind +(cid:12)x +(cid:21) +(cid:17) +(cid:23) (1) j;t 0 j;t 1 1 j;t 1 j;t 1 t j j;t (cid:0) (cid:0) (cid:0) where Emp is (log) employment and x is a vector of control variables, (cid:17) is a (cid:133)rm (cid:133)xed j;t j;t 1 j (cid:0) e⁄ect, (cid:21) is a year (cid:133)xed e⁄ect, and (cid:23) is a random error term assumed to be correlated within t j;t (cid:133)rm observations and potentially heteroskedastic (Petersen (2006)). The parameter of interest is (cid:11) , which represents the impact of a covenant violation on employment (i.e., the treatment 1 e⁄ect). Because of the inclusion of a (cid:133)rm-speci(cid:133)c e⁄ect, identi(cid:133)cation of (cid:11) comes only from those 1 (cid:133)rms that experience a covenant violation. Therefore, we restrict our attention to the subsample of (cid:133)rms that experience at least one covenant violation; however, the estimated treatment e⁄ect using the entire sample of (cid:133)rms is qualitatively similar. Note that, since our focus is on changes in employment and employment is highly persistent, we include one lag of the dependent variable (Emp ) in our speci(cid:133)cation. However, we verify that our results are robust to considering a j;t 1 (cid:0) speci(cid:133)cation without the lagged dependent variable. As noted in Chava and Roberts (2008), the nonlinear relation in equation (1) provides for identi(cid:133)cation of the treatment e⁄ect under very mild conditions. In fact, in order for the treatment e⁄ect (cid:11) to not be identi(cid:133)ed, it must be the case that the unobserved component of employment 1 13

((cid:23) )exhibitsanidenticaldiscontinuityasthatde(cid:133)nedinequation(1), relatingtheviolationstatus j;t to the underlying accounting variable. That is, even if (cid:23) is correlated with the di⁄erence, z z0, j;t it (cid:0) it our estimate of (cid:11) is unbiased as long as (cid:23) does not exhibit precisely the same discontinuity as 1 j;t Bind . it Because the discontinuity is the source of identifying information, we also estimate equation (1) on the subsample of (cid:133)rm-year observations that are close to the point of discontinuity. We follow Chava and Roberts (2008) and we formally de(cid:133)ne the (cid:147)Discontinuity Sample(cid:148)as those (cid:133)rm-year observations for which the absolute value of the relative distance between the accounting variable (current ratio or net worth) and the corresponding covenant threshold is less than 0.20. This restriction reduces our sample size by about 60% to 1,970 (cid:133)rm-year observations. For robustness, we also include smooth functions of the distance from the technical default boundary into our speci(cid:133)cation. More precisely, Default Distance (CR) and Default Distance (NW) are de(cid:133)ned as Default Distance (CR) = I(Current Ratio ) (Current Ratio - Current it it (cid:2) Ratio0); Default Distance (NW) = I(Net Worth ) (Net Worth - NetWorth0), where I(Current it it (cid:2) it it Ratio ) and I(Net Worth ) are indicator variables equal to one if the (cid:133)rm-year observation it it is bound by a current ratio or net worth covenant, respectively. The Current Ratio0 and Net it Worth0 variables correspond to the covenant thresholds. As noted in Chava and Roberts (2008), it in addition to isolating the treatment e⁄ect to the point of discontinuity, including these variables in the regression speci(cid:133)cation enables us to address the concern that the distance to the covenant threshold contains information about future investment opportunities not captured by the other determinants. 14

2.4 Results Table 2 presents our results for the entire sample consisting of loans containing either a current ratio or a net worth covenant (Panel A, "Entire Sample"), and for the "Discontinuity Sample" (Panel B). Column 0 of Panel A shows that we essentially replicate the results of Chava and Roberts (2008) on covenants and investment in our sample. Moving on to employment, Column 1 shows that covenant violations are associated with a signi(cid:133)cant decline in employment on the order of 12.5% per year. Relative to a median yearly employment drop of approximately 5% in the entire sample (and, indeed, in the entire Compustat population), this estimate translates into job cuts that are twice as large as for the median (cid:133)rm. Column 1 in Panel B shows that covenant violations lead to signi(cid:133)cant employment drops also in the Discontinuity Sample. In the Discontinuity Sample, the order of magnitude of an average yearly drop in employment following covenant violations is about 8%, which is still much larger than the median drop in the entire sample. Speci(cid:133)cations (2) through (6) incorporate additional control variables used in previous studies to address omitted variable concerns. In particular, we include (cid:133)rm size, total wages, current and lagged cash (cid:135)ows, and ROA. The inclusion of these additional controls, some of which have signi(cid:133)cant coe¢ cients (especially in the Discontinuity Sample), has little e⁄ect on the estimated impact of covenant violations. Finally, Column 7 in Panel A attempts to further isolate the discontinuitycorrespondingtothecovenantviolationbyincludingsmoothfunctionsofthedistance from the default boundary into the speci(cid:133)cation. While the coe¢ cient of net worth distance is signi(cid:133)cant, the coe¢ cient of the current ratio distance is largely insigni(cid:133)cant. Nonetheless, the estimated treatment e⁄ect of almost 10% per year remains economically and statistically large. Table 3 repeats the regression analysis in Table 2 by considering the e⁄ect of covenant vio- 15

lations on percentage employment drops, a variable studied in previous papers on (cid:133)nancing and employment (see, for example, Hanka (1998)). Columns (1) to (3) report results for the percentage employment drop using Compustat data, while columns (4) to (6) construct the same variable for Compustat data complemented with hand-collected information on 1,708 layo⁄ announcements from Wall Street Journal and other major news sources obtained from Factiva and Lexis Nexis news searches (see Ofek (1993) for a similar variable). A striking outcome of this analysis is that, as shown by columns (1) and (4), the e⁄ect of covenant violations is both qualitatively and quantitatively in line with the (cid:133)ndings reported in Table 2. In, particular, the analysis of job cuts in Table 3 reveals that covenant violations lead on average to a 7% yearly cut of a (cid:133)rm(cid:146)s workforce. 2.4.1 Robustness Table 4 veri(cid:133)es that our results are robust to controlling for additional factors that might a⁄ect employment. In particular, we include Tobin(cid:146)s Q (Column (1)), leverage (Column (2)), Altman(cid:146)s Z-score (Column (3)), and abnormal accounting accruals (Column (4)). Again, Panels A and B of Table 4 report results for the "Entire Sample" and for the "Discontinuity Sample," respectively. While Tobin(cid:146)s Q and abnormal accruals have statistically signi(cid:133)cant coe¢ cients, the magnitude of the impact of covenant violations on employment remains virtually unchanged and strongly signi(cid:133)cant. 3 Bond Covenant Analysis In this section, we study the role of bond covenants as an ex ante disciplining mechanism for (cid:133)rms. Creditors can use bond covenants that will limit the ability of managers to take actions that could have potentially adverse e⁄ects on creditors. Debt covenants, like the amount of debt, may reduce 16

operating (cid:135)exibility and, thus, may force managers to make the hard choices in order to avoid deterioration in (cid:133)rms(cid:146)(cid:133)nancial conditions. 3.1 Data and Sample Selection 3.1.1 Bond Data Similar to Billett, King, and Mauer (2007) and Chava, Kumar, and Warga (2004), our sample of public debt issues is from the Fixed Investment Securities Database (FISD), which contains detailed information on over 130,000 public debt issues across all rating categories. The version of FISD that we use includes debt issues that were issued through the last quarter of 2007 and that matured after 1989 and for which we have complete covenant information.5 We employ standard selection criteria and exclude U.S. government bonds, foreign bonds, bonds denominated in foreign currency, bonds issued by (cid:133)nancial (cid:133)rms and (cid:133)nance subsidiaries, and medium-term notes for which FISD does not record covenant information. We refer to Billett, King, and Mauer (2007) for details on the characteristics of the bond-level FISD data. 3.1.2 Sample Construction Since our objective is to examine (cid:133)rm employment policy, we create a (cid:133)rm-year panel database that matches the FISD debt issue data to issuer-level data from Compustat. To do so, we create a (cid:133)rm-year history of debt issues. Starting in 1960, we trace individual debt issues to their issuing (cid:133)rmsandthentrackthe(cid:133)rms(cid:146)portfoliosofdebtissuesovertime.6 Finally, wematchthishistorical 5As in Billett, King, and Mauer (2007), we include debt issues for which the subsequent data (cid:135)ag in the FISD is "yes." This (cid:135)ag indicates whether the issue proceeded beyond the initial input phase, containing data from a prospectus, pricing supplement, or other more detailed document or source. 6In particular, we make sure to use historical redemption information in FISD to account for the changing composition of a (cid:133)rm(cid:146)s debt issue portfolio by adjusting the outstanding principal of debt issues for sinking fund payments, calls, puts, conversions, and retirement at maturity. 17

