Contents lists available at ScienceDirect

Journal of Financial Economics

Journal of Financial Economics 112 (2014) 287–319

http://d0304-40

☆ WeanonymFinanceCentralUniversacknowfrom thEconom

n CorrE-m

journal homepage: www.elsevier.com/locate/jfec

Credit lines as monitored liquidity insurance: Theoryand evidence$

Viral Acharya a,b,c, Heitor Almeida c,d,n, Filippo Ippolito b,e,f, Ander Perez e,f

a Department of Finance, NYU Stern, 44 West Fourth Street, New York, NY 10012-1126, USAb CEPR, UKc NBER, USAd Department of Finance, University of Illinois, 515 E. Gregory Dr., Champaign, IL 61820, USAe Department of Economics and Business, Universitat Pompeu Fabra, Ramon Trias Fargas 25-27, 08005 Barcelona, Spainf Barcelona Graduate School of Economics, Spain

a r t i c l e i n f o

Article history:Received 1 October 2012Received in revised form10 January 2013Accepted 12 February 2013Available online 11 February 2014

JEL classification:G21G31G32E22E5

Keywords:Liquidity managementCash holdingsLiquidity riskHedgingCovenantsLoan commitmentsCredit line revocation

x.doi.org/10.1016/j.jfineco.2014.02.0015X & 2014 Elsevier B.V. All rights reserved.

are grateful to Igor Cunha and Ping Liu for exous referee for providing comments and sugAssociation Meetings, 2012 Society for EconBank, Universidade Nova de Lisboa, Univerity, Norwegian School of Economics at Berledges support from the Spanish Ministry oe Beatriu de Pinos post-doctoral programmia Monetaria, Financiera y Bancaria grant isesponding author at: Department of Financeail address: halmeida@illinois.edu (H. Almeid

a b s t r a c t

We propose a theory of credit lines provided by banks to firms as a form of monitoredliquidity insurance. Bank monitoring and resulting revocations help control illiquidity-seeking behavior of firms insured by credit lines. The cost of credit lines is thus greater forfirms with high liquidity risk, which in turn are likely to use cash instead of credit lines.We test this implication for corporate liquidity management by identifying exogenousshocks to liquidity risk of firms in corporate bond and equity markets. Firms experiencingincreases in liquidity risk move out of credit lines and into cash holdings.

& 2014 Elsevier B.V. All rights reserved.

cellent research assistance, to Roland Umlauft for providing the data on mutual fund flows, and to angestions. We also thank Francois DeGeorge (discussant) and seminar participants at the 2012 Europeanomic Dynamics Meetings, 6th Winter Finance Conference on Financial Intermediation, the Europeansity of Technology Sydney, University of New South Wales, University of Kentucky, Georgia Stategen, Norway, la Caixa, Universidad Carlos III, and the Boston Federal Reserve Bank. Ander Perezf Science and Innovation through the Grant ECO2009-12157. Filippo Ippolito acknowledges supporte grant. Financial support from the Banco de Espana - Excelencia en Educacion y Investigacion enalso gratefully acknowledged by Filippo Ippolito and Ander Perez., University of Illinois, 515 E. Gregory Dr., Champaign, IL 61820, USA. Tel.: þ1 217 333 2704.a).

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319288

1. Introduction

Theory suggests that the main difference between acredit line and standard debt is that a credit line allows thefirm to access precommitted debt capacity (e.g., Shockleyand Thakor, 1997; Holmstrom and Tirole, 1998). This pre-commitment creates value for credit lines as a corporateliquidity management tool, in that they help insulate thecorporation from negative shocks that could hinder access tocapital markets. In particular, credit lines can be an effective,and likely cheaper substitute for corporate cash holdings.Nevertheless, the results in Sufi (2009) challenge the notionthat credit lines have perfect commitment. Access to creditlines is often restricted precisely when the firm needs itmost, that is, following negative profitability shocks thatcause contractual covenant violations. In addition, the surveyevidence in Lins, Servaes, and Tufano (2010) suggests thatcorporate chief financial officers not only use credit lines asprecautionary savings against negative profitability shocks,but also to help fund future growth opportunities.

We propose and test a theory of corporate liquiditymanagement that bridges the gap between existing theoryand empirical evidence on credit lines. This theory explainshow credit line revocation following negative profitabilityshocks can be optimal, and it shows when the presence offuture growth opportunities could induce firms to use creditlines in their liquidity management. The theory generatesempirical predictions, which we test using a novel dataset oncredit lines, and a new identification strategy.

In the model, a fully committed credit line (that is, fulland irrevocable liquidity insurance) creates the followingproblem. While it protects firms from value-destroyingprofitability shocks, once full insurance is in place, firmscould gain incentives to engage in risky investments thatincrease the likelihood of liquidity shocks (illiquiditytransformation). Bank-provided credit lines can help elim-inate the firm's incentive to engage in illiquidity transfor-mation, because the bank retains the right (through creditline covenants, for example) to revoke access to the creditline if it obtains a signal that the firm could have engagedin illiquidity transformation. Crucially, bank monitoringand line revocation tend to happen in the same states inwhich the firm needs the credit line the most (liquidity-shock states). This coincidence arises because credit linedrawdowns are negative net present value (NPV) loans forthe bank. Thus, the bank's incentive to monitor is strongestprecisely when the firm attempts to draw on the creditline. And, in this way, credit line revocation providesincentives both for the firm to avoid illiquidity transforma-tion and for the bank to pay monitoring costs.

In this framework, the cost of liquidity insuranceprovided by credit line arises not only from direct mon-itoring costs, but also because credit line revocations causethe firm to pass on valuable investments. In equilibrium,firms could then choose to switch to cash holdings if thecost of credit lines is too high. In particular, the modelpoints to an important determinant of the choice betweencash and credit lines: the firm's total liquidity risk. Firmswith greater liquidity risk are monitored more often,causing direct and indirect monitoring costs (i.e., expectedcosts of credit line revocation) to increase and, as a result,

are particularly likely to forego monitored liquidity insur-ance and to switch to self-insurance (cash holdings).

We extend the model to allow firms to demandliquidity to be able to absorb negative profitability shocksand to pursue additional investment opportunities. Thefinancing of future investments interacts with liquidityshock insurance through two channels. First, the cost ofcredit line revocation increases because credit line revoca-tion both limits the continuation of existing projects andstops the firm from undertaking new investments. Second,future growth opportunities could provide incentives forfirms to avoid illiquidity transformation independently ofmonitoring. The first channel is particularly relevant forfirms that tend to have investment opportunities in stateswith low cash flows (in which credit lines are likely to berevoked). The second channel is particularly relevant forfirms that tend to have investment opportunities in highcash flow states (whose probability decreases with illi-quidity transformation). This setup implies that firms withlow hedging needs (high correlation between cash flowsand investment opportunities) are less likely to use cashrelative to credit lines and are also less likely to requirecredit line covenants and revocation when using creditlines for liquidity insurance.

Overall, our model provides two sets of empiricallytestable predictions, one set dealing with the relationbetween liquidity risk and liquidity management, andanother set dealing with the relation between hedgingneeds, liquidity management, and credit line covenantsand revocations. We test these predictions using a noveldataset on credit lines from Capital IQ (CIQ). The data covera large sample of firms in the United States for the period2002–2011. CIQ compiles detailed information on capitalstructure and debt structure by going through financialfootnotes contained in firms' 10-K Securities and ExchangeCommission (SEC) filings. In particular, CIQ containsdetailed information on the drawn and undrawn portionsof lines of credit.

We test the implication that an increase in liquidity riskdecreases reliance on credit lines using two differentidentification strategies. We first exploit the downgradeof General Motors (GM) and Ford Motor Co. (Ford) in 2005as a quasi-natural experiment. The downgrade came as anexogenous and unexpected shock, especially for firms notin the auto sector. Acharya, Schaefer, and Zhang (2008)examine the GM–Ford downgrade in detail and show thatit led to a market-wide sell-off of the corporate bondsissued by these two firms. The downgrade had a signifi-cant impact on inventory risk faced by financial interme-diaries that operated as market makers for the securitiesissued by the two automakers. The resulting effect oncorporate bond prices went beyond the bonds of GM andFord and of other producers in the auto sector, creating awidespread increase in liquidity risk that affected firmsthat relied on publicly traded bonds for their financing.

Consistent with the model's predictions, we find thattreated firms that experienced an exogenous increase inliquidity risk due to the GM–Ford downgrade – specifi-cally, firms that relied on bonds for financing in the pre-downgrade period – moved out of credit lines and intocash holdings in the aftermath of the downgrade, relative

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319 289

to the set of control firms.1 More precisely, we find thatwhile treated firms had on average around 7.4% less cashholdings as a share of total liquidity relative to controlfirms before the GM–Ford downgrade, following the down-grade the difference was reduced to around 4.2% onaverage.2 Further, we find that there is both an increase incash holdings and a decrease in credit line usage for thetreated firms relative to the control firms. A placebo test in aperiod outside of the downgrade episode reveals no sucheffect.

Given that the shift toward cash for bond-dependentfirms happens only following the GM–Ford downgrade,any alternative explanation for our results must alsopredict a similar shift that is specific to the downgradeepisode. While we find this possibility implausible, wecannot entirely rule it out given that a firm's bonddependence is endogenously determined. Thus, to provideadditional evidence for a causal effect of liquidity risk onliquidity management, we study the impact of mutualfund redemptions on corporate liquidity management. Wefollow Edmans, Goldstein, and Jiang (2012) and construct ameasure of price pressure associated with outflows frommutual funds. Edmans, Goldstein and Jiang use changes instock prices that are driven by large outflows by fundinvestors to measure variation in equity values that are notdriven by firm fundamentals. In our context, mutual fundoutflows should affect only liquidity management throughits effect on equity mispricing. Firms that are under pricepressure find it more costly to access equity financing and,as a result, face greater liquidity risk in the aftermath oflarge fund redemptions. Consistent with this hypothesis,we find that as downward price pressure increases, firmsshift from credit lines to cash holdings in their liquiditymanagement. This evidence provides further support tothe empirical implications of our model.

Next, we provide novel evidence linking corporate hedgingneeds to liquidity management and credit line contracting.Following existing literature (Acharya, Almeida, and Campello,2007; Duchin, 2010), we measure hedging needs by correlat-ing firm cash flows with investment opportunities. Wemeasure investment opportunities using two alternativeindustry-level proxies: median industry annual investmentactivities and median industry Tobin's Q. In addition wecollect information from LPC Dealscan on covenants attachedto new credit lines issued to the firms in our sample and fromNini, Smith, and Sufi (2009) on covenant violations.

In line with the predictions of our model, we find thatlow hedging needs firms (those with the highest correla-tions between cash flows and investment opportunities) arethe most likely to use credit lines. Depending on themeasure chosen, moving from the bottom quintile to thetop quintile of hedging needs is associated on average witha decrease of between 19.5% and 27.8% in the probability ofhaving a line of credit. The credit lines of low hedging needs

1 We measure bond dependence at the three-digit Standard IndustryClassification (SIC) industry level, to prevent firm-level unobservedcharacteristics that vary within an industry from driving our results.

2 Because ours is a difference-in-differences empirical design, time-invariant variables (even unobservable ones) are differenced out and donot compromise identification.

firms are also less likely to contain covenants and to berevoked by banks. These findings support the model'simplications that credit lines are less costly for low hedgingneeds firms and that, for such firms, the credit line provider(the bank) is less likely to retain revocation rights throughcovenants and to exercise them. In particular, these findingsstem uniquely from the feature of our model that firmstaking out lines of credit are monitored by banks thatprovide the lines by inclusion of contract terms such ascovenants and by invoking them ex post; the high expectedcost of monitoring and the ex post revocation costs forhigh-hedging needs firms (which face a greater illiquiditytransformation problem) induce these firms in equilibriumto rely more on cash instead of on lines of credit.

While we see the new identification strategy for theliquidity risk tests and the hedging needs tests (whichtogether provide support for lines of credit being a form ofmonitored finance) as the main empirical contributions ofthe paper, we show that our results are also consistent withexisting empirical evidence. Our main goal is to show thatthe Capital IQ data deliver results that closely resemble thoseobtained with other datasets. For example, Capital IQ dataconfirm earlier findings that profitable, safer, low Tobin's Qand high tangibility firms are more likely to have credit linesand less likely to use cash for liquidity management. Creditline users tend to have higher bond ratings and are morelikely to have a rating when compared with firms that usecash for liquidity management. We also confirm the Sufi(2009) finding that credit lines tend to get revoked followingdecreases in profitability. In addition, we find evidence thatcredit line drawdowns are relatively infrequent relative tocash reductions in situations in which firms are likely to havea liquidity need, which we define as a year in which profit-ability is negative. For instance, the likelihood of a credit linedrawdown to fund a liquidity shock (among credit line users)is close to ten times lower than the likelihood of a reductionin cash holdings to fund a similar liquidity shock among cashusers. This result shows that the ex post usage of credit linesand cash to meet liquidity needs is consistent with ex antedifferences in liquidity risk exposure across firms.

Our evidence is related to the Sufi (2009) result thatfirms with low profitability and high cash flow risk are lesslikely to use credit lines and more likely to use cash forliquidity management because they face a greater risk ofcovenant violation and credit line revocation.3 Our theoryand tests provide the new insight that credit line revoca-tion following negative profitability shocks can be anoptimal way to incentivize firms to not strategicallyincrease liquidity risk of projects and banks to monitorfirms to contain the illiquidity transformation problem.4

Our findings on the role of hedging needs for liquiditymanagement are related to those in Disatnik, Duchin, andSchmidt (2010). They find that firms that are moreexposed to traded sources of risk such as foreign currency

3 The growing empirical literature on the role of credit lines incorporate finance also includes Yun (2009), Campello, Giambona,Graham, and Harvey (2010), and Acharya, Almeida, and Campello (2013).

4 This insight distinguishes our paper from previous theoretical workon corporate liquidity management such as Acharya, Almeida, andCampello (2007).

5 To see this, let the Date 0 investment produce a Date 1 cash flowequal to ~r , which is random. The cash flow ~r can be equal to either r, withprobability ð1�bλÞ, or zero, with probability bλ . The required Date 1investment is equal to I1. If we let r¼ I1, we obtain the set up abovewith ρ¼ I1.

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319290

and commodity price risk (due to their industry affiliation)use more derivatives-based hedging. These firms are alsomore likely to use credit lines instead of cash to managethe liquidity risk that remains after derivatives usage.While we do not explicitly model the firm's demand forderivatives, this result is consistent with our theory.Industry-related derivatives usage reduces the probabilitythat the firm will face a cash shortfall and, thus, the firm'sliquidity risk. According to our model, this firm faces lowermonitoring costs and is more likely to demand monitoredliquidity insurance through credit lines. In this sense, ourmodel can provide an explanation for the Duchin, Ozbas,and Sensoy's (2010) results.

The theory in our paper is closely related to theliterature that focuses on bank specialness, such as Fama(1985), Houston and James (1996), and Holmstrom andTirole (1997). In these models, the bank improves resourceallocation by creating pledgeable income through a reduc-tion in private benefits, improved project screening, andother mechanisms. In contrast, in our paper, the specialrole of the bank is to provide monitored credit lines thatallow the bank to control firms' liquidity choices. Kashyap,Rajan, and Stein (2002) and Gatev and Strahan (2006) dofocus on liquidity provision by banks through credit lines.These papers suggest a different rationale for credit lines(diversification synergies with deposits). Our model can beconsidered as nesting the view of banks as efficientproviders of liquidity insurance under the view of banksas efficient monitors.