debt issue database to Compustat data. Following Billett, King, and Mauer (2007), we start the (cid:133)rm-year sample in 1990 to allow su¢ cient time for a (cid:133)rm(cid:146)s debt issue portfolio to develop. The (cid:133)nal sample consists of 11,324 (cid:133)rm-year observations, representing 1,918 di⁄erent (cid:133)rms, over the period from 1990 to 2007.7 Variable de(cid:133)nitions are detailed in Appendix A. Bond Covenants We follow Billett, King, and Mauer (2007) and group bond covenants into 15 categories to create (cid:133)rm-year indices of covenant protection (see Chava, Kumar, and Warga (2004) and Appendix therein for details on covenants in FISD). In our primary index, for a (cid:133)rm in a given sample year, we start by creating 15 covenant indicator variables that equal one if at least one debt instrument in its FISD debt issue portfolio has the given covenant and zero otherwise. We then sum the covenant indicator variables and divide by 15 to create an index that varies from zero - no covenant protection - to one - complete covenant protection. This index gives equal weight to the various covenant categories, an assumption that we will explicitly address in our empirical analysis by also examining covenant index components separately. We also construct a weighted covenant index to address concerns about ascribing covenant protection to the overall (cid:133)rm if any debt issue has a covenant. Thus, for a (cid:133)rm in a given year, we (cid:133)rst compute 15 covenant indicator variables for each of a (cid:133)rm(cid:146)s outstanding debt issues. For each covenant, we then weight each debt issue(cid:146)s covenant indicator variable by the amount outstanding relative to the total amount outstanding, and sum the weighted covenant indicator variable across issues. We then sum the weighted covenant indicator variables and divide by 15 to compute the weighted covenant index. The 15 covenant categories can be grouped into four sub-groups as follows: 7As noted in Billett, King, and Mauer (2007), FISD debt coverage relative to Compustat debt measures is reasonably representative of the (cid:133)rm(cid:146)s outstanding debt (in our matched sample, the median ratios of the sum of FISD debt outstanding to Compustat long-term debt is 0.71. 18

1. Covenants restricting payouts to equityholders and others (two categories). An issue has a dividend restriction if there is a covenant limiting the dividend payments of the issuer or a subsidiary of the issuer. Typical subsidiary restrictions limit dividend payments to the parent, thereby preventing the parent from draining the subsidiary(cid:146)s assets. An issue has a share repurchase restriction if there is a covenant limiting the issuer(cid:146)s freedom to make payments (other than dividend payments) to shareholders and others. Note that this covenant would also restrict the issuers(cid:146)ability to redeem subordinate debt. 2. Covenants restricting (cid:133)nancing activities (seven categories). A funded debt restriction prevents the issuer and/or subsidiary from issuing additional debt with a maturity of one year or longer. The next three covenants restrict the issuer from issuing additional subordinate, senior, and secured debt, respectively. Note that the secured debtcovenantisreferredtoasanegativepledge, andtypicallyspeci(cid:133)esthattheissuercannot issue secured debt unless it secures the current issue on a pari passu basis. The category of covenants that we refer to as "total leverage tests" includes a variety of accounting-based restrictions on leverage, ranging from a requirement that the issuer maintain a speci(cid:133)ed minimum net worth to a requirement that the issuer maintain a speci(cid:133)ed minimum ratio of earningsto(cid:133)xedcharges. Asaleandleasebackcovenantrestrictstheissuerand/orsubsidiary from selling and then leasing back assets that provide security for the debtholder. This provision usually requires that the proceeds from the sale be used to retire debt or acquire substantially equivalent property. Finally, the stock issue restriction restricts the issuer and/or subsidiary from issuing additional common or preferred stock. 3. Event-driven covenants (three categories). An issue has a rating or net worth trigger if certain provisions are triggered (e.g., a put 19

option) when either the credit rating or net worth of the issuer falls below a speci(cid:133)ed level. An issue has a cross-default provision if default (or acceleration of payments in default) is triggered in the issue when default (or acceleration of payments in default) occurs for any other debt issue. Finally, we include the poison put provision as a separate category, since it is triggered in the event of a change in control. 4. Covenants restricting investment policy (three categories). An issue has an asset sale clause if the issuer and/or subsidiary are required to use the net proceeds from the sale of certain assets to redeem the issue at par or at a premium to par. Investment policy restrictions proscribe certain risky investments for the issuer and/or subsidiary. Finally, a merger restriction typically speci(cid:133)es that the surviving entity must assume the debt and abide by all of the covenants in the debt. Other (cid:133)rm characteristics We control for a number of variables that have been previously employed in the literature on bond covenants (Billett, King, and Mauer (2007) and Johnson (2003)). In particular, we include (cid:133)rm size, pro(cid:133)tability, market-to-book asset ratio, leverage, debt maturity, and Altman(cid:146)s Z-score.8 Each variable is measured at the (cid:133)scal year-end prior to the year in which employment is measured. Since these variables are not integral to our predictions, we preserve space by not including their testable predictions here and refer the interested reader to the discussion in Johnson (2003). Renegotiation and Bargaining in Default We use variables fromDavydenko andStrebulaev(2004)tomeasuredebtstructurecomplexityandasaproxyforhowdi¢ cultitistorenegotiate 8Wealsoverifythatourresultsarerobusttoincludingthefollowingadditionalvariables: investmenttaxcredit, net operating loss, and regulated (cid:133)rm dummy. 20

the (cid:133)rm(cid:146)s debt. Our empirical variables for the dispersion of debtholders(cid:146)interest are the number and the Her(cid:133)ndhal Index of public bond issues outstanding. Table5presentsdescriptivestatisticsforourmatchedFISD-Compustatsample. Fordescriptive purposes, we report the unscaled versions of the unweighted and weighted covenant indices. Panel A shows that the median (cid:133)rm-year has 5 covenant categories, with (cid:133)rm-years ranging from 0 to 13 covenant categories. The weighted covenant index has a similar distribution, with a median of 4 andarangeof0to12covenantcategories. Thesimilarityoftheunweightedandweightedcovenant indices suggests that the FISD debt issues of a (cid:133)rm do not have vastly di⁄erent covenants. For the other variables, the last two columns of Panel A report mean and median values for all other non(cid:133)nancial (cid:133)rms in Compustat with complete data over the sample period from 1990 to 2007. The (cid:133)rms in our sample are clearly larger, more highly leveraged, have less short-term debt, but have similar market-to-book ratios to other Compustat (cid:133)rms. Panel B of Table 5 presents correlations among leverage and the covenant indices and debt maturity, the market-to-book ratio, and the number of employees. Several correlations are notable. First, leverage is negatively related to the market-to-book ratio and positively related to the covenant indices. Second, the covenant indices are negatively related to the market-to-book ratio, a result that is consistent with the previous literature that examines the determinants of covenants in individual debt issues. Third, the covenant indices and short-term debt are negatively related, consistent with the view that they are substitutes in addressing stockholder-bondholder con(cid:135)icts. Finally, both the leverage and the covenant indices are negatively related to the number of employees, which is consistent with Jensen(cid:146)s (1986) discipline argument, but could also be driven by the negative relation between leverage (and covenants) and growth opportunities. This last set of correlations is particularly important, since it highlights the need to control for endogeneity issues 21

when investigating the relation between creditor rights and employment. 3.2 Empirical Speci(cid:133)cation and Estimation Approach To test the impact of creditor rights on employment we build on dynamic factor demand models as in Arellano and Bond (1991) and subsequent literature (see Bond and Van Reenen (2007) for a survey). This speci(cid:133)cation is based on the dynamic optimization "Euler condition" for (cid:133)rms that accumulate productive factors of production with a quadratic adjustment cost technology. The advantage of this approach is that it controls for expectations, thus helping to overcome a majorchallengefacingempiricalworkon(cid:133)nancingconstraints, speci(cid:133)callytheneedtoseparatethe in(cid:135)uence of variables that measure access to (cid:133)nance fromtheir possible role as proxies for expected futurepro(cid:133)tability. TheEulerequationestimationapproacheliminatestermsinthesolutiontothe optimization problem that depend on unobservable expectations and it replaces expected values of observable variables with actual values plus an error orthogonal to predetermined instruments. If (cid:133)rms do not face (cid:133)nancing constraints, Bond and Van Reenen (2007) survey the literature showing that current or lagged (cid:133)nancial variables should not enter the speci(cid:133)cation merely as proxies for expected future pro(cid:133)tability. Nickell (1984) shows that the Euler equation leads to the following empirical dynamic employment equation speci(cid:133)cation in the absence of (cid:133)nancing constraints: Emp = (cid:11) Emp +(cid:11) Emp +(cid:12)x +(cid:21) +(cid:17) +(cid:23) (2) j;t 1 j;t 1 2 j;t 2 j;t 1 t j j;t (cid:0) (cid:0) (cid:0) where Emp is the logarithm of employment for (cid:133)rm j in period t, and the vector x contains j;t j;t 1 (cid:0) the following set of explanatory variables: log-assets, log-wages, and log-industry sales.9 The 9In particular, Nickell (1984) derives a log-linear approximation to the Euler equation for a (cid:133)rm maximis- 22