A close parallel also exists between our paper's mainhypotheses and standard models of insurance, startingwith Rothschild and Stiglitz (1976). In Rothschild andStiglitz, for example, high risk agents are willing to paymore for insurance and self-select into contracts withmore coverage. In our model, cash-based liquidity man-agement can be interpreted as liquidity insurance withgreater coverage, because there is no ex post revocation ofinsurance under cash management. In equilibrium, cashmanagement is chosen by firms that find it optimal toinvest in the riskier, illiquid project, while credit lines(which provide partial insurance only) are chosen by firmsthat opt to invest in the less risky liquid project. Theeconomics literature on insurance has also discussed anotion that is close to illiquidity transformation. Arnottand Stiglitz (1988), for example, show that an agent who isoffered a contract with greater insurance coverage endo-genously chooses to take more risk to take advantage ofthe increased insurance. In our model, this moral hazardeffect of insurance creates a role for imperfect insurancethrough revocable credit lines.

We start in the next section by introducing our model ofcredit line revocation as an incentive device for the firm tocommit to a liquid investment choice. Section 3 summarizesthe empirical implications of the model. Section 4 intro-duces and presents our empirical analysis, and Section 5concludes the paper.

2. The model

Our model introduces two innovations to the standardliquidity management model of Holmstrom and Tirole (1998).

First, we allow the firm to engage in illiquidity-seekingbehavior after acquiring insurance against liquidity shocks,to explore the role of credit line revocation as a monitoringmechanism. Second, we introduce a future investment oppor-tunity whose financing must be planned for. The correlationbetween the probability of arrival of this investment oppor-tunity and short-term cash flow varies across firms. Thisinnovation allows us to characterize the impact of hedgingneeds on the firm's liquidity policy.

2.1. Basic structure

The timing of the model is depicted in Fig. 1. At theinitial date (Date 0), each firm has access to an investmentproject that requires fixed investment I at Date 0 and anadditional investment at Date 1, of uncertain size (thefirm's liquidity need). The Date 1 liquidity need can beeither equal to ρ, with probability bλ, or zero, with prob-ability ð1�bλÞ. We can interpret state bλ (1�bλ) as a state inwhich the firm produces a low (high) cash flow at Date 1.5

In state bλ, a firm continues its Date 0 investment onlyuntil Date 2 if it can meet its Date 1 liquidity need.Otherwise, the firm is liquidated and the project producesnothing. If the firm continues, it produces total expectedDate 2 cash flow equal to cρ1 from the original project. As inHolmstrom and Tirole (1998), the basic friction in this modelis that some of this expected cash flow is not pledgeable tooutside investors. In short, we assume that conditional oncontinuation, the original project produces pledgeableincome equal to cρ0 ocρ1 . In the Appendix we describe amoral hazard structure (identical to that in Holmstrom andTirole) that generates limited pledgeability.

If the firm continues, it has access at Date 1 to anadditional investment opportunity that arrives with prob-ability υ. This Date 1 investment requires an investment ofτ and produces a Date 2 cash flow of ρτ4τ. For simplicity,we assume that this Date 2 cash flow generates zeropledgeable income (this assumption can be easily relaxed).

The probability of arrival of the new investment oppor-tunity depends on the Date 1 state. It is equal to υ¼ υH instate ð1�bλÞ and υ¼ υL in state bλ. The key differencebetween the liquidity need ρ and the investment oppor-tunity τ is that τ can arrive in both states of the world,whereas the liquidity need ρ arises only when firm cashflows are low (state bλ). The probability bλ is endogenous.Specifically, bλ is either equal to λ or equal to λ04λ. Themanager chooses the probability bλ after the initial invest-ment has been made. The choice of probability is unob-servable at Date 1 to outside parties, who can observe onlywhether or not the firm has a liquidity need at Date 1 (e.g.,ρ is observable). There is no discounting.

The manager's choice of bλ impacts the original project'scash flows and pledgeable income in the following way.If the manager chooses bλ ¼ λ0, the Date 2 cash flow cρ1 isequal to ρ01. If the manager chooses bλ ¼ λoλ0, the Date 2

Fig. 1. Timeline of model.

6 In the moral hazard framework of the Appendix, this condition is aconsequence of the assumption that illiquidity transformation increasesboth the project's verifiable cash flow and private benefit in a way thatleaves pledgeable income constant. Refer to Appendix A.1.

7 The firm's ability to use a credit line to manage liquidity distin-guishes our theory from Acharya, Almeida, and Campello (2007), whoconsider only cash as a liquidity management device.

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319 291

cash flow is ρ1oρ01. This structure allows us to interpretthe choice of bλ ¼ λ0 as the illiquidity transformation by themanager, because it results in a high Date 2 cash flowconditional on continuation, as well as on a greaterprobability of a liquidity shock at Date 1. We refer to thechoice of bλ ¼ λ as the liquid project, and the choice of bλ ¼ λ0

as the illiquid project.To economize on notation denote the following quantities:

ð1�λÞυHþλυL � υ;

ð1�λ0ÞυHþλ0υL � υ 0: ð1ÞThus, υ (υ 0) is the expected arrival rate of the new investmentopportunity when the manager chooses λ (λ0). Because λ oλ0,υ4υ 0.

We make the following assumptions:

ρ1� I�λρþυðρτ�τÞ4ρ01� I�λ0ρþυ 0ðρτ�τÞ40; ð2Þ

ρ0 ¼ ρ00; ð3Þ

max½Iþλρþυτ; Iþλ0ρþυ 0τ�rρ0; ð4Þ

ρ0oρoρ1; ð5Þ

ð1�υLÞτoρ: ð6ÞAssumption (2) means that the liquid project has

higher NPV than the illiquid project, net of monitoring

costs. But the illiquid project is also positive NPV. Assump-tion (3) means that both the liquid and the illiquid projectproduce the same pledgeable income ρ0.6 Assumption (4)means that both the liquid and the illiquid project gen-erate enough pledgeable income to fund the initial invest-ment and the Date 1 investment opportunity. Assumption(5) means that, conditional on the Date 1 liquidity shock,the firm does not have sufficient pledgeable income tocontinue the original project (ρ0oρ) though continuationis efficient irrespective of the new investment opportunity(ρoρ1). Assumption (6) captures the intuitive conditionthat the firm's demand for liquidity should be greater inthe low cash flow state, as we will see below.

2.2. Liquidity management and illiquidity transformationincentives

The firm canmanage its liquidity either using a bank creditline or cash holdings.7 As in Holmstrom and Tirole (1998), the

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319292

firm holds cash by buying a riskless, liquid security (such as aTreasury bond) at Date 0. The price of the bond is equal to q,which we take as exogenous. Holmstrom and Tirole endo-genize the cost of cash and show that if the demand for liquidsecurities is high enough the firm might need to pay aliquidity premium to transfer cash across time (that is, q41in equilibrium). In what follows we assume that q¼1, suchthat there is no liquidity premium. However, the modelimplications also hold when q41. In Section 3, we discussthe implications of the liquidity premium for the modelpredictions.

The credit line works similarly to an insurance contract.The firm pays a commitment fee y to the bank in the statesin which it does not need additional liquidity, in exchangefor the right to draw on additional funds (up to amaximum equal to w) in other states.8 We assume thatthe bank can provide the credit line at zero deadweightcost, that is, the bank can offer contracts that correspondto actuarially fair insurance.9

Assume that to implement the liquid project, the firmraises capital to finance the initial investment and opens acredit line with the bank. In exchange for the financing,the firm commits to making a payment D to the bank outof the Date 2 cash flow.10 The bank's break-even constraintrequires that

ð1�λÞ½ð1�υHÞDþυHðD�τÞ�þλ½ð1�υLÞDþυLðD�τÞ�ρ� ¼ I;

ð7Þwhile the pledgeability constraint requires that Drρ0. Thefirm makes payments to the bank in the states in which itdoes not need additional liquidity [such as stateð1�λÞð1�υHÞ], in exchange for additional liquidity trans-fers in the other states (such as state λ). As long as Eq. (4)holds, such that Iþλρþυτrρ0, it is possible to find apayment Drρ0 such that the break-even constraint issatisfied.

Once this contract is in place, does the firm haveincentives to stick to the liquid investment? Doing soproduces a payoff for the firm equal to

ð1�λÞðρ1þυHρτ�DÞþλðρ1þυLρτ�DÞ ¼ ρ1�Dþυρτ: ð8ÞThat is, the firm uses the credit line to continue the projectand fund the new investment, and it repays D to the bankin both states. If the firm deviates and shifts funds into theilliquid project, the payoff is

ð1�λ0Þðρ01þυHρτ�DÞþλ0ðρ01þυLρτ�DÞ ¼ ρ01�Dþυ 0ρτ : ð9Þ

8 In this model, the firm does not need to repay the credit linedrawdown w, that is, the drawdown of the credit line generates noliability. More generally, we can interpret w as the difference between thecredit line drawdown and the present value of the repayments from thefirm to the bank. As discussed by Tirole (2006), the key insurance featureof the credit line is that it forces the bank to make loans that are (ex post)negative NPV for the bank. The bank breaks even through commitmentfees y.

9 Acharya, Almeida, and Campello (2013) show that banks should beable to offer fair insurance to firms that have idiosyncratic liquidity riskbut might need to increase the cost of credit line provision if liquidity riskis correlated across firms. In Section 3, we discuss the impact of suchcosts for the model implications.

10 This payment covers both the initial investment and the commit-ment fee on the credit line.

Deviation thus pays off for the firm if ρ01þυ 0ρτ4ρ1þυρτ .Under this deviation, the firm faces a liquidity shock withgreater probability (λ04λ). This increase in the cost of theproject is irrelevant for the firm once it has secured a fullycommitted credit line with the bank. The bank mustfinance the liquidity shock ρ with greater probability butis not compensated for it (the firm still pays D).11 Thus, thefirm could have incentives to deviate the project funds intothe illiquid investment.12

2.3. Bank monitoring and the role of credit line revocation

To avoid illiquidity transformation, the firm might needa commitment device. This mechanism creates a role formonitored liquidity insurance. We assume that the moni-tor (the bank) can pay a cost c at Date 1 to receive a signal sthat gives information about the manager's liquiditychoice. Specifically, we have that

Probðs¼ s0=bλ ¼ λÞ ¼ μo1;

Probðs¼ s0=bλ ¼ λ0Þ ¼ 1: ð10ÞThat is, if the firm chooses bλ ¼ λ0, bank monitoring revealsthat the firm made the wrong choice. But the bankreceives an imperfect signal in case the firm makes thecorrect choice. This signal s is verifiable, so contracts thatare contingent on s are feasible.

We do not assume that the bank can commit tomonitor. Instead, the bank monitors only if it has sufficientincentives to do so. Because credit line drawdowns arelarger in the low cash flow state (state bλ), the bank'sincentive to monitor is greater in this state. Suppose thefirm has committed to a payment DMrρ0 to the bank.13

If the manager chooses bλ ¼ λ, the bank has incentives tomonitor in the low cash flow state when

μðρþυLτ�DMÞ4c: ð11ÞThe expression ρþυLτ�DM is the difference between theexpected credit line drawdown ρþυLτ and the paymentthat the firm makes to the bank, DM. This expression isgreater than zero by assumption (5) and because DMrρ0.Thus, revoking the credit line (with probability μ) gener-ates a benefit for the bank. As long as the expected benefitfrom revocation is greater than the monitoring cost, thebank has incentive to monitor. This incentive to monitorarises because the credit line allows the firm to access

11 Illiquidity transformation is similar to, but different from theJensen and Meckling's (1976) risk-shifting problem. In Jensen and Meck-ling, a firm with high leverage could have incentives to take riskyinvestments that shift value from debt to equity. The main implicationis that firms prone to risk-shifting should have low leverage ratios.In contrast, the main implication of our model is that firms could requiremonitored liquidity insurance in the form of a revocable credit line tocontrol incentives to engage in illiquidity transformation. In particular,this implication holds irrespective of the firm's leverage ratio.

12 One could wonder whether the firm can use derivatives tomitigate the incentives to engage in illiquidity transformation. A calloption on firm cash flows, for example, could transfer the benefits ofsuccess to outside investors. Such a strategy would not work in our setup,because any cash flow in excess of the pledgeable income ρ0 cannot bepaid out to investors because of the underlying moral hazard problem.

13 DM is the promised payment to the bank in the monitoredsolution. We solve for DM below.

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319 293

more funding than what its pledgeable income allows. Theinsurance role of the credit line and bank incentives tomonitor are inherently linked.

In contrast, the bank's incentives to monitor in the highcash flow state are weaker because the firm's demand forliquidity is lower. In that state, the credit line could berequired to help fund the new investment opportunity, ifτ4ρ0. Thus, the potential benefit of monitoring in the highcash flow state is given by μðτ�DMÞ�c. Because τoρþυLτ[by assumption (6)], the incentives to monitor are greaterin the low cash flow state.

Given that the bank is expected to monitor in the lowcash flow state, does the firm have incentives to avoidilliquidity transformation? The firm now anticipates that ifits cash flow is low, the bank monitors and could revokethe credit line. Because illiquidity transformation increasesthe probability of the low cash flow state, the firm has astronger incentive to avoid it when it expects the bank tomonitor. As we show in the proof to Proposition 1, ifcondition (27) holds, the optimal contract can rely oncredit line revocation to provide incentives for the firm toavoid illiquidity transformation.

In particular, this framework helps explain why creditline revocation tends to happen precisely in states of theworld when the firm needs the credit line the most.Monitoring happens in low cash flow states because thebank gains the most from credit line revocation in thesestates and because this revocation provides incentives forthe firm to avoid actions that increase the probability oflow cash flow states, such as illiquidity transformation. Themonitored credit line entails both the direct monitoringcost c and the indirect costs of credit line revocation,which arise from the bank's ability to revoke access to thecredit line with probability μ. The resulting trade-offgenerates the main predictions of the theory.

2.4. Main results

We start by deriving the firm's payoffs under themonitored credit line and when it uses cash to manageliquidity.

2.4.1. Project payoffs and optimal project choiceWe first derive the firm's payoff if it chooses the liquid

project.

Proposition 1. If ρ01þυ 0ρτrρ1þυρτ , then monitoring is notrequired and the payoff of the liquid project is

UL � ρ1� I�λρþυðρτ�τÞ: ð12ÞIf ρ01þυ 0ρτ4ρ1þυρτ , the liquid project can be implementedonly with monitoring. Let DM be defined as

ð1�λÞ½DM�υHτ�þλð1�μÞðDM�υLτ�ρÞ ¼ I: ð13ÞIf Eqs. (11) and (27) hold, the payoff of the liquid project withmonitoring is given by

Un

L ¼UL�λ½cþμ½ρ1�ρþυLðρτ�τÞ��: ð14ÞIf either Eq. (11) or Eq. (27) does not hold, the liquid projectcannot be implemented.

All results are proved in the Appendix. The conditionρ01þυ 0ρτrρ1þυρτ establishes whether the firm can accessfully committed liquidity insurance or not. If the incentiveto engage in illiquidity transformation is high (ρ01þυ 0ρτrρ1þυρτ), then implementing the liquid project requiresmonitoring. As long as monitoring is incentive compatibleboth for the firm and the bank, the liquid project can beimplemented resulting in the payoff Un

L . The termλcþμ½ρ1�ρþυLðρτ�τÞ� denotes the (ex ante) cost of mon-itoring. It comprises the direct monitoring cost and thecost of credit line revocation, which is the loss of theoriginal project and the new investment opportunity withprobability μ. Finally, if monitoring is not incentive com-patible for both the firm and the bank, then the liquidproject cannot be implemented when there are incentivesto engage in illiquidity transformation.

If the firm chooses the illiquid project to begin with,then illiquidity transformation is not an issue. Givenassumptions (2) and (4), the illiquid project can alwaysbe implemented:

Proposition 2. The payoff of the illiquid project is given by

UC ¼ ρ01�λ0ρ� Iþυ 0ðρτ�τÞ: ð15ÞIn this case the firm promises a payment D0 to outsideinvestors, which is given by

ð1�λ0ÞðD0 �υHτÞþλ0ðD0 �υLτ�ρÞ ¼ I: ð16Þ

Corollary 1 follows from Propositions 1 and 2.

Corollary 1. If ρ01þυ 0ρτrρ1þυρτ , the firm chooses the liquidproject. If ρ01þυ 0ρτ4ρ1þυρτ , then the firm chooses the liquidproject if Eqs. (11) and (27) hold and Un

L 4UC . It chooses theilliquid project if Un

L rUC or if either Eq. (11) or Eq. (27) doesnot hold.

If ρ01þυ 0ρτ4ρ1þυρτ then the liquid project requiresmonitoring. Thus, the firm chooses the liquid project ifmonitoring is feasible and the monitoring cost is not toohigh, and it chooses the illiquid project otherwise.

2.4.2. Implementation using cash and credit linesWe now characterize the implementation of the liquid

and illiquid projects using cash and credit lines. We focusfirst on the case in which monitoring is required for theimplementation of the liquid project and is incentive-compatible for the firm and for the bank. In the extensionssubsection (Section 2.4.4) we discuss the implications thatarise from studying other cases.

Proposition 3. If Un

L 4UC , then the firm implements the liquidproject with a monitored credit line of size ρþτ�ρ0. Thecredit line is revoked with probability μ if the firm produceslow date-1 cash flow (in state λ). If Un

L rUC , then the firmimplements the illiquid project by holding an amount of cashequal to C ¼ ρþτ�ρ0.

If illiquidity transformation is an issue, the firm mustemploy monitored liquidity insurance to implement theliquid project. The natural solution is then to use arevocable credit line, which gives the bank the right to

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319294

deny access to the credit line if it receives a signal that thefirm has engaged in illiquidity transformation.14

In principle, monitored liquidity insurance can also beimplemented with cash that is held by the firm, providedthat the bank or another outside investor can monitor thefirm and control whether the firm can use the cash tofinance expenditures. However, given that the firm has thedefault control of cash in this case, such a contract mustspecify exactly and, in an enforceable way, when the firmhas priority over the usage of cash and when the cashmust be returned to the bank. In general, this solution isnot as robust as the bank-provided credit line.15

To illustrate the issues that might arise, suppose, forexample, that the firm learns about the liquidity shock ρbefore the bank does.16 Under the credit line implementa-tion, the firm must contact the bank to request a draw-down of the credit line because the firm does not havesufficient resources to pay for the liquidity shock. Thiscontact would prompt the bank to monitor, and possiblyrevoke, the credit line (as modeled above). Under analternative cash implementation, because the firm hascontrol over the cash, the contract would need to ensurethat the firm has correct incentives to report the liquidityshock before using the cash to pay for it. If the firmdeviates from the contract and spends the cash, the bankcan, for example, demand immediate payment and liqui-date the firm. However, in this case liquidation becomes(ex post) inefficient for the bank because continuationproduces pledgeable income ρ0, while liquidation pro-duces nothing. The bank has ex post incentives to renegefrom monitoring and liquidation and, thus, the monitoredsolution breaks down.

While cash is not the best option to implement theliquid project when monitored liquidity insurance isrequired, it does allow the firm to implement the illiquidproject. In particular, cash is a better alternative for thefirm in this case than a non-monitored credit line. Theproblem with the credit line alternative is that monitoringis conditionally efficient for the bank in state λ0. In fact, thebank's incentives to monitor are very strong if the firmchooses the illiquid project, because the credit line isrevoked with probability one given monitoring.17 Unlessthe bank can perfectly commit not to monitor, the firmrisks being liquidated with probability one in state λ0 undercredit line implementation.

This result suggests that cash-based liquidity manage-ment tends to be associated with illiquid projects that

14 This result also implies that the firm would not fully substitute themonitored credit line for derivatives, even if the underlying source ofcash flow risk is traded on the market. Unlike credit lines, derivatives donot allow for monitoring and revocation. Full insurance through deriva-tives would thus induce firms to make illiquid investments ex ante, in theexpectation that investors would pay for cash shortfalls ex post.

15 Our model explains why the credit line could have an unique rolein implementing monitored liquidity insurance, but it does not explainwhy credit lines are provided by banks. See Kashyap, Rajan, and Stein(2002), Gatev and Strahan (2006), and Gatev, Schuermann, and Strahan(2009) for explanations that focus on synergies between the deposit-taking and the credit line providing roles of banks.

16 For example, the firm could have inside information about Date 1cash flows.

17 The bank would benefit from monitoring if ρþυLτ�D04c.

require frequent liquidity infusions. Firms that endogen-ously choose to invest in projects with high liquidity riskfind it optimal to self-insure against such shocks, whilefirms that choose to invest in projects with low liquidityrisk manage liquidity through a monitored credit line toensure that they do not engage in illiquidity transforma-tion after the bank has provided liquidity insurance. Weshow that the link between liquidity risk and liquiditymanagement extends to a case in which firms are hetero-geneous with respect to liquidity risk.

2.4.3. Introducing heterogeneity in liquidity risk and hedgingneeds

We now introduce two sources of firm heterogeneity inthe model, with the goal of deriving testable empiricalimplications.

Liquidity risk and the choice between cash and creditlines: Suppose first that there are two types of firms, L andH. Firm L has lower liquidity risk than firm H irrespectiveof project choice, that is, bλLobλH (which is equivalent tosaying that λLoλH and λ0Loλ0H). This difference in liquidityrisk can be interpreted as arising from firm characteristicssuch as the risk of the underlying business and thecorrelation between cash flows and investment needs.Specifically we make the following assumption:

λ0j ¼ λjþt for j¼ ðL;HÞ: ð17Þ

This assumption means that the effect of illiquidity trans-formation on the probability of the liquidity shock is thesame for both types of firms. This assumption is sufficientbut not necessary for our results – all that is needed is thatthe potential increase in illiquidity risk is not much largerfor firms of type H. Given this assumption, Proposition 4follows from the analysis in the previous section:

Proposition 4. Firms with low liquidity risk (type L) are morelikely to choose credit lines for liquidity management, andfirms with high liquidity risk (type H) are more likely tochoose cash.

The intuition for this result is straightforward. As theprobability of the liquidity shock increases, monitoringbecomes increasingly expensive due to the direct monitor-ing cost and revocation of credit line access. Thus, firmswith high liquidity risk prefer to avoid monitored liquidityinsurance and use cash for liquidity management.

Hedging needs and the choice between cash and credit lines:We now allow firms to vary with respect to their correlationbetween Date 1 cash flows and the investment opportunity τ.Specifically, we compare two types of firms. A firm with lowhedging needs (LHN) has υH4υL and, consequently, υ 0 �υo0[notice that υ 0 �υ ¼ �ðυH�υLÞðλ0 �λÞ]. This firm has a greaterprobability of having the investment opportunity τ in the highcash flow state ð1�bλÞ. A firmwith high hedging needs (HHN)has υH ¼ υL � υ, or the same probability of the investmentopportunity in both states. We let υ¼ υ4υL, so that theexpected arrival of the investment opportunity is identical forthe two types of firms. This setup also implies that υ 0 ¼ υ forthe high hedging needs firm. Both firms have the sameprobability bλ, that is, the implications of this section hold

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319 295

while controlling for variation in liquidity risk. We can showProposition 5:

Proposition 5. The firm with low hedging needs is more likelyto use credit lines for liquidity management, when comparedwith the firm with high hedging needs. That is, ðUC�Un

L ÞHHN4 ðUC�Un

L ÞLHN . Thus, if ðUC�Un

L ÞHHN4 is lower thanzero, then ðUC�Un

L ÞLHN must be lower than zero.

Two effects differentiate low and high hedging needsfirms. First, the firm with high hedging needs faces agreater cost of using the monitored credit line because itsinvestment opportunities tend to be concentrated in stateswith low cash flow (in which the credit line is likely to berevoked). Second, the firm with low hedging needs has agreater incentive to avoid the low cash flow state becauseits investment opportunities are positively correlated withcash flows. This effect increases the benefit of the liquidinvestment and the monitored credit line for the firm withlow hedging needs.

18 For simplicity, in this extension we shut down the new investmentopportunity by making υ¼ 0.

2.4.4. ExtensionsIn this subsection we consider three extensions of the

basic framework above. First, we analyze the implementa-tion of cases in which the incentives to engage in illiquid-ity transformation are weak. Second, we analyze theimplications of the case in which monitoring might notbe incentive compatible for both the firm and the bank.Third, we extend the model to allow for a continuousdistribution of liquidity shocks. The goal is to generate alink between pledgeable income and liquidity risk.

Hedging needs and fully committed liquidity insurance:If ρ01þυ 0ρτrρ1þυρτ , then illiquidity transformation is notan issue. In such a case, Corollary 1 shows that the liquidproject is always chosen. Because monitoring is notrequired to implement the optimal solution in this case,the firm can access fully committed liquidity insurance.

This case is more likely to happen when υ�υ 0 is high.Because υ�υ 0 ¼ ðυH�υLÞðλ0 �λÞ, this difference increases ashedging needs decrease (that is, as υH�υL becomes larger).Thus, firms with low hedging needs should find it easier toobtain fully committed liquidity insurance for liquid pro-jects. This result arises from the same incentive effectmentioned in the proof to Proposition 5: The firmwith lowhedging needs has a greater incentive to avoid the lowcash flow state because its investment opportunities arepositively correlated with cash flows.

In the absence of other frictions, the firm can use bothcash or a credit line to implement the liquid project in thiscase. However, even a small liquidity premium woulddrive the firm to prefer a fully committed credit line tocash. The implication that follows is that fully committedcredit lines should be more common for low hedgingneeds firms, which are less likely to require monitoringand credit line revocation.

The other implications of the model continue to holdonce we allow for this case. In particular, the relationbetween hedging needs and the choice of liquiditymechanism (Proposition 5) is unchanged, because fullycommitted liquidity insurance is more likely to be optimalfor low-hedging needs firms.

Monitoring not incentive-compatible: If there are incen-tives for illiquidity transformation (ρ01þυ 0ρτ4ρ1þυρτ) butmonitoring is not incentive compatible [that is, if eitherEq. (11) or Eq. (27) does not hold], the liquid project cannotbe implemented and the firm will choose the illiquidproject (Corollary 1). This particular case is more likely toarise when the monitoring cost c or the payoff of theilliquid project ρ01 is high.

As in Section 2.4.2, the firm should prefer to use cash toprovide liquidity insurance for the illiquid project. In thissense, the key implication of Section 2.4.2 continues tohold if we allow some firms not to satisfy the incentivecompatibility conditions for monitoring: Firms that chooseto invest in high liquidity risk projects find it optimal toself-insure against such shocks using cash, and firms thatchoose to invest in projects with low liquidity risk manageliquidity through a monitored credit line.

Pledgeable income and liquidity risk: In the model abovethe probability λ is an exogenous parameter that establishesthe probability that the firm suffers a liquidity shortfall. Thisprobability should be a function of variables such as thefirm's cash flow risk and the firm's ability to raise externalfinance. The cash flow risk interpretation directly matchesthe model presented above (see footnote 5). However, inthe model, no link exists between the firm's ability to raiseexternal finance (captured by ρ0) and liquidity risk.

This lack of link between liquidity risk and pledgeableincome is an artificial feature that is caused by thebinomial structure of the model. It is straightforward toextend the model to a more general case in which the Date1 liquidity shock bρ can assume values in a range [0; ρmax]according to a distribution function bF ð�Þ. This extensionallows us to derive implications relating pledgeableincome and the choice between cash and credit lines.18

The main result, which we prove in the Appendix, isthat a decrease in pledgeable income ρ0 makes it morelikely that the firm chooses cash instead of the credit line.The intuition for this result is that a decrease in pledgeableincome increases the firm's liquidity risk (the probabilitythat it requires liquidity infusions from the credit line) and,consequently, the monitoring cost of the credit line.As pledgeable income decreases, the bank's incentive tomonitor the firm increases, increasing direct and indirectmonitoring costs.

3. Empirical implications

The first set of empirical predictions of the modelfocuses on the role of revocations of lines of credit as amonitoring mechanism to prevent illiquidity transforma-tion by firms and the implications of this monitoringmechanism for optimal liquidity management. The keyimplication that we test is as follows.

1.

An increase in liquidity risk causes firms to switch fromcredit lines to cash for their liquidity management.

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319296

Firms that suffer a high risk of facing credit line revoca-tion (those with high liquidity risk) find it more costly to

employ monitored liquidity insurance and switch to cashholdings.19 A natural determinant of firm's liquidity risk isthe variability in firm cash flows. Firms with greater cashflow variance face a higher risk of facing liquidity shortfalls.This line of reasoning suggests Prediction 1a.

1a.

1

assupremanceholdprovof thpremamoliqui

2

typicencoexist

Firms with low cash flow risk are more likely to use creditlines instead of cash for liquidity management, whencompared with firms with high liquidity risk.

In addition, the model relates liquidity risk to pledge-able income. For a given level of cash flow risk, firms thathave higher pledgeable income face a lower risk of facingfuture liquidity shortfalls. Thus, the model also deliversPrediction 1b.

1b.

Firms with high pledgeable income are more likely to usecredit lines instead of cash for liquidity management,when compared with firms with low pledgeable income.

Pledgeable income is a direct function of the firm'sability to raise external funds. Credit ratings should cap-ture heterogeneity in pledgeable income, to the extent thatthey capture the ease of accessing public bond markets.The model would thus predict that rated firms should bemore likely to use credit lines than non-rated firms, andfirms with high credit ratings should be more likely to usecredit lines than firms with low credit ratings. An alter-native proxy for pledgeable income is the tangibility offirm's assets. Firms with more tangible assets should bemore likely to use credit lines for liquidity management.Finally, one could argue that firm size is correlated withpledgeable income. Larger firms are more transparent andhave easier access to external finance.

These implications are broadly consistent with otherresults reported in the literature. The standard approach inthe existing empirical literature is to examine the cross-sectional association between firm-level variables (such ascash flow risk, credit ratings and tangibility) and corporateliquidity policy.20

Nevertheless, these existing tests suffer from the usuallimitations associated with cross-sectional regressions.Unobservable firm-level variation can make it difficult tointerpret the coefficients on standard proxies for liquidityrisk. For example, some firms might not have access to

9 As discussed in Section 2.2, this implication is derived under themptions that the firm can carry cash without incurring a liquidityium and that the bank can provide actuarially fair liquidity insur-through credit lines. However, this implication would continue toif we had introduced liquidity premia and costly credit line

ision in the model. The key point is that these costs are independente liquidity risk mechanism. For example, having a positive liquidityiumwould make it less likely that a firm would use cash (for a givenunt of liquidity risk), but does not change the comparative statics ondity risk itself.0 Standard datasets that contain information on credit lines areally short in the time dimension. For example, Sufi's (2009) datampass the period 1996–2003. Thus, most of the variation drivinging results in the literature is cross-sectional.

bank-provided insurance due to lack of reputation andtrack record. These firms are also likely to have lowercredit ratings and to be riskier than other firms. In thiscase, a negative correlation between liquidity risk proxiesand credit line usage cannot be interpreted as evidencethat liquidity risk causes firms to switch to cash holdings.Reverse causality is also a possibility. To wit, firms couldhave high credit ratings because they have access to creditline insurance.

To provide evidence that liquidity risk causes firms toswitch from credit lines to cash-based liquidity insurance,one would like to trace the effects of a shock that changesthe liquidity risk (but not other fundamentals) of a groupof firms, while leaving other firms unaffected. It is verydifficult to think of a shock that changes cash flow riskwithout affecting other fundamentals such as the produc-tivity of investment. Shocks that affect the ability of agroup of firms to raise external funds, while arguablyleaving fundamentals unaffected, provide a more suitablesetting for a test of Prediction 1.