speci(cid:133)cation also contains time e⁄ects, (cid:21) ; to control for, among other things, aggregate demand t shocks and movements in the aggregate cost of labor and tax rates, and (cid:133)rm-speci(cid:133)c e⁄ects, (cid:17) ; to j control for time-invariant determinants of (cid:133)rm-level employment demand.10 Thisemploymentequationisstandardintheliterature(seeNickell(1986)foracompletesurvey) and has been estimated on U.K. time series data by Layard and Nickell (1986) and on micro data by Nickell and Wadhwani (1991) and Arellano and Bond (1991). Its parameters can be interpreted as functions of the parameters of the original optimization problem. The structural model implies that (cid:11) is positive, and the coe¢ cient on lagged log-assets, and log-industry sales are positive. 1 A signi(cid:133)cant advantage of this modeling approach is that the resulting empirical speci(cid:133)cation, although generated from an explicit optimization problem, has a form that corresponds to an intuitive, dynamic employment regression. To explore the role of (cid:133)nancing constraints on employment we add variables that correspond to the (cid:133)rm(cid:146)s access to both internal and external (cid:133)nancing. In particular, we add the following variables: 1. Contemporaneousandlaggedleverage. Theuseofthismeasureofcorporatecapitalstructure is standard in the literature on (cid:133)nancing and employment (Ofek (1993), Hanka (1998), and Sharpe (1994)). Bond and Meghir (1994) and Brown, Fazzari, and Petersen (2008) include similar variables in their investment and R&D regressions. ing the present discounted value of pro(cid:133)ts as E (Emp ) = (cid:14) + (2 + r)Emp (1+r)Emp + t 1 j;t 0 j;t 1 j;t 2 a(1+r) Emp Emp ; where r is a real(cid:0)discount rate, and x = (w ;k(cid:0) ; (cid:0) (cid:27) ); i.e., the lo(cid:0)g of the real produc j t ;t (cid:0)w 1 a (cid:0) ge, the (cid:3)j l ; o t (cid:0)g 1 of capital, and a measure of industry dema j n ;t d(cid:0)s 1 hocks ( j a ;t s m j; e t asu j r ;t ed by log industry (cid:2) (cid:3) sales), respectively. The latter are from the standard log-linear labour demand equation (see, for example, Layard and Nickell (1986). Replacing the conditional expectation by its realisation and introducing an expectational error (cid:23) yields a model with the form in the text. (cid:3)j;t 10If (cid:133)rms satisfy the Euler equation period by period and use all information dated t-1 or earlier to form rational expectations, the residual term, v , will be an i.i.d forecast error. A number of factors, however, might induce a j;t (cid:133)rmspeci(cid:133)c MA(1) component in the residuals, including short-run deviations from strict rational expectations or autocorrelated optimization errors. We compare regressions with instruments that are valid with i.i.d. errors with regressions that use longer instrument lags necessary with MA(1) errors and the results are robust. 23

2. Contemporaneous and lagged gross cash (cid:135)ow, scaled by beginning of the period assets, the standard measure of internal (cid:133)nancing in the (cid:133)nancial constraint literature. Based on arguments in Bond and Meghir (1994), gross cash (cid:135)ow might matter even without (cid:133)nancial constraints, due to imperfect product market competition and/or decreasing returns to scale. However, without (cid:133)nancial constraints, imperfect competition implies that the coe¢ cient of lagged cash (cid:135)ow has a negative sign. We split the data into high versus low bond covenant (cid:133)rms. The baseline Euler equation (2) should best describe employment for low covenant (cid:133)rms and the (cid:133)nancing variables should have signi(cid:133)cant e⁄ects for (cid:133)rms with high covenants if the Jensen (1986) conjecture on creditor rights is important for employment. We estimate these equations using the (cid:133)rst-di⁄erence GMM procedure developed by Arellano and Bond (1991) for dynamic panel models with lagged dependent variables. We treat all righthandsidevariablesaspotentiallyendogenousanduselaggedlevelsdatedt-3andt-4asinstruments. Theinstrumentsmustbelaggedatleastthreeperiodsiftheerrortermfollowsa(cid:133)rm-speci(cid:133)cMA(1) process (see Bond and Van Reenen (2007)). This is the case for our data, since employment is highly persistent. A number of authors have raised concerns, however, about the weakness of lagged levels as instruments in (cid:133)rst-di⁄erence GMM regressions. Blundell and Bond (1998) show that a weak instrument problem arises if the time-series process for the regression variables is close to AR(1). Thus, to insure that weak instruments are not a signi(cid:133)cant source of bias, we follow Blundell and Bond (1998) and use two-step "system" GMM estimation. 24

3.3 Results Table 6 presents two-step GMM coe¢ cient estimates and standard errors for equation (2) in the 1990 to 2007 period. The standard errors are corrected for the well-known downward bias in small samples (e.g., Arellano and Bond (1991) and Windmeijer (2005)). Moreover, the standard errors are robust to heteroskedasticity and any arbitrary pattern of within-(cid:133)rm serial correlation (Petersen (2006)). The instruments are lags dated t-3 and t-4. The (cid:133)rst column reports results for the baseline speci(cid:133)cation estimated in the entire sample. The p-values for the m1 statistic indicate (cid:133)rst-order autocorrelation in the errors, which is expected with (cid:133)rst-di⁄erence estimation. The m2 statistics do not reject the null of no second-order autocorrelation. The Sargan test does not reject the validity of the instruments. Neither contemporaneous nor lagged leverage have a statistically signi(cid:133)cant e⁄ect on employment. 3.3.1 Comparison of Firms with Low and High Covenant Protection The second and third columns of Table 6 report results for the two-subsample split, based on whether (cid:133)rms have relatively low (below mean covenant index) or relatively high (above mean covenant index) covenant protection. The results indicate a strong negative impact of leverage on employment, but only for (cid:133)rms with relatively high covenant protection. The coe¢ cient estimate on lagged leverage implies that, for (cid:133)rms with relatively high covenant protection, a one standard deviation increase in leverage leads to a drop in employment of 5.6%, a drop which is about as large as the median employment drop in our sample of 5.2%. By contrast, for (cid:133)rms with low covenant protection, neither current nor lagged leverage have a statistically signi(cid:133)cant e⁄ect on employment. To lend further con(cid:133)dence to these results, Table 7 veri(cid:133)es that they continue to hold when 25

we split the sample based on (cid:133)ner sub-categories of the overall covenant index. In particular, the (cid:133)rst two columns of Table 7 report results for a split based on the category of covenants restricting payout policies, and the third and forth columns reports results for a split based on covenants restricting (cid:133)nancing decisions. Again, we use the mean values of each covenant class to de(cid:133)ne the respective thresholds. The coe¢ cient of lagged leverage for both splits con(cid:133)rms that the strong negative impact of leverage on employment holds only among (cid:133)rms with relatively high covenant protection. While our analysis so far has focused on identifying average employment e⁄ects, the results reported in Table 8 explore whether there is cross-sectional variation in these e⁄ects. We do this by focusing only on (cid:133)rms with relatively high covenant protection. We then further stratify this sub-sample by measures of the severity of (cid:133)nancial constraints (Panel A) and of the bargaining power of bondholders in renegotiation or bankruptcy (Panel B). In Panel A we report results for two measures of the severity of (cid:133)nancial constraints: (cid:133)rm cash asset holdings ((cid:133)rst and second columns) and free cash (cid:135)ows (third and fourth columns). We (cid:133)nd that, even within relatively high covenant protection (cid:133)rms, lagged leverage has a negative and statistically signi(cid:133)cant coe¢ cient only if cash holdings and free cash (cid:135)ows are relatively low (below mean). In Panel B we report results for two measures of the bargaining power of bondholders in renegotiation or bankruptcy, based on Davydenko and Strebulaev (2004): the number of public bond issues outstanding ((cid:133)rst and second columns) and the Her(cid:133)ndhal Index of public bond issues outstanding (third and fourth columns). Davydenko and Strebulaev (2004) argue that, due to free-rider issues, multiple creditors are less able to enforce their rights in case of bankruptcy or renegotiation triggered by covenant violation. Consistent with this intuition we (cid:133)nd that the negative e⁄ect of leverage on employment for high covenant (cid:133)rms is concentrated within the sub-sample of (cid:133)rms with relatively simpler debt structures (fewer and more similar bonds). 26