We propose two identification strategies, which wediscuss further in Section 4.3. We first rely on a quasi-natural experiment and examine the evolution of liquidityratios during the downgrade to junk status of GeneralMotors Corp. and Ford Motor Co. that occurred in thespring of 2005. Due to their importance as issuers inthe public bond market, the downgrade had an impacton the liquidity of the bond market as a whole. Firmsthat depended on bond financing and needed financingbecame suddenly exposed to the effects of the downgradeand found it more difficult to raise debt in the form ofbonds. Notably, the firms that were most affected by thisshock were well-established, mature firms that have goodaccess to bank financing. And for many of these firms(notably those outside of the auto sector), fundamentalswere likely not affected by the bond market shock. In thissense, evidence that bond-reliant firms switch from creditlines to cash in the aftermath of the GM–Ford crisis isconsistent with the hypothesis that changes in liquidityrisk causes firms to change their liquidity policy.21 Second,we examine the effect that the price pressure associatedwith mutual fund redemptions has on liquidity ratios. Ourhypothesis is that firms whose equity is subject to down-ward price pressure could face a higher cost of issuingequity and that this lower ability to raise equity decreasesfirms' pledgeable income and increases their liquidityrisk.22

The second set of empirical implications of the modelrelates to the relation between credit lines, hedging needs,covenants, and revocations. Proposition 5 states that firmswith low hedging needs are more likely to use credit linesfor liquidity management, when compared to firms withhigh hedging needs. An empirical proxy for the firm'shedging needs is the correlation between cash flows and

21 Deadweight costs of credit line provision by banks (which areabsent from the model) could have amplified this effect, because banksthemselves faced greater costs of borrowing from the bond market in theaftermath of the GM–Ford crisis.

22 We thank our referee for suggesting this alternative identificationstrategy.

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319 297

investment opportunities (as in Acharya, Almeida, andCampello, 2007). Empirical Prediction 2 stems from themodel:

2.

Firms with a high correlation between cash flows andinvestment opportunities (low hedging-needs firms)should be more likely to use credit lines instead of cashfor liquidity management, when compared with firmswith a low correlation (high hedging needs firms).

Our model also has predictions for the relation betweenhedging needs and covenants in credit line contracts.As we discuss in Section 2.4.4, the presence of a futureinvestment opportunity could provide incentives for firmsto avoid illiquidity transformation independently of mon-itoring, and these incentives are stronger for firms whoseinvestment opportunities tend to arrive in high cash flowstates (low hedging needs firms). Prediction 3 derives fromdifferential incentive effect.

3.

Credit lines for low hedging needs firms are less likely tocontain covenants than credit lines for high hedging needsfirms.

Because low-hedging needs firms can access creditlines that contain less contractual restrictions, Prediction4 should result.

4.

Low hedging needs firms are less likely to face revocationof bank credit lines than high hedging needs firms.

We also use our data to track down the followingpredictions about other characteristics of the monitoringmechanism, which according to the theory is behind theempirical relation between liquidity risk and liquiditymanagement.

5.

Credit line contracts contain covenants contingent on firmprofitability, and access to credit lines is sometimesrestricted when firm profitability decreases.

In the model, this pattern is part of an optimal liquiditymanagement policy that discourages credit line users fromengaging in illiquidity transformation. In particular, themodel explains why credit line revocation is concentratedin states in which the firm reports negative profitabilityshocks. Such shocks increase the demand for credit lineinsurance and, thus, the bank's incentives to monitor toavoid credit line drawdowns by the firm.

Existing evidence on the frequency of credit line revo-cations is mixed. Some papers report significant revo-cation following covenant violations that are triggeredby declines in profitability (e.g., Sufi, 2009), while otherpapers argue that revocations are infrequent (Barakova andParthasarathy, 2012; Berrospide, Meisenzahl, and Sullivan,2012). We use the Capital IQ data to provide additional largesample evidence on the frequency and magnitude ofrevocations.

The revocability of credit line access is not incompatiblewith credit lines' role as liquidity insurance. If revocation

does not occur, a firm facing a liquidity shock coulddraw down on the credit line to meet the shortage ofliquidity.

6.

Credit line drawdowns are more likely to happen follow-ing decreases in firm profitability.

Thus, both drawdowns and revocations should be morelikely in low profitability states. In addition, if credit lineusers have low liquidity risk when compared with firmsthat employ cash, the following prediction should hold.

7.

Credit line drawdowns are relatively infrequent, so thatcredit line drawdowns to meet liquidity needs are sig-nificantly less frequent than reductions in cash holdings tomeet liquidity needs.

The ex post usage of credit lines and cash to meetliquidity needs should be consistent with ex ante differ-ences in liquidity risk exposure across firms. We use ourdata to verify whether these predicted patterns are con-sistent with data on real world credit lines.

4. Empirical tests

In this section we present our empirical analysis.

4.1. Sample construction and description

We obtain firm-level data from the Capital IQ andCompustat databases for the period 2002–2011. Werestrict ourselves to US firms covered on both databasesand traded on AMEX, NASDAQ, or NYSE. We removeutilities (Standard Industrial Classification(SIC) codes4900-4999) and financial firms (SIC codes 6000-6999).Following Bates, Kahle, and Stulz (2009), we furtherremove firm-years with negative revenues, as well asnegative or missing assets, obtaining in the end a sampleof 32,671 firm-years involving 4,741 unique firms.

CIQ compiles detailed information on capital structureand debt structure by going through financial footnotescontained in firms' 10-K SEC filings. Most important forour purposes, firms provide detailed information on thedrawn and undrawn portions of their lines of credit in theliquidity and capital resources section under the manage-ment discussion, or in the financial footnotes explainingdebt obligations, and CIQ compiles these data. We use theinformation from CIQ to construct a dummy for thepresence of a credit line, which is equal to one if the firmhas a positive amount of undrawn credit reported in the10-K. Following Sufi (2009), we also construct a measureof the amount of undrawn credit expressed as a percen-tage of net book assets. Following Lemmon, Roberts, andZender (2008), we compute a set of firm characteristicssuch as profitability, market-to-book (M/B), and tangibility.We also compute firm-year rating as the average monthlyrating by Standard and Poor’s (S&P) (item 280), afterconverting the S&P rating into numbers. Credit spread isthe spread on U.S. corporate bond yields between Moody'sAAA and BAA provided by Datastream, based on averages

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319298

of seasoned issues. We compute the firm's asset (unlev-ered) beta, calculated from equity (levered) betas anda Merton-KMV formula as in Acharya, Almeida, andCampello (2013). Finally, following standard procedures,all variables are winsorized at the 0.5% level in both tails ofthe distribution. See Table A8 in the appendix for acomplete set of definitions.

4.2. Descriptive statistics

Table 1 provides univariate evidence on the differencesin firm characteristics across the samples of firms with andwithout a line of credit. In Column 1, we report mean andmedian values for the entire sample. Columns 2 and 3contain values for the subsamples of firms with andwithout a line of credit, respectively.

Table 1 allows for a broad comparison of firms with andwithout a line of credit. The main picture that emerges is

Table 1Comparison of firms with and without credit lines.

This table provides summary statistics for the entire sample and for the restrconsists of non-utilities (excluding Standard Industrial Classification (SIC) codescovered by both Capital IQ and Compustat from 2002 to 2011. We remove firmabove filters, the sample consists of 32,671 firm-year observations involving 4,74All variables are winsorized at the 0.5% in both tails of the distribution. The laundrawn credit using the unequal variances T-test and the two-sample Wilcoxo

(1) (2)

Entire sample Sample of fiwith a cre

line

Variable Mean Mean[Median] [Median

Cash/Net Asset 0.657 0.350[0.147] [0.094]

Credit Lines/Net Asset 0.106 0.156[0.071] [0.124]

Cash/Net Asset (market value) 0.115 0.093[0.054] [0.043]

Credit Lines/Net Asset (market value) 0.067 0.096[0.041] [0.071]

Profitability 0.059 0.097[0.107] [0.119]

Size 2181.0 2618.9[317.2] [453.8]

Book Leverage 0.211 0.232[0.152] [0.192]

M/B 1.728 1.576[1.232] [1.152]

Tangibility 0.248 0.273[0.167] [0.197]

NWC/Asset 0.050 0.076[0.042] [0.064]

Capex/Asset 0.054 0.057[0.033] [0.036]

R&D/Sales 0.370 0.136[0.005] [0.000]

Dividend Payer Dummy 0.275 0.330[0.000] [0.000]

Beta KMV 1.245 1.176[1.099] [1.054]

Rating Dummy 0.265 0.327[0.000] [0.000]

Number of observations 32,671 22,186

that the sample of firms with a line of credit is significantlydifferent from the rest along all the dimensions reported inthe table. Firms with a line of credit are more profitable,are more leveraged, are more likely to pay dividends, havelower beta, and are more likely to be rated. These firmsalso invest more in working capital and capex, but theyhave lower Research and Development (R&D) expenses.Overall, these characteristics suggest that firms with a lineof credit are more established, mature firms with fewergrowth opportunities and more stable cash flows. Table 1is also informative on the relative sizes of lines of creditand cash across the two samples. Focusing on firms with aline of credit, Column 2 shows that average cash holdingsare larger than average credit lines, as a percentage of netbook value assets (35% versus 15.6%). However, the mag-nitude of the ratio for cash and credit lines is similar if wedivide by the market value of assets: respectively, 9.3%and 9.6%.

icted samples of firms with and without a credit line. The entire sample4900-4949) and non-financials (excluding SIC codes 6000-6999) US firms-years with negative revenues and negative or missing assets. After the1 unique firms. In this table, size is measured as the book value of assets.st two columns test for differences between samples with and withoutn rank-sum (Mann-Whitney) test.

(3) (4) (5)

rms Sample of firms Test of differencedit without a credit with vs. without a credit line

line

Mean T-test Mann-Whitney] [Median] P-value P-value

1.517 45.658 80.044[0.576] (0.000) (0.000)

0.185 33.527 50.649[0.100] (0.000) (0.000)

�0.048 �46.411 �52.245[0.045] (0.000) (0.000)952.8 �30.560 �55.191[118.5] (0.000) (0.000)0.151 �30.678 �49.071[0.017] (0.000) (0.000)2.195 28.841 32.753[1.582] (0.000) (0.000)0.177 �39.567 �47.168[0.094] (0.000) (0.000)�0.019 �43.110 �44.129[-0.015] (0.000) (0.000)0.045 �16.503 �32.671[0.023] (0.000) (0.000)1.034 29.731 63.398[0.106] (0.000) (0.000)0.117 �49.989 �40.888

[0.000] (0.000) (0.000)1.419 13.428 12.227[1.247] (0.000) (0.000)0.092 �59.701 �46.715[0.000] (0.000) (0.000)10,485

Fig. 2. Distribution of liquidity shocks. This figure presents evidence onthe differences in the distribution of profitability and cash flows overassets between firms with a credit line (LC Firms) and firms without one(Cash Firms). The probability density displayed is an estimate using thedata in the sample based on a normal kernel function and evaluated atone hundred equally spaced points that cover the range of the datawinsorized at the 0.5% level on both tails.

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319 299

Several of our model's predictions are consistent withthis univariate evidence. Firms that use credit lines forliquidity management have more tangible assets andhigher credit ratings than those that use cash for liquiditymanagement. This evidence is also in line with that reportedin previous literature.23

Fig. 2 illustrates the distribution of profitability andcash flows over assets for firms with and without a line ofcredit (LC firms and cash firms, respectively). The figureshows that firms that use cash have a higher probability ofhaving low profitability and low cash flows than firms thatuse credit lines.

23 In Table A1 in the Appendix, we provide multivariate evidence onthe relation between credit line usage and firm characteristics. Thepatterns that we uncover using Capital IQ data are very similar to thosereported in existing literature.

4.3. Implication 1: liquidity risk and liquidity management

Our model predicts that an increase in liquidity riskshould cause firms to switch from credit lines to cash intheir liquidity management. We test this implication in thissubsection using two alternative identification strategies.

4.3.1. The GM–Ford downgradeIn May 2005 Standard and Poor's downgraded the

ratings of bonds issued by GM and GMAC, its financialarm, from BBB� to BB, two notches below investmentgrade. Similarly, Ford and FMCC, its financial subsidiary,had their ratings reduced from BBB� to BBþ , one notchbelow investment grade. According to Acharya, Schaefer,and Zhang (2008), the downgrade led to a market widesell-off of the corporate bonds issued by these two firms.It also had a significant impact on inventory risk faced byfinancial intermediaries that operated as market makersfor the securities issued by the two automakers. Theeffect of the downgrade went beyond the bond marketsof GM and Ford and of other producers in the auto sector,and it was perceived at the time to be potentially long-lasting. Because of their size and their importance in thedebt markets, the credit deterioration of the two giantautomakers affected the functioning of corporate-bondmarkets as a whole. Acharya, Schaefer, and Zhang showthat, simultaneously with the downgrade, there wasexcess comovement in the fixed-income securities of allindustries, not just in those of auto firms.

An article that appeared in the May 12, 2005 issue ofThe Economist observes that “…the most obvious instancesof this [rising interest rates on American companies'bonds] have been the bonds of two giant carmakers,General Motors and Ford, which Standard & Poor's (S&P),a rating agency, downgraded to junk status on May 5th.However, the malaise in the market for corporate debtgoes far wider than these two big names….Severalplanned bond issues have been postponed because inves-tors are becoming more demanding. Plenty of companiesbesides GM and Ford have been marked down.” All of thishappened in the context of a booming economy, and thesame article comments that “[b] By many measures,America's economy continues in strikingly good health”.The last statement suggests that the effect of the GM–Forddowngrade was a shock primarily to the corporate bondmarket not to the economy at large.

The downgrade of GM–Ford offers an opportunity toidentify the effects of liquidity risk on liquidity policy. Ourempirical strategy is to examine whether firms that faced alarger increase in liquidity risk (firms that are heavilyfinanced by publicly traded bonds, the treatment group)increased their cash holdings, as a share of total liquidity,more than the rest of the firms (control group). In normaltimes, bond-dependent firms are less likely to be exposedto liquidity risk because they tend to be less financiallyconstrained and more profitable. However, the key identi-fication assumption for our tests is that the GM–Fordshock increases liquidity risk for bond-dependent firms,more so than for firms that do not rely as much on bondmarkets for their financing.

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319300

The other key assumption behind our identificationstrategy is that no unobservable variables explain both afirm's reliance on bonds (which is endogenously deter-mined) and its sensitivity to the GM–Ford shock. This isthe exclusion restriction for our test. As a first step towardensuring that our exclusion restriction holds, we sort firmsaccording to the average exposure to bond financingmeasured at the industry level. This sorting eliminatesthe possibility that a firm-level unobserved characteristicthat varies within industries could explain the results.24

Specifically, to identify the treatment firms, we constructthree sorting variables: (1) the ratio of a firm's outstandingbonds over the book value of its assets; (2) a dummy forfirms with outstanding bonds greater than 50% of debt;(3) a dummy for being rated. In our main specification, thecrisis period lasts from December 2004 to May 2005.25

Accordingly, for each of these variables we compute theindustry average in fiscal year 2003 at the three-digit SICcode and their average over the whole sample in the years2003 and 2004. All the firms whose industry has anaverage higher than the sample average are identified asbelonging to the treatment group. For example, if anindustry has a higher percentage of rated firms in 2003than the sample average, all firms in the industry belong tothe treatment group for both fiscal years 2003 and 2004.

We conduct a difference-in-difference (DID) analysis ina regression framework. For all observations that occurduring the crisis period the variable crisisi;t takes the valueof one, and zero otherwise. We exclude from the analysisall firms with a rating below B� as they are too close todefault. To focus only on a pure liquidity risk event, and toexclude possible supply effects, we drop all firm-years forwhich reporting occurred after May 2005. In other words,we include only firm-years for which reporting occurredduring any month of the fiscal years 2002–2004 (accord-ing to Compustat, May 2005 belongs to fiscal year 2004).Finally, we require all firms to have data for every period infiscal years 2003–2005.