Table 9 explores the robustness of our results to a variety of alternative speci(cid:133)cations. In particular, in Panel A we verify that our results are robust to including Tobin(cid:146)s Q, a measure of (cid:133)rm growth opportunities, in our speci(cid:133)cation ((cid:133)rst and second columns); and to including a measure of debt maturity (third and fourth columns). These two additional speci(cid:133)cations address theimportantconcernthat, asemphasizedbyBillett, King, andMauer(2007)andasshownbyour discussion of the correlation table (Panel B) in Table 5, leverage, growth opportunities, and debt maturityarebestthoughtasjointlydeterminedvariables. PanelBofTable9veri(cid:133)estherobustness of our results to using book leverage instead of market leverage ((cid:133)rst and second columns), using a measure of inside (cid:133)nancing (cash (cid:135)ow) rather than leverage (third and fourth columns), and, (cid:133)nally, to using a value-weighted average covenant index rather than an equally weighted index ((cid:133)fth and sixth columns). Overall, these results provide strong support for the Jensen (1986) conjecture that creditor rights increase employment risk by strengthening the disciplinary role of debt. 4 Conclusion This paper shows that stronger creditor rights increase employment risk. We consider both the e⁄ects of ex post loan covenant violations and the ex ante disciplinary role of bond debt covenants. We document reliable evidence that both loan covenant violations and bond covenant protection have signi(cid:133)cant adverse e⁄ects on employment risk. In particular, in response to a loan covenant violation employment drops by approximately 8% to 12% per year and in response to a one standarddeviationincreaseinleverageitdropsby5.6%,butonlyfor(cid:133)rmswithrelativelyhighbond covenant protection. Ours is the (cid:133)rst direct evidence that there are con(cid:135)icts of interest between creditors and other stakeholders with priority claims, since credit contracts between debtholders 27

and shareholders have spillover e⁄ects on parties that are not directly subject to credit contracts, in this case employees. In addition, our evidence shows that the real e⁄ects of (cid:133)nancial contracting are operative even before debt default or bankruptcy, which suggests that debtholders do not wait until a (cid:133)rm enters distress to exercise in(cid:135)uence. 28

References [1] Altman, E., 1984, "A Further Empirical Investigation of the Bankruptcy Cost Question," Journal of Finance 39, 1067-1089. [2] Andrade, G. and and S. Kaplan, 1998. "How Costly is Financial (not Economic) Distress?" Evidence from Highly Leveraged Transactions that Became Distressed," Journal of Finance, 53, 1443-1494. [3] Arellano, M, and S R. Bond, 1991, "Some Speci(cid:133)cation Tests for Panel Data: Monte Carlo Evidence and an Application to Employment Equations," Review of Economic Studies 58, pp. 277-298. [4] Beneish, M. and E. Press, 1993, "Costs of Technical Violation of Accounting-Based Debt Covenants,(cid:148)The Accounting Review, 68, 233-257. [5] Bertrand, M. and S. Mullainathan, 2003, (cid:148)Enjoying the Quiet Life? Managerial Behavior Following Anti-Takeover Legislation(cid:148), Journal of Political Economy, 11, 1043-1075 [6] Billett, M. T., D. K. Tao-Hsien, and D. C. Mauer, 2007, "Growth Opportunities and the Choice of Leverage, Debt Maturity, and Covenants," forthcoming, Journal of Finance. [7] Blundell, R. and S. Bond, 1998, "Initial Conditions and Moment Restrictions in Dynamic Panel Data Models," Journal of Econometrics 87, pp.115-143. [8] Bond, S.andC.Meghir, 1994, "DynamicInvestmentModelsandtheFirm(cid:146)sFinancialPolicy," Review of Economic Studies 61, pp.197-222. 29

[9] Bond, S. R., and J. Van Reenen, 2007, "Microeconometric Models of Investment and Employment," in J. J. Heckman and E. E. Leamer eds.: Handbook of Econometrics, Volume 6A (Elsevier, Amsterdam). [10] Brown, J. R., S. M. Fazzari, and B. C. Petersen, 2008, "Financing Innovation and Growth: Cash Flow, External Equity, and the 1990s R&D Boom," forthcoming, Journal of Finance. [11] Chava,S.,P.Kumar,andA.Warga,2004,(cid:147)AgencyCostsandthePricingofBondCovenants," mimeo, University of Houston [12] Chava, S. and M.R. Roberts, 2008, "How Does Financing Impact Investment? The Role of Debt Covenants," forthcoming, Journal of Finance. [13] Cronqvist, H., F. Heyman, M. Nilsson, H. Svaleryd, and J. Vlachos, 2008, "Do Entrenched Managers Pay Their Workers More?" forthcoming, Journal of Finance. [14] Davis, S. J., J. Haltiwanger, R. S. Jarmin, J. Lerner, and J. Miranda, 2008, "Private Equity and Employment," mimeo, HBS. [15] Davydenko S. A. and I. A. Strebulaev, 2004, "Strategic Actions and Credit Spreads: An Empirical Investigation," forthcoming, Journal of Finance. [16] Dichev, I. D. and D. J. Skinner, 2002. (cid:147)Large Sample Evidence on the Debt Covenant Hypothesis,(cid:148)Journal of Accounting Research, 40, 1091 (cid:150)1123. [17] Garleanu, N. and J. Zwiebel, 2007, (cid:147)Design and Renegotiation of Debt Covenants,(cid:148)Journal of Finance, forthcoming. 30

[18] Grossman and Hart, 1982, (cid:148)Corporate Financial Structure and Managerial Incentives(cid:148), in John J. McCall, ed: The Economics of Information and Uncertainty, University of Chicago Press, Chicago, Ill. [19] Hanka, G., 1998, "Debt and the Terms of Employment,(cid:146)Journal of Financial Economics, 48, 252-282 [20] Harris, M. and A. Raviv, 1990, "Capital Structure and the Informational Role of Debt," Journal of Finance 45, 297-356. [21] Holtz-Eakin, D., Newey, W. and Rosen, H., 1988, "Estimating Vector Autoregressions with Panel Data", Econometrica, 56, pp.1371-1395. [22] Jensen, M., 1986, "Agency Costs of Free Cash Flow, Corporate Finance, and Takeovers," American Economic Review, 76 (2), 323-329. [23] Jensen, M., andW.Meckling, 1976, "TheoryoftheFirm: ManagerialBehavior, AgencyCosts and Capital Structure," Journal of Financial Economics, 3, 11-25. [24] Jensen, M. and J. Warner, 1988, "Power and Governance in Corporations," Journal of Financial Economics 20, 3-24. [25] Johnson, S. A., 2003, "Debt Maturity and the E⁄ects of Growth Opportunities and Liquidity Risk on Leverage," Review of Financial Studies 16, pp.209-236. [26] Kaplan, S., 1989. "The E⁄ect of Management Buyouts on Operations and Value," Journal of Financial Economics 24, pp.217-254. [27] Layard, R. and Nickell, S. J., 1986, "Unemployment in Britain", Economica, 53, Supplement, pp.5121-5169. 31

[28] Muscarella, C., and Vetsuypens, M., 1990, "E¢ ciency and Organizational Structure: A Study of Reverse LBOs," Journal of Finance 45, pp.1389-1413. [29] Nickell, S. J., 1984, "An Investigation of the Determinants of Manufacturing Employment in the United Kingdom", Review of Economic Studies, 51, pp.529-557. [30] Nickell, S.J., 1986, "Dynamic Models of Labor Demand," in Handbook of Labor Economics (V.1), Ashenfelter O. and R. Layard (eds.), Elsevier. [31] Nickell, S. J. and Wadhwani, S., 1991, "Employment Determination in British Industry: Investigations Using Micro-Data," Review of Economic Studies, 58, pp. 955-969. [32] Nini, G., D. C. Smith, and A. Su(cid:133), 2009a, "Creditor Control Rights and Firm Investment Policy," Journal of Financial Economics, forthcoming. [33] Nini, G., D. C. Smith, and A. Su(cid:133), 2009b, "Creditor Control Rights, Corporate Governance, and Firm Value," mimeo, University of Chicago. [34] Ofek, E., 1993, "Capital Structure and Firm Response to Poor Performance: An Empirical Analysis," Journal of Financial Economics, 34 (1), 3-30. [35] Opler, T. and S. Titman, 1994, "Financial Distress and Corporate Performance," Journal of Finance 49, 1015-1040. [36] Petersen, M., 2006, "Estimating Standard Errors in Finance Panel Data Sets: Comparing Approaches," forthcoming Review of Financial Studies. [37] Rajan, R. and A. Winton, 1995. (cid:147)Covenants and Collateral as Incentives to Monitor,(cid:148)Journal of Finance 47, 1367-1400. 32

[38] Roberts, M. and A. Su(cid:133), 2007, (cid:147)Control Rights and Capital Structure: An Empirical Investigation,(cid:148)forthcoming, Journal of Finance. [39] Sharpe, S., 1994, "Financial Market Imperfections, Firm Leverage, and the Cyclicality of Employment," American Economic Review, 84 (4), 1060-1074. [40] Smith, C., 1993. (cid:147)A Perspective on Violations of Accounting Based Debt Covenants,(cid:148)Accounting Review, 68(2), 289-303. [41] Smith, C. and J. Warner, 1979, "On Financial Contracting: An Analysis of Bond Covenants," Journal of Financial Economics 7, 117-161. [42] Stein, J., 2003, "Agency, Information and Corporate Investment," in G.M. Constantinides, M. Harris and R. Stulz, eds.: Handbook of the Economics of Finance (Elsevier, Amsterdam). [43] Stulz, R., 1990, "Managerial DiscretionandOptimal FinancingPolicies,"Journal of Financial Economics 26, 3-27. [44] Sweeney, A. P., 1994, "Debt Covenant Violations and Managers(cid:146)Accounting Responses," Journal of Accounting and Economics 17, pp.281(cid:150)308. [45] Windmeijer, F., 2005, "A Finite Sample Correction for the Variance of Linear E¢ cient Two- Step GMM Estimators," Journal of Econometrics 126, pp.25-51. 33

Appendix A: Variable De(cid:133)nitions The variables used in this paper are extracted from four major data sources: Fixed Investment SecuritiesDatabase(FISD),LoanPricingCorporation(cid:146)s(LPC)Dealscandatabase,COMPUSTAT, and CRSP. For each data item, we indicate the relevant source in square brackets. The speci(cid:133)c variables used in the analysis are de(cid:133)ned as follows: Bond Covenants [FISD] (see Billett, King, and Mauer (2007) for additional details): (cid:15) (cid:150)Covenant Index is the sum of the (cid:133)rm(cid:146)s 15 covenant indicator variables, where covenant indicator variables are equal to one if any of the (cid:133)rm(cid:146)s outstanding debt issues have a given covenant. (cid:150)Weighted Covenant Index is the weighted sum of the (cid:133)rm(cid:146)s 15 covenant indicator variables, with each covenant indicator variable is weighted by the ratio of the debt issue(cid:146)s amount outstanding to the total amount outstanding. (cid:150)Payout Covenants is the sum of the (cid:133)rm(cid:146)s two payout-speci(cid:133)c covenant indicator variables scaled by two. (cid:150)Financing Covenants the sum of the (cid:133)rm(cid:146)s seven (cid:133)nancing-speci(cid:133)c covenant indicator variables scaled by seven. Loan Covenants [Dealscan]: (cid:15) (cid:150)NW is the net worth covenant threshold (cid:150)CR is the current ratio covenant threshold Outcome measures: (cid:15) (cid:150)((Log) Employment is de(cid:133)ned as the log of the total number of employees (item 29). [Compustat] (cid:150)Decline in employment (Compustat) is de(cid:133)ned as percent decline in employment from previous year (left-censored at zero). As in Hanka (1998), this measure only includes employment reductions. [Compustat] (cid:150)Decline in employment (Compustat and layo⁄s) is de(cid:133)ned as percent decline in employment from previous year (left-censored at zero). As in Hanka (1998), this measure only includes employment reductions. This measure complements Compustat data with information on 1708 hand-collected layo⁄announcements from Wall Street Journal and other major news sources (obtained from Factiva and Lexis Nexis news searches). Controls: (cid:15) (cid:150)Size is log of the book value of assets (item 6), de(cid:135)ated by CPI in 1990. [Compustat] (cid:150)Total Wages is the log of total labor expenses (item 42), de(cid:135)ated by CPI in 1990. [Compustat] 34

(cid:150)Leverage is de(cid:133)ned as long term debt (item 9) plus debt in current liabilities (item 34) over the sum of long term debt (item 9) plus debt in current liabilities (item 34) plus market value of equity (item 25*item199). [Compustat] (cid:150)R&D is the ratio of R&D expenditures (item 46, or 0 is missing) over lagged sales (item 12). [Compustat] (cid:150)Advertising is the ratio of advertising expenditures (item45, or 0 if missing) over lagged total sales (item 12). [Compustat] (cid:150)Cash Holdings is de(cid:133)ned as the ratio of cash holdings (item 1) to total assets (item 6). [Compustat] (cid:150)Free Cash(cid:135)ow is de(cid:133)ned as the ratio to total assets (item 6) of operating income before depreciation (item 13) less interest expense (item 15) and income taxes (item 16) and capital expenditures (item 128). [Compustat] (cid:150)Tobin(cid:146)s Q is de(cid:133)ned as the market value of assets divided by the book value of assets (item 6), where the market value of assets equals the book value of assets plus the market value of common equity less the sum of the book value of common equity (item 60) and balance sheet deferred taxes (item 74). [Compustat] (cid:150)Debt Maturity is de(cid:133)ned as the fraction of a (cid:133)rm(cid:146)s total debt that matures in three years or less. [Compustat] (cid:150)Investment is capital expenditures (item 128) over net property, plant and equipment at the beginning of the (cid:133)scal year (item 8). [Compustat] (cid:150)Return on assets (ROA) is the ratio of operating income after depreciation (item 178) over lagged total assets (item 6). [Compustat] (cid:150)Current Ratio is the ratio of current assets to current liabilities. [Compustat] (cid:150)Net Worth is total assets minus total liabilities. [Compustat] (cid:150)Tangible Net Worth is de(cid:133)ned as current assets plus net physical plant, property, and equipment plus other assets minus total liabilities. [Compustat] (cid:150)Cash Flow is de(cid:133)ned as the ratio of income before extraordinary items plus depreciation and amortization over the ratio of net property, plant and equipment at the beginning of the (cid:133)scal year to total assets. [Compustat] (cid:150)Altman(cid:146)s Z-Score is the sum of 3.3 times pre-tax income, sales, 1.4 times retained earnings, and 1.2 times net working capital all divided by total assets. [Compustat] (cid:150)Accruals TWW and Accruals DD are as de(cid:133)ned in Chava and Roberts (2008). [Compustat] 35

Table 1: Bank Covenant Sample: Summary Statistics The sample consists of 4986 (cid:133)rm-year observations from the Dealscan database corresponding to loans containing a covenant that restricts its current ratio or net worth to lie above a certain threshold. De(cid:133)nitions for all variables are in Appendix A. Dealscan-Compustat Other Compustat Mean Median Mean Median Main Financial Characteristics Leverage 0.26 0.20 0.23 0.21 Cash Flow 0.06 0.09 -0.21 0.05 Net Worth 622 130 488 31 Tangible Net Worth 228 56 136 11 Current Ratio 1.91 2.42 3.45 1.74 Bind 0.16 0 Employment Characteristics Employees (000) 8.21 1.59 5.11 0.40 Decline in Employment (%, left 5.20 0 5.30 0 censored at zero) [Compustat] Decline in Employment (%, left 5.32 0 5.36 0 censored at zero) [Compustat and hand-collected layo⁄s] Labor Costs ($000 per employee) 45.92 45.35 58.34 43.00 Other Firm Characteristics Tobin(cid:146)s Q 1.94 1.36 1.89 1.41 ROA 0.04 0.07 0.03 0.07 Size (Sales $M) 1689 321 1007 60 Investment/Capital 0.06 0.04 0.07 0.04 36