Because this is a DID empirical design, time-invariantvariables (even unobservable ones) are differenced out anddo not compromise identification. We further verify ouridentification assumptions by conducting a placebo test thatshows that the differential change in liquidity managementacross bond-reliant and other firms obtains only in the after-math of GM–Ford. The placebo crisis is defined as the periodthat goes from December 2003 to May 2004. For this exercise,we classify firms as being in the treatment group dependingon whether they were rated in fiscal year 2002. This placebotest helps narrow down the set of possible alternative

24 Measuring bond reliance at the industry-level does not alleviate allendogeneity concerns, because the unobservable characteristic thatexplains both bond reliance and the sensitivity to GM–Ford could becorrelated across firms in the same three-digit SIC industry. We provideadditional support for our identification strategy below.

25 The crisis began with the downgrade of GM and GMAC by S&P andMoody's in October 2004. In March 2005, GM issued a profit warning andwas subsequently downgraded by Fitch and Moody's. In April 2005, Fordissued a profit warning, which subsequently led to its downgrade by allthree rating agencies. In May 2005, both automakers were excluded fromMerrill's and Lehman's investment-grade indices. See Acharya, Schaefer,and Zhang (2008) for a detailed timetable of the events.

explanations for the results. If the results obtain only followingGM–Ford, any alternative explanation must also predict a shifttowards cash for bond-dependent firms that happens only inthe treatment period, not in other periods.

Table 2 illustrates the results of our analysis of the GM–

Ford crisis period.26 Our base specification is provided incolumn 1 as follows:

cashcashþundrawn

� �i;t¼ α0þα1 crisisi;tþα2 treatmenti

þα3ðcrisisi;tntreatmentiÞþ firm controlsi;tþ industry meani;tþεi;t ð18Þ

The coefficient of interest is α3, which captures theestimate of the difference-in-difference. We expect treat-ment firms to be more exposed to the liquidity effects ofthe GM–Ford downgrade. Therefore, α3 should be positivein Column 1 across Panels A–C.

Column 1 of Panel A shows that treatment firms had onaverage around 7.4% less cash as a share of total liquidityrelative to control firms before the crisis (α2).27 During thecrisis control firms decreased their cash by 1.9% (α1), andtreatment firms reduced their cash by 3.2% less thancontrol firms. Therefore, the difference in cash holdingsbetween the two groups decreased in absolute terms toapproximately 4% during the crisis (α3�α2Þ. These findingsare confirmed for the other sortings reported in Column 1of Panels B and C. Overall these results are in line with ourprediction that bond dependent firms shifted from lines ofcredit to cash as a result of an increase in liquidity risk.

In Columns 2–4 we examine which margins were at workduring the GM–Ford crisis, by looking at the behavior of cash,undrawn credit lines, and the sum of cash plus undrawncredit lines, all computed as a percentage of net assets. Wefind that while cash holdings decreased around 2% on averagein the control group during the crisis, they remained essen-tially unchanged for treatment firms ðα3�α1Þ. This evidence isconsistent with our interpretation of the GM–Ford event as arise in future liquidity risk instead of as a current liquidityshock. Under the liquidity shock interpretation, treatmentfirms would have been expected to decrease their cash as apercentage of assets during the crisis period.

In Column 3 we find that undrawn credit lines did notvary for the control group, and they decreased by around1% as a percentage of assets during the crisis for thetreated group. The drop in credit lines for treatment firmscould indicate that they drew down on their outstandingcredit lines or opened less or smaller credit lines, bothpossibly as a result of increased liquidity risk and bothconsistent with our theoretical predictions. It could alsomean, however, that treatment firms faced a restriction inaccess to credit lines during the GM–Ford event that didnot affect control firms as much. While this mechanism isnot inconsistent with our model, it suggests that the down-grade affected the liquidity of banks, which reacted byrevoking their outstanding credit lines. Likely, revocations

26 In Table A2 in the Appendix we report the complete set ofcoefficients which are not shown in Table 2.

27 This difference could reflect the lower liquidity risk of bond-dependent firms.

Table 2The effect of the General Motor (GM)–Ford crisis on liquidity management.

This table examines how the GM–Ford crisis affected liquidity management. We construct three measures of bond dependence: bonds to assets, dummyfor firms with bonds greater than 50% of debt and a dummy for being rated. For each of these variables, we compute the average at the three-digit StandardIndustrial Classification in 2003 and the average over the whole sample. The industry dummies take value one if the industry average is above the sampleaverage. These dummies identify three separate treatment groups, respectively, employed in Panels A–C. The dummy for the crisis equals one if the annualreport falls in the period that goes from December 2004 to May 2005. The interaction term is the product of the treatment group dummies and the crisisdummy. We regress four measures of liquidity (Columns 1–4) on the crisis, treatment and interaction dummies controlling for (unreported) firmProfitability, Size, M/B, Tangibility, NWC/Assets, Capex/Assets, R&D/Sales, Dividend Payer Dummy, Beta KMV and Rating Dummy. We drop firms in the autosectors. We include only firm-years for which reporting occurred during any month of the fiscal years 2003–2004. We cluster errors at the firm level. nnn, nn,and n denote statistical significance at the 1%, 5%, and 10% level, respectively.

(1) (2) (3) (4)Cash/ Cash/ Undrawn credit Total liquidity

(cash þ credit lines) net assets lines/net assets /net assets

Panel A: Sort by industry average bonds to assets ratio

Crisis dummy �0.019nn �0.024nnn 0.002 �0.023nnn

(0.009) (0.004) (0.003) (0.005)Treatment dummy �0.074nnn �0.021nnn 0.011nn �0.009

(0.014) (0.006) (0.005) (0.007)Treatment n Crisis 0.032nn 0.020nnn �0.012nn 0.008

(0.013) (0.006) (0.005) (0.008)R-squared 0.390 0.405 0.177 0.197

Panel B: Sort by industry percentage of firms with at least 50% of debt in bonds

Crisis dummy �0.018nn �0.023nnn 0.002 �0.022nnn

(0.009) (0.005) (0.003) (0.006)Treatment dummy �0.054nnn �0.010n 0.013nnn 0.002

(0.014) (0.006) (0.005) (0.007)Treatment n Crisis 0.025nn 0.015nn �0.011nn 0.005

(0.013) (0.006) (0.005) (0.008)R-squared 0.387 0.403 0.178 0.197

Panel C: Sort by industry percentage of firms that are rated

Crisis dummy �0.015n �0.029nnn 0.000 �0.029nnn

(0.008) (0.005) (0.003) (0.006)Treatment dummy �0.082nnn �0.020nnn 0.017nnn �0.003

(0.016) (0.006) (0.005) (0.007)Treatment n Crisis 0.028nn 0.032nnn �0.009n 0.024nnn

(0.013) (0.006) (0.005) (0.007)R-squared 0.391 0.405 0.179 0.198

Number of observations 4,705 4,689 4,707 4,689

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319 301

occurred at a faster rate for treated firms because theirliquidity position had deteriorated more. Alternatively,treatment firms could have violated some of the covenantsassociated with their credit lines, and these violations couldhave led to a revocation of the line.28 Finally, Column 4shows that overall liquidity dropped for all firms during thecrisis, but relatively less for treatment firms (Panel C).29

The analysis of a placebo crisis is reported in Table 3.It shows that all firms had lower cash holdings during the

28 In unreported analysis we find that treatment firms were notexposed to an increase in revocations of credit lines during the crisismonths relative to control firms and that treatment firms did not sufferan increase in covenant violations relative to control firms. This evidencesuggests that bank credit line revocations were not behind the decreasein undrawn credit experienced by treatment firms in the aftermath of theGM–Ford downgrade.

29 In unreported tests, we collect analyst earnings forecasts data fromInstitutional Broker Estimate System (I/B/E/S) and construct a proxy forasymmetric information based on analyst forecast dispersion. All ourresults are robust to the inclusion of this variable as an additional control.

proposed placebo crisis and that treatment firms did notbehave differently from control firms. These results sup-port our identification assumptions.30

The results show that bond-dependent firms shiftedfrom credit lines to cash in the aftermath of the GM–Fordshock.31 Does this result simply reflect a shift fromexternal to internal finance engendered by an increase inthe cost of external finance? To address this concern, weperform two complementary tests.

30 In Table A3 of the Appendix we examine whether we can findevidence of changes in liquidity policy following changes in marketconditions. A natural hypothesis is that the results should be symmetric,that is, bond-dependent firms should increase their reliance on creditlines in periods of hot bond markets. We find evidence consistent withthis hypothesis. The results must be interpreted with caution becausechanges in market conditions are associated with changes in firmfundamentals.

31 Table A4 provides some evidence that in 2005 there was a partialreversal of the reported effects on cash for treatment firms. Specifically,rated firms decreased their reliance on cash holdings.

Table 3Liquidity management during a placebo event.

This table provides a robustness check on Table 2 using a different definition of crisis. We construct a dummy for the crisis if the annual report falls in theperiod that goes from December 2003 to May 2004. nnn, nn, and n denote statistical significance at the 1%, 5%, and 10% level, respectively.

Cash/(Cash þ Credit Cash/Net Assets Undrawn Credit Total LiquidityLines) Lines/Net At /Net Assets(1) (2) (3) (4)

Panel A: Sort by industry average bonds to assets ratio

Crisis dummy �0.056nnn 0.001 0.010nnn 0.010n

(0.008) (0.005) (0.003) (0.005)Treatment dummy �0.051nnn �0.017nnn 0.004 �0.013n

(0.013) (0.006) (0.005) (0.007)Treatment n Crisis 0.004 0.013nn �0.001 0.012n

(0.013) (0.006) (0.004) (0.007)

R-squared 0.309 0.409 0.134 0.205

Panel B: Sort by industry percentage of firms with at least 50% of debt in bonds

Crisis dummy �0.058nnn 0.001 0.011nnn 0.012nn

(0.009) (0.005) (0.003) (0.006)Treatment dummy �0.060nnn �0.014nn 0.009nn �0.006

(0.013) (0.006) (0.004) (0.007)Treatment n Crisis 0.008 0.011n �0.004 0.007

(0.013) (0.006) (0.004) (0.008)R-squared 0.310 0.409 0.135 0.204

Panel C: Sort by industry percentage of firms that are rated

Crisis dummy �0.052nnn 0.004 0.007nnn 0.011nn

(0.008) (0.005) (0.003) (0.005)Treatment dummy �0.072nnn �0.023nnn 0.009n �0.013n

(0.015) (0.006) (0.005) (0.007)Treatment n Crisis �0.002 0.006 0.004 0.011

(0.013) (0.006) (0.005) (0.007)R-squared 0.312 0.410 0.136 0.204

Number of observations 4,775 4,759 4,777 4,759

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319302

First, in Table 4 we examine how cash was raisedduring the GM–Ford crisis. Panel A relates the change incash holdings to the main components of the cash flowstatement, both during the GM–Ford event and outside theevent. The results show that some of the channels thatoperate in normal times were shut down during the crisisin the subset of bond-dependent firms, in particular debtissues. At the same time, however, equity issues remainedan active source of cash during the crisis for bond-dependent firms. These results suggest that firms thattypically rely on bonds were forced to rely more on equityissuance to raise cash during the GM–Ford crisis period.These results also suggest that bond-reliant firms were notcut off from all sources of external financing in the after-math of GM–Ford, and could use equity issuances to raisecash. This flexibility could be due to the fact that bond-reliant firms are relatively less financially constrained thanan average firm.

Panel B relates changes in undrawn credit to draw-downs, controlling for a set of firm characteristics. Therelation between drawdowns and changes in undrawncredit became stronger during the crisis, suggesting thatfirms used their existing lines of credit possibly to increasetheir cash holdings.

Second, in Table A5 in the Appendix we sort firmsaccording to their external finance dependence, computed

as in Rajan and Zingales (1998). If the effects of GM–Fordsimply reflect a substitution of external for internalfinance, then we might expect externally dependent firmsto be affected more than the rest of the firms. On thecontrary, we find that during the crisis more externallydependent firms decreased their cash ratios as a percen-tage of total liquidity and as a percentage of assets andthen increased their exposure to credit lines. This evidencesuggests that the GM–Ford event was not a widespreadshock to externally dependent firms, but rather a specificshock to bond-dependent firms.

We also investigate if covenants on credit lines becamestricter during the GM–Ford event for firms in our controlgroup. For this purpose, we first collect the list of cove-nants and the associated ratios on all the credit linesissued by the firms in our sample over the period Decem-ber 2004–May 2005 (i.e., the same employed for the GM–

Ford tests). We adjust all ratios by their standard deviationand construct an index of covenant slackness based on thestandardized ratios. A higher value of the index means lesstight covenants. We then replicate the GM–Ford tests ofTable 2, using the index of covenant slackness as adependent variable. The regressions are executed at thefacility-firm-year level, instead of at the firm-year level, aseach facility carries different covenants. The results of theanalysis are reported in Table A6 and they show that

Table 4How was the increase in cash financed during the General Motors (GM)–Ford event?

In Panel A this table examines how the increase in cash holdings was financed for bond-dependent firms during the GM–Ford event. In Panel B, it showshow much of the credit lines were drawn. We construct three measures of bond dependence: a dummy for bonds greater than 5% of assets, a dummy forfirms with bonds greater than 50% of debt and a dummy for being rated. We construct a dummy for the crisis if the annual report falls in the period thatgoes from December 2004 to May 2005. In Panel A, we regress the change in cash holdings on various items from the cash flow statement, for firms withbonds during the crisis (Columns 1, 3 and 5) outside the crisis (Columns 2, 4 and 6) in the three subsets of bond-dependent firms. In Panel B, we regress thechange in undrawn credit lines on the change in drawn credit lines, controlling for a number of firm characteristics, across different subsets as in Panel A.Controls include (unreported) firm Profitability, Size, M/B, tangibility, NWC/Assets, Capex/Assets, R&D/Sales, Dividend Payer Dummy and Rating Dummy.We cluster errors at the firm level. nnn, nn, and n denote statistical significance at the 1%, 5%, and 10% level, respectively.

Firms with bonds Firms with bonds/debt 450% Rated firms(1) (2) (3) (4) (5) (6)

In crisis Not in crisis In crisis Not in crisis In crisis Not in crisis

Panel A: Change in cash holdings year on year

Net income þ depreciation 0.101nnn 0.094nnn 0.102nnn 0.110nnn 0.077nnn 0.085nnn

(0.028) (0.011) (0.029) (0.011) (0.029) (0.010)Capex �0.032 �0.061nnn �0.022 �0.083nnn �0.002 �0.049nnn

(0.045) (0.015) (0.048) (0.017) (0.046) (0.015)Change working capital �0.245nnn �0.241nnn �0.236nnn �0.259nnn �0.240nnn �0.256nnn

(0.048) (0.020) (0.049) (0.021) (0.049) (0.020)Dividends 0.107 �0.126nnn 0.028 �0.184nnn 0.171 �0.115nn

(0.130) (0.046) (0.132) (0.049) (0.132) (0.047)Net equity issues 0.114nnn 0.066nnn 0.126nnn 0.078nnn 0.119nnn 0.069nnn

(0.024) (0.011) (0.026) (0.011) (0.026) (0.011)Net debt issues 0.044 0.073nnn 0.044 0.086nnn 0.015 0.050nnn

(0.028) (0.012) (0.032) (0.014) (0.030) (0.012)Constant 3.416 6.608nnn 4.868n 8.355nnn 6.492 8.395nnn

(2.548) (1.019) (2.592) (1.035) (4.419) (1.694)Number of observations 1,002 6,045 949 5,590 666 4,093R-squared 0.211 0.111 0.213 0.126 0.189 0.105

Panel B: Change in undrawn credit lines year on year

Change in drawn credit lines �0.572nnn �0.481nnn �0.564nnn �0.474nnn �0.568nnn �0.473nnn

(0.135) (0.042) (0.156) (0.051) (0.154) (0.047)Number of observations 833 5,066 790 4,656 585 3,547R-squared 0.089 0.077 0.091 0.074 0.081 0.066

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319 303

covenants were slacker outside the crisis for treatmentfirms (positive coefficient on the non-interacted treatmentterm), but they became tighter during the crisis (negativecoefficient on the interaction term). This finding is con-sistent with the hypothesis that banks anticipated poten-tial refinancing difficulties for treatment firms andaccordingly imposed tighter covenants. It is also consistentwith the main prediction of the model that treatmentfirms shifted to cash during the crisis as managing liquid-ity via credit lines became more costly in terms of tightercovenants.

32 See Appendix A of Edmans, Goldstein, and Jiang (2012) for adetailed description of how MFFLow is constructed.

4.4.2. Mutual fund redemptionsOur interpretation of the evidence relies on a key

exclusion restriction: no unobservable variables explainboth a firm's reliance on bonds and its sensitivity to theGM–Ford shock. Because bond dependence is endogen-ously determined, we cannot entirely rule out this possi-bility. Thus, to provide additional evidence for the causaleffect of liquidity risk on liquidity management, we com-plement the GM–Ford test with an alternative identifica-tion strategy. This alternative strategy relies on thehypothesis that firms whose equity is subject to down-ward price pressure could face refinancing problems.

Insofar as a firm has refinancing needs that are to be metby issuing equity, price pressure increases liquidity risk.

Following Edmans, Goldstein, and Jiang (2012), wecollect data on mutual fund redemptions and use themto construct a measure of price pressure for the firms inour sample. We compute fund flows (MFFlow) as the saleof a firm's equity as a percentage of its dollar tradingvolume, associated with mutual fund outflows greaterthan 5%.32 The main argument for using this measure ofprice pressure is that it is associated with mutual fundtrading due to investor flows, not firm fundamentals. Thus,such price pressure should affect liquidity policy onlythrough its effects on equity mispricing. Furthermore, theadvantage of using equity price pressure is that it is notdirectly related to debt financing and can, therefore, bemore exogenous when examining changes in firm demandfor credit lines.

Our main evidence on fund redemptions is reported inTable 5. We construct two measures of outflow: (1) adummy equal to one if the firm has been subject to anoutflow and (2) a continuous measure Outflow defined as

Table 5The effect of mutual fund redemptions on liquidity management.

This table examines how price pressure by fund outflows relates to liquidity management. We compute fund outflows as the absolute value of MFFlow asin Edmans, Goldstein, and Jiang (2012). Our outflow variable measures the sale of a firm's equity as a percentage of its dollar trading volume, associatedwith mutual fund outflows greater than 5%. All regressions include the following (unreported) firm characteristics: industry CF Volatility, Size, M/B,Tangibility, NWC/Assets, Capex/Assets, R&D/Sales, Dividend Payer Dummy and Rating Dummy. We cluster errors at the firm level.

n, nn, nnn denotes two-tail statistical significance at 10%, 5% and 1%, respectively. nnn, nn, and n denote statistical significance at the 1%, 5%, and 10% level,respectively.

Cash/(cash þ Cash/net assets Undrawn credit Total liquiditycredit lines) lines/net asset /net assets

(1) (2) (3) (4)

Panel A: Dummy equals one if the firm has been subject to an outflow

Dummy outflow 0.013nn 0.004 �0.005nn �0.002(0.007) (0.009) (0.003) (0.009)

R-squared 28,934 28,854 28,946 28,814Number of observations 0.357 0.280 0.101 0.229

Panel B: Continuous measure of outflow

Outflow 0.473nnn �0.590nnn �0.123n �0.712nnn

(0.103) (0.058) (0.071) (0.086)R-squared 28,934 28,854 28,946 28,814Number of observations 0.356 0.280 0.101 0.229

Panel C: Continuous measure of outflow conditional on having an outflow

Outflow 0.555nnn �0.589nnn �0.177nnn �0.766nnn

(0.101) (0.089) (0.066) (0.079)Number of observations 12,173 12,131 12,176 12,115R-squared 0.385 0.257 0.110 0.186

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319304

the absolute value of MFFlow. We run a set of regressionsin which our measures of liquidity management arerelated to the two measures of outflows across the wholesample (Panels A and B) and for the subsample of firms forwhich an outflow occurred (Panel C). In all regressions wecontrol for a large set of firm characteristics. The mainresult of the table is to show that there is an effect ofoutflows on liquidity management: As downward pricepressure increases, firms shift to cash holdings in theirliquidity management (Column 1).

One important difference between the results of Table 5and those following the GM–Ford shock is that firm cashholdings and total liquidity tend to decrease followingmutual fund outflows (Columns 2 and 4), while both cashand total liquidity increased for bond-dependent firmsfollowing GM–Ford. This decrease in cash is probably dueto the fact that firms that face downward price pressuredue to outflows are forced to use up some of their cash tofinance operations.33 Still, their usage of credit linesdecreases even more (Column 3), generating the increasein the ratio of cash to total liquidity shown in Column 1.Thus, firms rely more on cash for their liquidity manage-ment following a shock to their ability to raise equity.These results provide further support to the empiricalimplications of our model.

33 Bond-dependent firms have higher credit quality than an averagefirm and, thus, could be shielded from the need to use up cash to financeoperations following GM–Ford.

4.4. Liquidity management and hedging needs

In this subsection we test the predictions of the modelregarding the relation between hedging needs and a firm'sliquidity policy. We build on Acharya, Almeida, and Campello(2007) and Disatnik, Duchin, and Schmidt (2010) to constructproxies for industry annual cash flow at the three-digit SICcode. Similarly we compute the mean industry annualinvestment activities (Hedging based on Investment Activ-ities), and mean industry annual market-to-book ratio (Hed-ging based on Tobin's Q).

More precisely we first compute the mean industryTobin's Q. We then compute the correlation between cashflows and, respectively, investment activities and market-to-book. We use industry-level variables to mitigate thepossibility that cash flows are endogenously related withinvestment and, therefore, with Tobin's Q. To furtheraddress this issue, we calculate industry-level investmentopportunities using data only for financially unconstrainedfirms. Firms are defined to be financially unconstrained ifthey pay dividends, have assets above $500 million and arerated above Bþ .

4.4.1. Prediction 2: hedging needs and the use of linesof credit

We first test the empirical prediction that low hedgingneeds firms are more likely to use lines of credit instead ofcash for liquidity management, when compared with highhedging needs firms. Table 6 compares several measures ofliquidity policy by sorting firms into high and low hedgingneeds categories. High (low) hedging needs firms are those

Table 6Credit line usage across top and bottom quintiles of hedging needs.

This table provides summary statistics across groups of high versus low correlation of investment opportunities and cash-flow (low and high hedgingneeds, respectively). Hedging needs are calculated at the three-digit Standard Industrial Classification code industry level as the correlation between meanindustry annual cash flow and, respectively, mean industry annual investment activities (Hedging based on Investment Activities, item 311 þ 46), andmean industry annual Tobin's Q (Hedging based on Tobin's Q). High (low) hedging needs firms are those with a correlation in the bottom (top) quintile.In Panel A, we examine the relationship between hedging needs and four liquidity variables. In Panel B we first regress the four liquidity variables on a setof firm characteristics (Profitability, Size, Book Leverage, M/B, Tangibility, NWC/Assets, Capex/Assets, R&D/Sales, Dividend Payer Dummy, CF Volatility, BetaKMV, and Rating), and then relate the residuals of those regressions to hedging needs. In both panels we include a test for differences between sampleswith high and low hedging needs using the unequal variances T-test.

Presence of a credit Undrawn credit/ Undrawn credit/ Revolving creditline (dummy) net assets total liquidity /net assets

Panel A: Variables:Mean (Median)

Industry median investmentactivities

High hedging needs 0.540 0.079 0.233 0.023N¼6,704 (1.000) (0.015) (0.054) (0.000)Low hedging needs 0.818 0.126 0.536 0.050N¼6,545 (1.000) (0.102) (0.595) (0.000)

T-test 5.88 6.16 7.23 2.61(P-value) (0.000) (0.000) (0.000) (0.011)

Industry median Tobin's QHigh hedging needs 0.602 0.096 0.304 0.027N¼6,611 (1.000) (0.044) (0.177) (0.000)Low hedging needs 0.797 0.129 0.493 0.044N¼6,133 (1.000) (0.105) (0.538) (0.000)

T-test 2.37 2.17 2.25 2.18(P-value) (0.020) (0.032) (0.026) (0.031)

Panel B: Residuals: Mean (median)

Industry median investmentactivities

High hedging needs �0.055 �0.011 �0.029 �0.003N¼6,815 (0.295) (�0.043) (�0.064) (�0.010)Low hedging needs 0.236 0.006 0.039 0.003N¼6,518 (0.410) (�0.017) (0.083) (�0.019)

T-test 2.98 1.86 1.37 0.84(P-value) (0.004) (0.067) (0.173) (0.402)

Industry median Tobin's QHigh hedging needs 0.105 0.005 0.042 0.001N¼6,611 (0.405) (�0.029) (0.009) (�0.008)Low hedging needs 0.182 0.004 0.010 0.002N¼6,133 (0.396) (�0.020) (0.035) (�0.019)

T-test 1.11 �0.13 �1.38 0.23(P-value) (0.268) (0.898) (0.172) (0.822)

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319 305

with a correlation in the bottom (top) quintile.34 The tabledisplays the differences in the presence of a line of credit,the amount of undrawn credit as a share of net assets, theshare of undrawn credit in total liquidity, and total revol-ving credit over net assets. In Panel A we calculate the rawstatistics for the four liquidity variables across the sampleof high and low hedging needs firms. In Panel B we relatehedging needs with the residuals obtained from regres-sions in which we control for a set of firm characteristics.By looking at regression residuals, instead of at the rawvariables, we can evaluate the relation between hedgingneeds and our four liquidity variables after controlling forfirm characteristics.

The evidence of Table 6 is consistent with the predic-tion of the model. Using the measure of hedging needs

34 We report other cutoffs in Table A7 in the Appendix.

based on industry median investment activities, Panel Ashows that the percentage of firms with a credit lineis 81.8% among firms with low hedging needs and it isonly 54.0% for firms with high hedging needs. Low hed-ging needs firms have more undrawn credit both as apercentage of assets (12.6%) and as a percentage of totalliquidity (53.6%), than high hedging needs firms (respec-tively 7.9% and 23.3%). Low hedging needs firms also havea higher percentage of revolving credit as a percentage ofnet assets (5.0%) than high hedging needs firms (2.3%).Panel B confirms the results provided in Panel A, usingregression residuals.

4.4.2. Prediction 3: hedging needs and the use of covenantsTable 7 estimates the relation between the covenants

attached to credit lines and hedging needs. We obtain dataon covenants from LPC Dealscan and examine all thecovenants on credit lines for the firms in our sample

Table 7Hedging needs and covenants on credit lines.

This table estimates the relation between hedging needs and the use of covenants on credit lines using a Poisson specification. We obtain data oncovenants from LPC Dealscan. We list all the covenants attached to credit lines for the firms in our sample during the period 2002–2008. When firms aregranted several new credit line facilities in the same year we report the median value. General Purpose LC is a dummy that takes value one if the statedpurpose of the line of credit is General Corporate Purposes, as reported in LPC Dealscan. All regressions include year fixed effects. nnn, nn, and n denotestatistical significance at the 1%, 5%, and 10% level, respectively.

Covenant Index Covenant Index Covenant Index Covenant IndexDrucker and Puri (2009) Demiroglu and James (2010) (Only CF Covenants) (Only Sweeps)

(1) (2) (3) (4) (5) (6) (7) (8)

Correlation investment activities �0.079nnn �0.077nnn �0.208nnn �0.170nnn

(0.026) (0.030) (0.041) (0.047)Correlation Tobin's Q �0.058nn �0.035 �0.148nnn �0.170nnn

(0.027) (0.031) (0.042) (0.047)Profitability 0.208n 0.205n �0.162 �0.175 0.248 0.221 �0.254 �0.244

(0.112) (0.112) (0.122) (0.122) (0.176) (0.176) (0.196) (0.198)Size �0.240nnn �0.241nnn �0.242nnn �0.243nnn �0.275nnn �0.279nnn �0.360nnn �0.362nnn

(0.008) (0.008) (0.010) (0.010) (0.013) (0.013) (0.015) (0.015)Book Leverage 0.698nnn 0.704nnn 0.718nnn 0.723nnn 0.764nnn 0.769nnn 1.405nnn 1.408nnn

(0.046) (0.046) (0.053) (0.053) (0.071) (0.071) (0.075) (0.075)M/B �0.100nnn �0.098nnn �0.139nnn �0.136nnn �0.103nnn �0.096nnn �0.200nnn �0.196nnn

(0.012) (0.012) (0.014) (0.014) (0.019) (0.019) (0.024) (0.024)Rated 0.155nnn 0.148nnn 0.141nnn 0.133nnn 0.119nnn 0.114nnn 0.343nnn 0.337nnn

(0.026) (0.026) (0.030) (0.030) (0.042) (0.042) (0.047) (0.047)CF Volatility �1.968nnn �2.030nnn �1.435nn �1.481nn �1.756n �1.840nn �1.872n �2.039n

(0.591) (0.592) (0.650) (0.652) (0.921) (0.922) (1.076) (1.083)General Purpose LC �0.226nnn �0.230nnn �0.249nnn �0.254nnn �0.218nnn �0.224nnn �0.395nnn �0.405nnn

(0.020) (0.020) (0.023) (0.023) (0.032) (0.032) (0.035) (0.035)Ln(Maturity) of LC 0.406nnn 0.407nnn 0.472nnn 0.473nnn 0.472nnn 0.473nnn 0.747nnn 0.744nnn

(0.022) (0.022) (0.026) (0.026) (0.036) (0.036) (0.045) (0.045)Facility Amount/Asset �0.600nnn �0.613nnn �0.185nn �0.209nnn �0.848nnn �0.889nnn �1.275nnn �1.298nnn

(0.073) (0.072) (0.079) (0.078) (0.116) (0.115) (0.136) (0.135)Number of observations 4,667 4,634 4,667 4,634 4,667 4,634 4,667 4,667Pseudo R2 0.0967 0.0995 0.0994 0.0826 0.0820 0.138 0.138 0.0967

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319306

during the period 2002–2008. In most cases, firms aregranted several new credit line facilities in the same year.For these cases we report the median number of covenantsacross facilities.

We use four measures of covenant intensity. The firstmeasure of covenant intensity is based on the covenantindex of Drucker and Puri (2009), which is computed asthe sum of all the covenants included in a loan agreement.Our second measure is borrowed from Demiroglu andJames (2010), and it is itself based on the covenantintensity index originally introduced by Bradley andRoberts (2004). This index consists of covenants that limitthe actions of the borrower or give lenders rights that arecontingent on adverse future events. The index is com-posed of six types of covenants: asset sales sweeps,collateral releases, debt issuance sweeps, dividend restric-tions, a dummy for financial covenants, and equity issu-ance sweeps. Our third measure of covenant intensityfocuses on the covenants that relate directly to cash flows.These covenants primarily require minimum profitabilitylevels, minimum interest coverage ratios, and restrictionson cash flow usage. Our fourth measure of covenantintensity is based on the presence of sweeps, such as assetsale sweeps, debt issuance sweeps, and excess cash flowsweeps. These covenants impose restrictions on managers'payout and reinvestment policy, giving preference to debtreimbursement over other possible uses of cash flows.