Table 2: Bank Covenant Violations and Employment This table presents regression results of log employment on a covenant violation measure ("Bind") and controls. The dependent variable in Column (0) is the ratio of capital expenditures to assets at the start of the period, to ensure comparability of samples to Chava and Roberts (2008). In all remaining columns, the dependent variable is log employment. All variable de(cid:133)nitions are in the Appendix. All independent variables, except cash (cid:135)ow, are lagged one year. Panel A presents the results for the entire sample. Panel B only uses (cid:133)rm-year observations in which (cid:133)rm is close to violating the covenant, de(cid:133)ned as a narrow range ( 20%) around the covenant threshold ("Discontinuity sample"). All speci(cid:133)cations (cid:6) include both (cid:133)rm and year (cid:133)xed e⁄ects. Standard errors robust to heteroskedasticity and within-(cid:133)rm serial correlation appear below point estimates. Levels of signi(cid:133)cance are indicated by *, **, and *** for 10%, 5%, and 1% respectively. Panel A: Entire Sample Investment Log(Employment) (0) (1) (2) (3) (4) (5) (6) (7) Lag Log(Employment) 0.617*** 0.587*** 0.589*** 0.606*** 0.606*** 0.605*** 0.601*** (0.044) (0.057) (0.057) (0.057) (0.057) (0.057) (0.057) Bind -0.011*** -0.125*** -0.124*** -0.124*** -0.120*** -0.122*** -0.118*** -0.098*** (0.002) (0.026) (0.026) (0.026) (0.026) (0.026) (0.026) (0.026) Log(Assets) 0.048 0.047 0.036 0.038 0.039 0.039 (0.033) (0.033) (0.033) (0.032) (0.032) (0.032) Total Wages 0.006 0.008 0.008 0.008 0.007 (0.010) (0.010) (0.010) (0.010) (0.010) Lag Cash Flow 0.023** 0.006 -0.004 -0.003 (0.010) (0.011) (0.009) (0.009) Cash Flow 0.053 0.020 0.024 (0.033) (0.063) (0.064) ROA 0.061 0.054 (0.071) (0.071) Default Distance (NW) 0.033** (0.014) Default Distance (CR) 0.010 (0.007) Intercept 0.090*** 0.164*** -0.402 -0.394 -0.330 -0.355 -0.367 -0.382 (0.009) (0.057) (0.444) (0.448) (0.436) (0.435) (0.433) (0.435) Firm Fixed E⁄ects Yes Yes Yes Yes Yes Yes Yes Yes Year Fixed E⁄ects Yes Yes Yes Yes Yes Yes Yes Yes Observations 4,986 4,934 4,923 4,923 4,907 4,904 4,904 4,904 37

Panel B: Discontinuity Sample Log(Employment) (1) (2) (3) (4) (5) (6) Lag Log(Employment) 0.518*** 0.459*** 0.466*** 0.467*** 0.457*** 0.461*** (0.112) (0.129) (0.125) (0.126) (0.123) (0.117) Bind -0.079*** -0.077*** -0.079*** -0.078*** -0.082*** -0.085*** (0.029) (0.029) (0.029) (0.029) (0.029) (0.030) Log(Assets) 0.103* 0.100* 0.097 0.106* 0.102* (0.061) (0.060) (0.062) (0.062) (0.058) Total Wages 0.022 0.022 0.026* 0.026* (0.015) (0.015) (0.015) (0.015) Lag Cash Flow 0.018 0.064 0.079 (0.066) (0.083) (0.088) Cash Flow 0.303** 0.343** (0.137) (0.153) ROA -0.098 (0.221) Intercept 0.378*** -1.003 -0.954 -0.919 -1.077 -1.019 (0.110) (0.806) (0.785) (0.819) (0.822) (0.760) Firm Fixed E⁄ects Yes Yes Yes Yes Yes Yes Year Fixed E⁄ects Yes Yes Yes Yes Yes Yes Observations 1,952 1,949 1,949 1,945 1,944 1,944 38

Table 3: Bank Covenant Violations and Employment Risk Thistablepresentsregressionresultsofemploymentdeclinesonacovenantviolationmeasure("Bind") and controls. The dependent variable in Columns (1) -(3) is percent decline in employment, based on Compustat data only. The dependent variable in Columns (4) -(6) is percent decline in employment, based on Compustat data combined with hand-collected layo⁄ data. All variable de(cid:133)nitions are in the Appendix. All independent variables, except cash (cid:135)ow, are lagged one year. Panel A presents the results for the entire sample. Panel B only uses (cid:133)rm-year observations in which (cid:133)rm is close to violating the covenant, de(cid:133)ned as a narrow range ( 20%) around the covenant threshold ("Discontinuity sample"). (cid:6) All speci(cid:133)cations include both (cid:133)rm and year (cid:133)xed e⁄ects. Standard errors robust to heteroskedasticity and within-(cid:133)rm serial correlation appear below point estimates. Levels of signi(cid:133)cance are indicated by *, **, and *** for 10%, 5%, and 1% respectively. Panel A: Entire Sample % Employment Drop [Compustat] % Employment Drop [Compustat & Layo⁄s] (1) (2) (3) (4) (5) (6) Bind 0.069*** 0.070*** 0.070*** 0.070*** 0.072*** 0.071*** (0.010) (0.010) (0.010) (0.010) (0.010) (0.010) Log(Assets) 0.037*** 0.037*** 0.037*** 0.038*** (0.007) (0.007) (0.007) (0.007) Total Wages -0.008** -0.008** -0.009** -0.009** (0.004) (0.004) (0.004) (0.004) Lag Cash Flow -0.003 -0.003 -0.003 -0.003 (0.004) (0.004) (0.005) (0.005) Cash Flow -0.008 -0.009 -0.010 -0.011 (0.014) (0.014) (0.014) (0.014) ROA -0.016 -0.015 -0.014 -0.012 (0.019) (0.019) (0.019) (0.019) Default Distance (NW) -0.002 -0.003 (0.005) (0.005) Default Distance (CR) -0.004 -0.004 (0.004) (0.003) Intercept 0.033*** -0.507*** -0.506*** 0.033*** -0.508*** -0.508*** (0.009) (0.099) (0.100) (0.009) (0.099) (0.101) Firm Fixed E⁄ects Yes Yes Yes Yes Yes Yes Year Fixed E⁄ects Yes Yes Yes Yes Yes Yes Observations 4,664 4,582 4,582 4,664 4,582 4,582 39

Panel B: Discontinuity Sample % Employment Drop [Compustat] % Employment Drop [Compustat & Layo⁄s] (1) (2) (3) (4) Bind 0.047*** 0.046*** 0.048*** 0.046*** (0.012) (0.012) (0.012) (0.012) Log(Assets) 0.054*** 0.053*** (0.012) (0.012) Total Wages -0.016*** -0.016*** (0.006) (0.006) Lag Cash Flow -0.008* -0.008* (0.005) (0.005) Cash Flow -0.038 -0.038 (0.037) (0.036) ROA -0.002 -0.000 (0.085) (0.084) Intercept 0.044** -0.714*** 0.044** -0.706*** (0.018) (0.176) (0.018) (0.172) Firm Fixed E⁄ects Yes Yes Yes Yes Year Fixed E⁄ects Yes Yes Yes Yes Observations 1,869 1,827 1,869 1,827 40

Table 4: Bank Covenant Violations and Employment: Robustness In this table, we check for robustness of our main result in Table 7. The dependent variable is log employment. All variable de(cid:133)nitions are in the Appendix. In addition to the set of controls in Table 7, Column (1) controls for market leverage, Column (2) controls for Tobin(cid:146)s Q, Column (3) controls for Altman(cid:146)s Z-score, and Column (4) controls for discretionary accruals. All independent variables, except cash (cid:135)ow, are lagged one year. Panel A presents the results for the entire sample. Panel B only uses (cid:133)rm-year observations in which (cid:133)rm is close to violating the covenant, de(cid:133)ned as a narrow range ( 20%) (cid:6) around the covenant threshold ("Discontinuity sample"). All speci(cid:133)cations include both (cid:133)rm and year (cid:133)xed e⁄ects. Standard errors robust to heteroskedasticity and within-(cid:133)rm serial correlation appear below point estimates. Levels of signi(cid:133)cance are indicated by *, **, and *** for 10%, 5%, and 1% respectively. Panel A: Entire Sample Log(Employment) (1) (2) (3) (4) Lag Log(Employment) 0.591*** 0.586*** 0.594*** 0.578*** (0.059) (0.059) (0.061) (0.060) Bind -0.118*** -0.118*** -0.119*** -0.117*** (0.027) (0.027) (0.028) (0.026) Log(Assets) 0.074** 0.073** 0.075** 0.087** (0.034) (0.034) (0.035) (0.036) Total Wages 0.006 0.006 0.006 0.005 (0.010) (0.010) (0.011) (0.011) Lag Cash Flow -0.009 0.010 -0.006 -0.084* (0.013) (0.046) (0.034) (0.043) Cash Flow 0.017 0.029 0.014 0.036 (0.057) (0.085) (0.060) (0.050) ROA 0.070 0.091 0.077 -0.019 (0.069) (0.058) (0.072) (0.094) Tobin(cid:146)s Q 0.131*** 0.130*** 0.130*** 0.134*** (0.019) (0.022) (0.021) (0.019) Leverage 0.016 (0.031) Z-Score -0.000 (0.004) Accruals TWW 0.091 (0.066) Accruals DD 0.090** (0.045) Intercept -0.957** -0.957** -0.957** -1.111** (0.464) (0.469) (0.474) (0.485) Firm Fixed E⁄ects Yes Yes Yes Yes Year Fixed E⁄ects Yes Yes Yes Yes Observations 4,611 4,594 4,497 4,500 41

Panel B: Discontinuity Sample Log(Employment) (1) (2) (3) (4) Lag Log(Employment) 0.456*** 0.463*** 0.455*** 0.462*** (0.120) (0.121) (0.120) (0.116) Bind -0.095*** -0.076** -0.092*** -0.092*** (0.031) (0.031) (0.030) (0.029) Log(Assets) 0.116* 0.122** 0.116** 0.117* (0.060) (0.060) (0.059) (0.062) Total Wages 0.026 0.024 0.025 0.025 (0.017) (0.017) (0.017) (0.017) Lag Cash Flow 0.056 0.043 0.058 0.094 (0.087) (0.081) (0.088) (0.075) Cash Flow 0.333** 0.356** 0.321** 0.333** (0.148) (0.143) (0.133) (0.145) ROA -0.126 -0.173 -0.150 -0.302 (0.226) (0.222) (0.311) (0.359) Tobin(cid:146)s Q 0.065 0.063 0.066 0.071 (0.044) (0.043) (0.044) (0.046) Leverage -0.220* (0.124) Z-Score 0.003 (0.020) Accruals TWW 0.160 (0.203) Accruals DD -0.070 (0.073) Intercept -1.331* -1.356* -1.323* -1.326* (0.778) (0.776) (0.752) (0.789) Firm Fixed E⁄ects Yes Yes Yes Yes Year Fixed E⁄ects Yes Yes Yes Yes Observations 1,866 1,865 1,853 1,853 42

Table 5: Bond Covenant Sample: Summary Statistics The sample consists of 1918 (cid:133)rms from FISD database in the 1990 to 2007 period. De(cid:133)nitions for all variables are in Appendix A. Panel A: Summary Statistics FISD-Compustat All Other Compustat Mean Median Standard Mean Median Deviation Main Financial Characteristics Leverage 0.36 0.34 0.25 0.23 0.21 Debt Maturity 0.21 0.14 0.24 0.47 0.33 Covenant Index 5.10 5 2.08 Weighted Covenant Index 4.44 4 1.88 Employment Characteristics Employees (000) 19.5 4.8 58 3.8 0.4 Labor Costs ($000 per employee) 48 47 23.5 40 35 Firm Characteristics Tobin(cid:146)s Q 1.64 1.37 1.07 1.86 1.37 Tangible Assets 0.37 0.32 0.24 0.29 0.23 Pro(cid:133)tability 0.12 0.12 0.11 0.05 0.11 Size (Sales $M) 3790 1047 8285 867 150 Panel B: Correlations Leverage Debt Tobin(cid:146)s Q Employees Decline in Maturity Employment Leverage 1 0.01 -0.44 -0.11 0.17 Covenant Index 0.24 -0.07 -0.08 -0.03 0.003 Weighted Covenant Index 0.24 -0.05 -0.07 -0.10 0.004 43

Table 6: Dynamic Employment Regressions: High vs. Low Bond Covenant Protection Firms This table reports dynamic employment regressions estimated with two-step GMM in (cid:133)rst di⁄erences. The dependent variable is log employment. All variable de(cid:133)nitions are in the Appendix. Column (1) reports results for all (cid:133)rms, columns (2) and (3) report results when the sample is split between (cid:133)rms with low (below sample mean) and high (above sample mean) values of the covenant index variable, respectively. Lagged variables dated t-3 and t-4 are used as instruments. Controls include log of total assets, total wages, and market leverage. Year dummies are included in all regressions. Standard errors robusttoheteroskedasticityandwithin-(cid:133)rmserialcorrelationappearbelowpointestimates. Thestatistics m1 and m2 test the null of no (cid:133)rst- and second-order autocorrelation in the (cid:133)rst-di⁄erenced residuals. Sargan is a test of the null that the overidentifying restrictions are valid. Levels of signi(cid:133)cance are indicated by *, **, and *** for 10%, 5%, and 1% respectively. Entire Sample Low Covenants High Covenants (1) (2) (3) Emp 0.932*** 0.853*** 0.935*** t 1 (cid:0) (0.021) (0.052) (0.023) Emp -0.017 -0.010 -0.000 t 2 (cid:0) (0.015) (0.029) (0.018) Size 0.536*** 0.485*** 0.519*** t (0.058) (0.062) (0.069) Size -0.468*** -0.345*** -0.467*** t 1 (cid:0) (0.063) (0.077) (0.072) Total Wages -0.046** -0.103*** -0.045** t (0.018) (0.033) (0.018) Total Wages 0.010 0.030 0.014 t 1 (cid:0) (0.012) (0.026) (0.013) Leverage -0.086 -0.008 0.021 t (0.067) (0.096) (0.085) Leverage -0.071 -0.048 -0.221*** t 1 (cid:0) (0.057) (0.095) (0.074) m1 (p-value) 0.000 0.000 0.000 m2 (p-value) 0.316 0.486 0.208 Sargan (p-value) 0.556 0.236 0.3667 Observations 11,324 4,998 6,326 44

Table 7: Dynamic Employment Regressions: Analysis of Finer Covenant Classes This table reports dynamic employment regressions estimated with two-step GMM in (cid:133)rst di⁄erences. The dependent variable is log employment. All variable de(cid:133)nitions are in the Appendix. Columns (1) and (2) report results for (cid:133)rms with low (below sample mean) and high (above sample mean) number of covenants that restrict payout activities. Columns (3) and (4) report results for (cid:133)rms with low (below sample mean) and high (above sample mean) number of covenants that restrict (cid:133)nancing activities. Lagged variables dated t-3 and t-4 are used as instruments. Controls include log of total assets, total wages, and market leverage. Year dummies are included in all regressions. Standard errors robust to heteroskedasticity and within-(cid:133)rm serial correlation appear below point estimates. The statistics m1 and m2 test the null of no (cid:133)rst- and second-order autocorrelation in the (cid:133)rst-di⁄erenced residuals. Sargan is a test of the null that the overidentifying restrictions are valid. Levels of signi(cid:133)cance are indicated by *, **, and *** for 10%, 5%, and 1% respectively. Payout Covenants Financing Covenants Low High Low High (1) (2) (3) (4) Emp 0.924*** 0.947*** 0.880*** 0.969*** t 1 (cid:0) (0.031) (0.031) (0.040) (0.023) Emp -0.013 -0.005 -0.009 -0.003 t 2 (cid:0) (0.017) (0.022) (0.022) (0.020) Size 0.563*** 0.541*** 0.544*** 0.600*** t (0.067) (0.053) (0.063) (0.066) Size -0.489*** -0.479*** -0.415*** -0.580*** t 1 (cid:0) (0.077) (0.054) (0.077) (0.068) Total Wages -0.039 -0.050** -0.114*** -0.030* t (0.024) (0.024) (0.031) (0.016) Total Wages 0.005 0.025 0.046* 0.017 t 1 (cid:0) (0.015) (0.020) (0.024) (0.011) Leverage -0.110 0.017 -0.093 0.057 t (0.145) (0.077) (0.082) (0.083) Leverage 0.022 -0.169** 0.029 -0.153* t 1 (cid:0) (0.129) (0.071) (0.086) (0.082) m1 (p-value) 0.000 0.000 0.000 0.000 m2 (p-value) 0.380 0.222 0.903 0.584 Sargan (p-value) 0.248 0.596 0.658 0.290 Observations 7,887 4,030 5,762 6,155 45