We run a set of Poisson regressions using our fourcovenant intensity measures as dependent variables. Among

the explanatory variables, the variables of main interest arethe two hedging measures already employed in Table 6, thefirst one based on investment activities and the second one onTobin's Q. We also control for a set of variables identified byDemiroglu and James (2010) as relevant for the use ofcovenants.

According to our model, we should expect a positiverelation between the intensity of covenants and hedgingneeds, meaning that credit lines for low hedging needs firmstend to carry fewer covenants. The evidence provided inTable 7 suggests that this is the case, as covenants are lessprevalent for firms that have high correlation between cashflows and investment opportunities (low hedging needsfirms). Focusing on the covenant intensity measure mostrelevant to the type of bank monitoring described in ourmodel, the cash flow covenant index, Column 5 shows that,for the hedging needs measure based on investment activities,an increase by two standard deviations in the hedgingvariable (0.751), i.e., an increase in correlation between cashflows and investment opportunities, is associated on averagewith having 14.4% (¼ 1�e�0:208�0:751) less covenants, hold-ing the other variables in the model constant.

4.4.3. Prediction 4: hedging needs and credit line revocationsTo test the implication that low hedging needs firms

are less likely to face revocation of credit lines, we limit thesample to firms that have undrawn credit available under acredit line in period t�1, and study whether access to theline of credit is restricted during period t. We consider a

Table 8Revocations and drawdowns.

This table presents Probit (marginal effects dF/dx) and ordinary least square regression results for the contemporaneous relation between restriction ofaccess to credit lines, profitability and hedging needs (Columns 1–4) and for the relationship between drawdown of credit lines, profitability, and hedgingneeds (Columns 5–6). In Columns 1–2 the dependent variable is a dummy for a full revocation of the credit line. In columns 3–4 the dependent variable is adummy for a decrease in undrawn credit greater than 50% of the outstanding amount. In columns 5–6 the dependent variable is the annual change indrawn credit lines as a percentage of total assets. Year, rating and exchange fixed effects included. Rating fixed effects are based on 22 rating dummies andthe unrated dummy. Robust standard errors clustered at the firm level are reported in parentheses. nnn, nn, and n denote statistical significance at the 1%, 5%,and 10% level, respectively. All regressions include a constant term (unreported).

Full revocation Full or partial (450%) Drawdown ofof credit line revocation of credit line credit lines(dummy) (dummy)

(1) (2) (3) (4) (5) (6)

Hedging based on investment activities �0.009nnn �0.012nn 0.001(0.003) (0.006) (0.001)

Hedging based on Tobin's Q �0.010nnn �0.014nn 0.003nn

(0.004) (0.006) (0.001)Profitability �0.051nnn �0.050nnn �0.113nnn �0.111nnn �0.024nnn �0.023nnn

(0.010) (0.010) (0.018) (0.018) (0.007) (0.007)Size �0.000 �0.001 �0.002 �0.002 0.001 0.001

(0.001) (0.001) (0.002) (0.002) (0.000) (0.000)Book Leverage �0.037nnn �0.038nnn �0.047nnn �0.050nnn 0.046nnn 0.046nnn

(0.008) (0.008) (0.013) (0.013) (0.005) (0.004)M/B 0.003nnn 0.004nnn 0.003 0.003n �0.001n �0.001n

(0.001) (0.001) (0.002) (0.002) (0.001) (0.001)Tangibility �0.021nn �0.016n �0.008 �0.004 �0.026nnn �0.026nnn

(0.009) (0.009) (0.014) (0.014) (0.004) (0.004)NWC/Assets �0.070nnn �0.070nnn �0.142nnn �0.142nnn �0.004 �0.005

(0.008) (0.008) (0.014) (0.014) (0.004) (0.004)Capex/Assets �0.033 �0.039 �0.226nnn �0.224nnn 0.172nnn 0.171nnn

(0.033) (0.032) (0.055) (0.054) (0.020) (0.019)R&D/Sales 0.001 0.001 �0.002 �0.001 �0.002 �0.002

(0.002) (0.002) (0.003) (0.003) (0.001) (0.001)Dividend Payer Dummy �0.014nnn �0.014nnn �0.020nnn �0.020nnn 0.005nnn 0.005nnn

(0.003) (0.003) (0.005) (0.005) (0.001) (0.001)CF Volatility 0.066 0.068 0.341nnn 0.338nnn �0.237nnn �0.239nnn

(0.054) (0.054) (0.088) (0.087) (0.031) (0.031)Beta KMV 0.002n 0.002n 0.002 0.002 0.000 0.000

(0.001) (0.001) (0.002) (0.002) (0.001) (0.001)

Number of observations 14,874 14,998 14,887 15,011 14,931 15,058R-squared 0.141 0.142 0.093 0.093 0.059 0.059

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319 307

full revocation to occur in period t when a firm has apositive amount of undrawn credit in period t�1 andnone in period t, and the amount of drawn credit has notincreased between the two periods. For robustness, wealso consider partial revocations by examining losses of atleast 50% of undrawn credit over one year.35 At theunivariate level, the annual frequency of revocations is4.1% and that of a reduction in credit lines greater than 50%is 7.9%. In the subsample of firms with negative profits,these percentages are, respectively, 15.3% and 23.5%.36

Table 8 provides evidence on the relation between revoca-tions of credit lines and hedging needs, controlling for a set of

35 As firms typically hold lines of credit with more than one bank,our measure of partial revocations could capture full revocations ofindividual lines.

36 Both our measures of revocations, full and partial, might over-estimate true revocations because they treat voluntary non-renovationsof credit lines as revocations. However, two reasons suggest that thisoverestimation might be small, particularly for full revocations. First,undrawn credit lines are not very expensive; annual fees on the undrawnportions of credit lines are around 25 basis points on average (Sufi, 2009).Second, firms could be unwilling to lose their credit lines as it could senda negative signal to investors.

key firm characteristics. The evidence reported is consistentwith our theory. Column 1 shows that an increase by twostandard deviations (0.751) in the hedging variable based oninvesting activities, i.e., a decrease in hedging needs, isassociated on average with a decrease of 0.7% (¼ �0:009�0:751) in the probability of facing a revocation, holdingother variables constant. This effect is similar for the hedgingneeds measure based Tobin's Q (Column 2). The evidence forpartial revocations is less strong, suggesting that the effect ofhedging needs on the probability of revocations is mainlyassociated with full revocations.

4.4.4. Prediction 5–7: profitability, revocations, anddrawdowns

With respect to the relation between profitability andrevocations, Columns 1–4 of Table 8 show that access toundrawn credit depends positively on profitability.37 A two

37 For each regression specification we provide several differentdefinitions of profitability: the standard definition of profitability com-puted as operating income before depreciation, scaled by assets; adummy for profitability being positive; a dummy for profitability beingabove 5%; the change in profitability, measured as the difference between

Table 9Probability of a liquidity event.

This table examines the probability of a drawdown in credit lines or a reduction in cash associated with a drop in profitability. In Panel A, we compute theprobability of profitability falling below 0% and 5%, respectively for firms with and without a credit line. In Panel B, we define liquidity events as follows: forcredit line drawdowns, a liquidity event occurs if there is an increase in drawn revolving credit (ΔRC40) while profitability is negative; for cash, a liquidityevent occurs if there is a reduction in cash (ΔCash and ST Investmentso0) while profitability is negative. The probability of a liquidity event for credit linesis computed as the ratio of credit line liquidity events divided by the number of firm-years with a credit line. The probability of a liquidity event for cash iscomputed as the ratio of cash liquidity events divided by the number of firm years without a credit line.

Panel A: Probability of a negative cash flow shock

With Without T-statistics Wilcoxoncredit line credit line (P-value) (P-value)

Probability of profitso0% 0.128 0.368 52.644 59.627(0.000) (0.000)

Probability of profitso5% 0.203 0.476 55.001 58.113(0.000) (0.000)

Panel B: Probability of a liquidity event

T-statistics WilcoxonCredit Line Cash (P-value) (P-value)

Probability of a liquidity event 0.016 0.185 40.494 58.691(0.000) (0.000)

Probability of a liquidity event 40:5% of assets 0.014 0.179 39.726 57.833(0.000) (0.000)

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319308

standard deviations increase in profitability (0.432) isassociated with a decrease of 2.2% (¼�0.051�0.432) inthe probability of facing a full revocation (Column 1) and a4.9% (¼�0.113�0.432) decrease in the probability offacing a revocation of at least 50% of undrawn credit(Column 3).

Table 8 also provides evidence on the relation betweencredit line drawdowns and profitability. The evidence onthe revocability of credit line access is not incompatiblewith credit lines' role as liquidity insurance and, if revoca-tion does not occur, a firm facing a liquidity shock coulddraw down on the credit line to meet the shortage ofliquidity. To see if this is the case, in Columns 5 and 6 ofTable 8, we regress the annual variation in revolving crediton profitability. The regression shows a negative andsignificant relation between profitability and variations intotal drawn lines of credit, as predicted by our model. Thisevidence shows that an increase in revolving credit isassociated with a drop in profitability, and it suggests thatcredit lines are employed by firms to withstand liquidityshocks resulting from a shortfall in cash flows.38

Prediction 7 says that firms that use credit lines forliquidity management should be less likely to draw on

(footnote continued)profitability in year t minus profitability in year t�1; and one variable forthe positive changes in profitability (increases in profitability) and one forthe negative changes (decreases in profitability). Results (unreported) arequalitatively unaffected for these other measures of profitability.

38 For robustness, we also run unreported tests in which we restrictvariations in revolving credit to be positive. In this way, we can betterisolate the variations in revolving credit that arise from a drawdown,from the variations that arise from repayments. We also run a specifica-tion in which the dependent variable takes value zero if the change inrevolving credit is non-positive. The results obtained in these additionalregressions confirm the above results.

credit lines to meet liquidity needs, when compared withfirms that rely on cash holdings. We first test this predic-tion by comparing the frequency of low profit realizationsfor firms with and without a credit line. In particular, wecompare the frequency of profitability being below 0% and5%. The evidence, reported in Panel A of Table 9, showsthat on average credit line users are significantly less likelyto face a negative or low profitability shock.

Next, we examine the probability that firms with andwithout credit lines use their liquidity to meet a shortfallin cash flows. For this purpose, we construct a newvariable that is a dummy taking the value of one when adrawdown occurs during a period in which profitability isnegative. We then calculate the frequency of this event(liquidity event) in the sample of firms with a line ofcredit. To compare this with firms that do not use creditlines, we calculate a similar variable for cash drawdownsand compare both frequencies. The evidence is reported inPanel B of Table 9. In line with the empirical prediction ofthe model, liquidity events associated with drawdowns ofcredit lines are significantly less frequent than liquidityevents associated with reductions in cash holdings.

5. Conclusions

Recent empirical evidence on corporate liquidity man-agement suggests that bank credit lines do not offer fullycommitted liquidity insurance and that they are used notonly for precautionary motives, but also to finance futuregrowth opportunities. In this paper, we propose and test atheory of corporate liquidity management that is consis-tent with these findings. We argue that a corporate creditline can be understood as a form of monitored liquidityinsurance, which controls illiquidity-seeking behavior byfirms through bank monitoring and credit line revocation.

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319 309

In addition, we allow firms to demand liquidity to hedgeagainst negative profitability shocks and to help financefuture growth opportunities. We show that bank monitor-ing is more costly for firms that have greater arrival ofgrowth opportunities in states in which cash flows are low(high hedging needs firms). Such firms find it morebeneficial to move to cash holdings, to avoid revocationcosts associated with credit lines.

We use a novel dataset on corporate credit lines toprovide empirical evidence that is consistent with thepredictions of the model. The evidence suggests that creditline users have lower liquidity risk than firms that use cashfor liquidity management. The causality from liquidity riskto liquidity management is supported by two alternativetests that both show that firms that were more exposed toan increase in liquidity risk moved away from credit linesinto cash holdings. In addition, we find evidence that firmswith high hedging needs are more likely to use cashinstead of credit lines for liquidity management. Thesefirms also face more covenants and credit line revocationswhen they do use credit lines, when compared with firmswith low hedging needs.

Several interesting avenues shed further light on therelation between illiquidity transformation and bank mon-itoring. One is the empirical relation between the ex postreported purpose of line of credit drawdowns and thepresence of covenants. Credit lines designated and primar-ily used for activities with low illiquidity-seeking risk, suchas working capital management, could reflect fewer fea-tures of monitored insurance than credit lines used foractivities with high illiquidity-seeking risk, such as mergersand acquisitions. Another avenue is to study carefully theaftermath of a covenant violation for a firm's line of creditand, in particular, the factors that determine whethercovenants are waived by banks providing the credit lineor whether they instead generate credit revocations. Finally,material adverse clauses (MACs) are another way throughwhich banks could employ their monitoring information.While invoked infrequently, the off-equilibrium threat ofMACs could have significant impact on firm incentives toengage in illiquidity-seeking activities.

Appendix A

In this appendix, we provide proofs and develop someof the arguments in the main body of the paper in greaterdetail.

A.1. Moral hazard and pledgeability

To see how moral hazard generates limited pledgeabil-ity, consider the following setup. If the firm continues untilDate 2, the investment produces a Date 2 cash flow bR,which obtains with probability p. With probability 1�p,the investment produces nothing. The probability of suc-cess depends on the input of specific human capital bythe firms' managers. If the managers exert high effort, theprobability of success is equal to pG. Otherwise, theprobability is pB, but the managers consume a privatebenefit equal to bB. While the cash flow bR is verifiable, themanagerial effort and the private benefit are neither

verifiable nor contractible. Because of the moral hazarddue to this private benefit, managers must keep a highenough stake in the project to be induced to exert effort.We assume that the investment is negative NPV if themanagers do not exert effort, implying the followingincentive constraint:

pGbRMZpBbRMþbB;or bRMZ

bBΔp

; ð19Þ

where bRM is the managers' compensation and Δp¼ pG�pB.This moral hazard problem implies that the firms' cashflows cannot be pledged in their entirety to outsideinvestors. In particular, we have that

cρ0 � pG bR� bBΔp

!ocρ1 � pGbR: ð20Þ

The manager's choice of bλ impacts the project's cashflows and private benefits in the following way. If themanager chooses the illiquid project (bλ ¼ λ0), we have thatbR ¼ R0 and bB ¼ B0. If the manager chooses the liquid project(bλ ¼ λ), we have that bR ¼ R and bB ¼ B. Finally, we have thatR04R and B04B. Thus, the illiquid project produces ahigher cash flow when it is successful and a higherassociated private benefit. We make the following assump-tion:

ρ00 � pG R0 � B0

Δp

� �¼ ρ0 � pG R� B

Δp

� �: ð21Þ

Thus, bR and bB change in a way that leaves ρ0 constant.

A.2. Proof of Proposition 1

Suppose the firm chooses the liquid project and pro-mises a payment D to the bank. Let us first show whenfully committed liquidity insurance is feasible.

Sticking to the liquid investment produces a payoff forthe firm equal to

ð1�λÞðρ1þυHρτ�DÞþλðρ1þυLρτ�DÞ ¼ ρ1�Dþυρτ : ð22ÞThat is, the firm uses the credit line to continue the projectand fund the new investment, and it repays D to the bankin both states. If the firm deviates and shifts funds into theilliquid project, the payoff is

ð1�λ0Þðρ01þυHρτ�DÞþλ0ðρ01þυLρτ�DÞ ¼ ρ01�Dþυ 0ρτ: ð23ÞThus, for any D, fully committed insurance is feasible

when ρ1þυρτZρ01þυ 0ρτ . In this case, D is defined as

D�υτ�λρ¼ I: ð24Þ

By Assumption (4), it is possible to find a solution Drρ0 tothis equation. Thus, the bank breaks even and the firmcaptures the NPV of the project:

UL � ρ1� I�λρþυðρτ�τÞ40; ð25Þby assumption (2).