Table 8: Dynamic Employment Regressions: High Bond Covenant Protection Firms with High vs. Low Cost of Debt This table reports dynamic employment regressions estimated with two-step GMM in (cid:133)rst di⁄erences for (cid:133)rms with high (above sample mean) values of the covenant index. The dependent variable is log employment. All variable de(cid:133)nitions are in the Appendix. Panel A reports results for (cid:133)rms with low (below sample mean) and high (above sample mean) values of cash holdings (Columns (1) and (2)) and free cash (cid:135)ows (Columns (3) and (4)), respectively. Panel B reports results for (cid:133)rms with low (below sample mean) and high (above sample mean) number (Columns (1) and (2)) and concentration (Columns (3)and(4))ofbondsoutstanding,respectively. Laggedvariablesdatedt-3andt-4areusedasinstruments. Controls include log of total assets, total wages, and market leverage. Year dummies are included in all regressions. Standard errors robust to heteroskedasticity and within-(cid:133)rm serial correlation appear below point estimates. The statistics m1 and m2 test the null of no (cid:133)rst- and second-order autocorrelation in the (cid:133)rst-di⁄erenced residuals. Sargan is a test of the null that the overidentifying restrictions are valid. Levels of signi(cid:133)cance are indicated by *, **, and *** for 10%, 5%, and 1% respectively. Panel A: Cash Holdings and Free Cash Flows Cash Holdings Free Cash Flows Low High Low High (1) (2) (3) (4) Emp 0.970*** 0.855*** 0.912*** 0.991*** t 1 (cid:0) (0.036) (0.330) (0.032) (0.024) Emp 0.006 -0.045 0.026 -0.018 t 2 (cid:0) (0.030) (0.353) (0.025) (0.021) Size 0.637*** 0.598 0.518*** 0.584*** t (0.067) (0.853) (0.077) (0.074) Size -0.596*** -0.398 -0.428*** -0.578*** t 1 (cid:0) (0.064) (0.860) (0.079) (0.076) Total Wages -0.060*** -0.204 -0.057** -0.024 t (0.019) (0.402) (0.027) (0.015) Total Wages 0.034* 0.145 0.015 0.017 t 1 (cid:0) (0.019) (0.440) (0.025) (0.012) Leverage 0.047 -0.036 0.161 -0.141 t (0.101) (1.199) (0.108) (0.100) Leverage -0.263** -0.022 -0.293*** -0.032 t 1 (cid:0) (0.106) (0.699) (0.090) (0.098) m1 (p-value) 0.000 0.000 0.000 0.000 m2 (p-value) 0.256 0.501 0.207 0.439 Sargan (p-value) 0.336 0.661 0.369 0.316 Observations 2,166 2,016 1,809 2,116 46

Panel B: Renegotiation Costs and Debtholder Bargaining Power in Bankruptcy Number of Bonds Her(cid:133)ndhal of Bonds Low High Low High (1) (2) (3) (4) Emp 0.944*** 0.967*** 0.983 0.936*** t 1 (cid:0) (0.034) (0.026) (1.139) (0.027) Emp -0.014 -0.005 -0.018 -0.007 t 2 (cid:0) (0.029) (0.021) (1.334) (0.021) Size 0.448*** 0.653*** 0.610 0.488*** t (0.063) (0.068) (1.463) (0.063) Size -0.371*** -0.609*** -0.571 -0.435*** t 1 (cid:0) (0.071) (0.069) (1.114) (0.067) Total Wages -0.047** -0.064*** -0.037 -0.030 t (0.021) (0.020) (0.076) (0.021) Total Wages 0.014 0.043*** 0.033 0.008 t 1 (cid:0) (0.018) (0.017) (0.022) (0.016) Leverage 0.143 -0.127* -0.096 0.130 t (0.112) (0.075) (1.095) (0.097) Leverage -0.350*** -0.093 -0.091 -0.312*** t 1 (cid:0) (0.107) (0.081) (0.358) (0.084) m1 (p-value) 0.000 0.000 0.000 0.000 m2 (p-value) 0.279 0.548 0.299 0.320 Sargan (p-value) 0.551 0.484 0.371 0.698 Observations 1,881 2,064 2,006 2,320 47

Table 9: Dynamic Employment Regressions and Bond Covenants: Robustness In this table, we check for robustness of our main result from dynamic employment regressions in Table 2. The dependent variable is log employment. All variable de(cid:133)nitions are in the Appendix. Panel A includes controls for Tobin(cid:146)s Q (Columns (1) and (2)) and debt maturity (Columns (3) and (4)). In Panel B, we use book, instead of market, leverage (Columns (1) and (2)), control for cash (cid:135)ow (Columns (3) and (4)), and use value-weighted instead of equal-weighted index of covenants (Columns (5) and (6)). Laggedvariablesdatedt-3andt-4areusedasinstruments. Othercontrolsincludelogoftotalassets,total wages, and the ratio of long-term debt to assets. Year dummies are included in all regressions. Standard errors robust to heteroskedasticity and within-(cid:133)rm serial correlation appear below point estimates. The statistics m1 and m2 test the null of no (cid:133)rst- and second-order autocorrelation in the (cid:133)rst-di⁄erenced residuals. Sargan is a test of the null that the overidentifying restrictions are valid. Levels of signi(cid:133)cance are indicated by *, **, and *** for 10%, 5%, and 1% respectively. Panel A Include Growth Opportunities Include Debt Maturity Low Covenants High Covenants Low Covenants High Covenants (1) (2) (3) (4) Emp 0.907*** 0.936*** 0.960*** 0.920*** t 1 (cid:0) (0.038) (0.023) (0.029) (0.024) Emp -0.040* 0.010 -0.053** 0.028* t 2 (cid:0) (0.022) (0.017) (0.023) (0.017) Size 0.486*** 0.593*** 0.539*** 0.654*** t (0.066) (0.059) (0.052) (0.055) Size -0.363*** -0.545*** -0.462*** -0.592*** t 1 (cid:0) (0.078) (0.060) (0.059) (0.058) Total Wages -0.088*** -0.039** -0.071*** -0.078*** t (0.033) (0.018) (0.020) (0.020) Total Wages 0.026 0.013 0.029 0.049*** t 1 (cid:0) (0.021) (0.013) (0.018) (0.016) Leverage 0.144 0.069 0.173 0.076 t (0.118) (0.098) (0.109) (0.095) Leverage -0.083 -0.209** 0.041 -0.187** t 1 (cid:0) (0.119) (0.085) (0.106) (0.088) Tobin(cid:146)s Q 0.108** 0.032 -0.007 0.008 t (0.049) (0.061) (0.046) (0.064) Tobin(cid:146)s Q -0.045 -0.019 -0.020 0.014 t 1 (cid:0) (0.040) (0.047) (0.041) (0.053) Debt Maturity 0.009 -0.076* t (0.044) (0.043) Debt Maturity -0.014 0.027 t 1 (cid:0) (0.035) (0.028) m1 (p-value) 0.000 0.000 0.000 0.000 m2 (p-value) 0.785 0.580 0.352 0.887 Sargan (p-value) 0.316 0.500 0.495 0.278 Observations 4,090 5,728 3,022 4,616 48

Panel B Book Leverage Inside Financing Weighted Average Index Low Cov High Cov Low Cov High Cov Low Cov High Cov (1) (2) (3) (4) (5) (6) Emp 0.919*** 0.963*** 0.944*** 0.941*** 0.845*** 0.952*** t 1 (cid:0) (0.057) (0.024) (0.053) (0.024) (0.063) (0.025) Emp -0.056* -0.028 -0.043 -0.004 -0.001 -0.011 t 2 (cid:0) (0.029) (0.020) (0.027) (0.020) (0.031) (0.018) Size 0.427*** 0.563*** 0.549*** 0.597*** 0.431*** 0.530*** t (0.068) (0.062) (0.071) (0.059) (0.063) (0.055) Size -0.318*** -0.518*** -0.491*** -0.534*** -0.282*** -0.493*** t 1 (cid:0) (0.090) (0.071) (0.088) (0.062) (0.088) (0.054) Total Wages -0.072** -0.025 -0.045** -0.056*** -0.110** -0.033 t (0.028) (0.021) (0.020) (0.017) (0.045) (0.022) Total Wages 0.033 0.015 0.016 0.023* 0.029 0.016 t 1 (cid:0) (0.023) (0.014) (0.019) (0.012) (0.031) (0.016) Leverage 0.162 0.096 -0.086 0.034 t (0.129) (0.069) (0.127) (0.085) Leverage -0.014 -0.156** -0.017 -0.247*** t 1 (cid:0) (0.022) (0.063) (0.126) (0.078) Cash Flow 0.010 -0.005 t (0.012) (0.013) Cash Flow 0.003 0.023** t 1 (cid:0) (0.006) (0.011) m1 (p-value) 0.000 0.000 0.000 0.000 0.000 0.000 m2 (p-value) 0.534 0.870 0.3546 0.680 0.565 0.326 Sargan (p-value) 0.231 0.260 0.432 0.301 0.292 0.365 Observations 5,030 6,549 5,584 7,318 4,111 5,280 49