If ρ1þυρτoρ01þυ 0ρτ , thenmonitoring is required to imple-ment the liquid project. Let the payment to the bank, DM,be defined as in Eq. (13). Because DM�υLτ�ρrρ0�υLτ�ρ

Table A1Access to credit lines.

This table relates firm characteristics to various liquidity measures. In Columns 1 and 2 we run probit regressions (marginal effects dF/dx) for thepresence of a credit line, where the dependent variable is a dummy that takes value one, if the firm has an undrawn credit line. Column 1 is for all the firmsin the sample, and Column 2 is for the subsample of rated firms (4CCC). In Columns 3 and 4, we run ordinary least squares (OLS) regressions for cash andshort-term investments as a percentage of total liquidity. Total liquidity is computed as cash and short-term investments plus undrawn credit. Column 3 isfor subsample of rated firms (4CCC), and column 4 is for the subsample of rated firms that have a credit line. In column 5, we run an OLS regression forcash and short-term investments as a percentage of total assets, on the subsample of rated firms (4CCC) with a credit line. Year and exchange fixed effectsare included. Robust standard errors clustered at the firm level are reported in parentheses. nnn, nn, and n denote statistical significance at the 1%, 5%, and10% level, respectively.

(1) (2) (3) (4) (5)

Presence of a credit line Cash/ Cash/(dummy) total liquidity net assets

All firms Rated Rated Rated with a line Rated with a lineof credit of credit

Profitability 0.362nnn 0.116n �0.190nn 0.041 �0.078(0.041) (0.069) (0.086) (0.087) (0.080)

Size 0.002 �0.013nn 0.048nnn 0.041nnn �0.003(0.005) (0.006) (0.007) (0.007) (0.004)

Book Leverage 0.271nnn 0.080nnn �0.338nnn �0.314nnn �0.202nnn

(0.034) (0.028) (0.037) (0.035) (0.032)M/B �0.035nnn �0.029nnn 0.082nnn 0.055nnn 0.061nnn

(0.005) (0.006) (0.009) (0.009) (0.011)Tangibility 0.201nnn 0.043 �0.156nnn �0.093nnn �0.119nnn

(0.039) (0.032) (0.038) (0.034) (0.018)NWC/Assets 0.504nnn 0.134nnn �0.430nnn �0.410nnn �0.286nnn

(0.036) (0.043) (0.056) (0.053) (0.036)Capex/Assets 0.210n 0.015 �0.553nnn �0.637nnn �0.126nnn

(0.126) (0.123) (0.137) (0.115) (0.043)R&D/Sales �0.023nn �0.304nnn 0.037n 0.878nnn 0.555nnn

(0.009) (0.084) (0.022) (0.155) (0.159)Div. Payer Dum. 0.076nnn 0.015 �0.071nnn �0.058nnn �0.006

(0.015) (0.011) (0.015) (0.014) (0.008)CF Volatility �2.172nnn �0.977nn 3.591nnn 2.948nnn 1.059nnn

(0.555) (0.382) (0.537) (0.596) (0.323)Beta KMV �0.004 0.002 0.029nnn 0.031nnn 0.017nnn

(0.003) (0.004) (0.005) (0.005) (0.003)Rating �0.009nnn �0.012nnn 0.019nnn 0.017nnn 0.008nnn

(0.001) (0.003) (0.004) (0.003) (0.002)Number of observations 23,653 6,848 6,839 6,166 6,172R-squared 0.272 0.259 0.287 0.286 0.239

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319310

[by assumption (5)], assumption (4) ensures that DMrρ0, sothis payment is feasible.

First, we show when the bank has incentives tomonitor and deny access to the credit line if s¼ s0, giventhat the firm follows the equilibrium strategy (bλ ¼ λ).Conditional on the liquidity shock, if the bank does notmonitor it must honor the credit line and its payoff is�ðρ�υLτ�DMÞ. If the bank does monitor, it obtains a signals¼ s0 with probability μ and it can deny access to the creditline. Its payoff is then �ð1�μÞðρ�υLτ�DMÞ�c. So as longas μðρþυLτ�DMÞ4c [Eq. (11)] it is incentive-compatiblefor the bank to monitor even when the bank anticipatesthat the firm has made the correct choice and picked bλ ¼ λ.Incentive compatibility is preserved because of the nega-tive NPV feature of the Date 1 loan. In terms of a creditline, the bank loses money when the firm draws on thecredit line. Thus, the optimal contract can rely on thebank's incentives to deny access to liquidity insurance (thecredit line) to induce good behavior by the firm.

Given that the bank is expected to monitor, if the firmchooses the right project its payoff (after the initial contractis written) is ð1�λÞðρ1þυHρτ�DMÞþλð1�μÞðρ1þυLρτ�DMÞ,

while illiquidity transformation produces the payoff ð1�λ0Þðρ01þυHρτ�DMÞ. So, ifð1�λÞðρ1þυHρτ�DMÞþλð1�μÞðρ1þυLρτ�DMÞ

4ð1�λ0Þðρ01þυHρτ�DMÞ; ð26Þ

then the firm chooses the liquid project.The bank breaks even and the firm captures the NPV of

the liquid project net of monitoring costs:

Un

L ¼ UL�λ½cþμ½ρ1�ρþυLðρτ�τÞ��: ð27ÞFinally, if either Eq. (11) or (26) does not hold, monitor-

ing is not incentive-compatible and, thus, the liquidproject cannot be implemented. □

A.3. Proof of Proposition 2

Suppose the firm chooses the illiquid project. It mustthen promise a payment D0 to outside investors such thatthey break even:

D0 �υ 0τ�λρ¼ I: ð28Þ

Table A2Table 2 with coefficients for all explanatory variables.

This table is an exact replica of Table 2 and it reports the entire list of coefficients for all the explanatory variables. nnn, nn, and n denote statisticalsignificance at the 1%, 5%, and 10% level, respectively.

Cash/ Cash/ Undrawn credit Total liquidity(cash þ credit net assets lines/ /net assets

lines) net assets(1) (2) (3) (4)

Panel A: Sort by industry average bonds to assets ratio

Crisis dummy �0.019nn �0.024nnn 0.002 �0.023nnn

(0.009) (0.004) (0.003) (0.005)Above mean industry bond/Asset �0.074nnn �0.021nnn 0.011nn �0.009

(0.014) (0.006) (0.005) (0.007)Crisis n above mean industry bond/asset 0.032nn 0.020nnn �0.012nn 0.008

(0.013) (0.006) (0.005) (0.008)Profitability �0.027 0.020 0.075nnn 0.094nnn

(0.039) (0.025) (0.015) (0.025)LnRevenues �0.027nnn �0.015nnn 0.003nn �0.012nnn

(0.004) (0.002) (0.001) (0.002)M/B 0.029nnn 0.022nnn �0.002 0.020nnn

(0.004) (0.002) (0.001) (0.002)Tangibility �0.398nnn �0.200nnn 0.021n �0.179nnn

(0.038) (0.013) (0.012) (0.016)NWC/Asset �0.699nnn �0.212nnn 0.186nnn �0.026

(0.034) (0.014) (0.014) (0.018)Capex/Asset �0.048 �0.007 0.064 0.061

(0.140) (0.050) (0.048) (0.062)R&D/Sales �0.001 0.001 0.000 0.001

(0.001) (0.001) (0.000) (0.001)Dividend Payer Dummy �0.054nnn 0.002 0.030nnn 0.033nnn

(0.014) (0.005) (0.005) (0.006)CF Volatility �0.000 0.765nnn 0.311nnn 1.077nnn

(0.218) (0.133) (0.096) (0.157)Rated �0.076nnn �0.007 0.013nn 0.006

(0.017) (0.006) (0.006) (0.007)Beta KMV 0.029nnn 0.006nnn �0.005nnn 0.001

(0.003) (0.002) (0.001) (0.002)Constant 0.885nnn 0.245nnn 0.043nnn 0.287nnn

(0.027) (0.014) (0.009) (0.016)

Number of observations 4,705 4,689 4,707 4,689R-squared 0.390 0.405 0.177 0.197

Panel B: Sort by industry percentage of firms with at least 50% of debt in bonds

Variable

Crisis dummy �0.018nn �0.023nnn 0.002 �0.022nnn

(0.009) (0.005) (0.003) (0.006)Industry percentage 50% of bonds �0.054nnn �0.010n 0.013nnn 0.002

(0.014) (0.006) (0.005) (0.007)Crisis n Industry percentage 50% of bonds 0.025nn 0.015nn �0.011nn 0.005

(0.013) (0.006) (0.005) (0.008)Profitability �0.022 0.022 0.076nnn 0.097nnn

(0.040) (0.025) (0.015) (0.025)LnRevenues �0.026nnn �0.015nnn 0.003n �0.012nnn

(0.004) (0.002) (0.001) (0.002)M/B 0.029nnn 0.022nnn �0.002 0.020nnn

(0.004) (0.002) (0.001) (0.002)Tangibility �0.409nnn �0.206nnn 0.020n �0.186nnn

(0.038) (0.013) (0.012) (0.016)NWC/Asset �0.691nnn �0.211nnn 0.185nnn �0.026

(0.034) (0.015) (0.014) (0.018)Capex/Asset �0.029 �0.002 0.063 0.065

(0.138) (0.050) (0.048) (0.062)R&D/Sales �0.001 0.000 0.000 0.001

(0.001) (0.001) (0.000) (0.001)Dividend Payer Dummy �0.056nnn 0.001 0.030nnn 0.031nnn

(0.014) (0.005) (0.005) (0.006)CF Volatility 0.038 0.780nnn 0.312nnn 1.091nnn

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319 311

Table A2 (continued )

Cash/ Cash/ Undrawn credit Total liquidity(cash þ credit net assets lines/ /net assets

lines) net assets(1) (2) (3) (4)

(0.219) (0.133) (0.096) (0.157)Rated �0.082nnn �0.010n 0.012nn 0.003

(0.017) (0.006) (0.006) (0.007)Beta KMV 0.029nnn 0.006nnn �0.005nnn 0.001

(0.003) (0.002) (0.001) (0.002)Constant 0.874nnn 0.241nnn 0.043nnn 0.283nnn

(0.027) (0.014) (0.009) (0.016)

Number of observations 4,705 4,689 4,707 4,689R-squared 0.387 0.403 0.178 0.197

Panel C: Sort by industry percentage of firms that are rated

Variable

Crisis dummy �0.015n �0.029nnn 0.000 �0.029nnn

(0.008) (0.005) (0.003) (0.006)Industry percentage of rated �0.082nnn �0.020nnn 0.017nnn �0.003

(0.016) (0.006) (0.005) (0.007)Crisis n Industry percentage of rated 0.028nn 0.032nnn �0.009n 0.024nnn

(0.013) (0.006) (0.005) (0.007)Profitability �0.016 0.022 0.075nnn 0.096nnn

(0.038) (0.025) (0.014) (0.025)LnRevenues �0.024nnn �0.015nnn 0.002n �0.012nnn

(0.004) (0.002) (0.001) (0.002)M/B 0.027nnn 0.022nnn �0.002 0.021nnn

(0.004) (0.002) (0.001) (0.002)Tangibility �0.388nnn �0.203nnn 0.015 �0.188nnn

(0.039) (0.013) (0.012) (0.016)NWC/Asset �0.678nnn �0.210nnn 0.182nnn �0.028

(0.034) (0.015) (0.014) (0.018)Capex/Asset �0.039 �0.003 0.065 0.067

(0.139) (0.050) (0.048) (0.062)R&D/Sales �0.002n 0.000 0.000 0.001

(0.001) (0.001) (0.000) (0.001)Dividend Payer Dummy �0.056nnn 0.002 0.030nnn 0.032nnn

(0.014) (0.005) (0.005) (0.006)CF Volatility 0.128 0.789nnn 0.293nnn 1.081nnn

(0.217) (0.133) (0.096) (0.157)Rated �0.075nnn �0.009 0.011nn 0.002

(0.017) (0.006) (0.005) (0.007)Beta KMV 0.029nnn 0.006nnn �0.005nnn 0.001

(0.003) (0.002) (0.001) (0.002)Constant 0.862nnn 0.241nnn 0.046nnn 0.287nnn

(0.026) (0.014) (0.009) (0.016)

Number of observations 4,705 4,689 4,707 4,689R-squared 0.391 0.405 0.179 0.198

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319312

By assumption (4), there is a solution D0rρ0 that satisfiesthis equation. Given this, and because illiquidity transfor-mation is not an issue, the firm captures the NPV of theilliquid project, which is given by

UC ¼ ρ01�λ0ρ� Iþυ 0ðρτ�τÞ: □ ð29Þ

A.4. Proof of Proposition 3

In the case of the liquid project, the firm is given acredit line of size ρþτ�ρ0. This credit line is sufficient tomeet the firm's liquidity demand if both the liquidityshock ρ and the new investment opportunity τ arrive at

Date 1. If the firm reports a liquidity shock ρ, the bank hasincentives to monitor given that condition (11) holds.Monitoring triggers revocation with probability μ. Thisrevocation creates incentives for the firm to avoid illiquid-ity transformation, because condition (26) holds. Thissolution allows the firm to achieve a payoff equal to Un

L .The bank receives a payment equal to DM upon continua-tion, at Date 2, which ensures that the bank breaks even.

In case of the illiquid project, the firm borrows anamount equal to Iþρþτ�ρ0 at Date 0 and saves ρþτ�ρ0as cash. In Date 1, the firm uses the cash to finance theliquidity shock and the new investment opportunity, and itreturns excess cash to investors, who also receive thepayment D0 at Date 2. □

Table A3How does liquidity management vary over different states of the economy?

This table examines how liquidity management varies across different states of the economy, measured by bond yields, equity market returns, and grossdomestic product (GDP) growth. Bond yields are the average spreads on AAA and BAA Moody's rated bonds. Hot Bond Market is a dummy that takes one ifthe average bond spreads are one standard deviation below the median over the whole period. We compute GDP growth as the annual percentage changein GDP in the US. We compute equity returns as the annual percentage change on the Standard and Poor's 500 index. Industry mean cash flow volatilityand bonds percentage of assets are computed at the three-digit Standard Industry Classification level. All regressions include the following (unreported)firm characteristics: Size, M/B, Tangibility, NWC/Assets, Capex/Assets, R&D/Sales, Dividend Payer Dummy and Rating Dummy. We cluster errors at the firmlevel. nnn, nn, and n denote statistical significance at the 1%, 5%, and 10% level, respectively.

(1) (2) (3) (4)Cash/(cash þ Cash/net Undrawn Total liquiditycredit lines) assets credit lines/net /net assets

asset

Panel A: Bond spreads and average industry bonds to assets

Bond spreads 0.036nnn 0.025nnn �0.005nnn 0.019nnn

(0.003) (0.005) (0.001) (0.005)Industry mean bonds/assets �0.062nnn 0.021 0.001 0.023

(0.013) (0.020) (0.005) (0.020)Bond spreads n Industry mean bonds/assets 0.000 �0.007 0.000 �0.007

(0.004) (0.007) (0.002) (0.008)Number of observations 28,948 28,814 28,946 28,814R-squared 0.330 0.277 0.089 0.226

Panel B: Hot bond markets and average industry bonds to assets

Hot bond market �0.036nnn �0.040nnn 0.001 �0.038nnn

(0.003) (0.008) (0.002) (0.008)Industry mean bonds/assets �0.061nnn 0.005 0.002 0.006

(0.009) (0.012) (0.003) (0.012)Hot bond market n Industry mean bonds/assets �0.005 0.002 0.001 0.003

(0.005) (0.011) (0.002) (0.011)Number of observations 28,948 28,814 28,946 28,814R-squared 0.326 0.278 0.088 0.226

Panel C: Bond spreads and industry percentage of firms with at least 50% of debt in bonds

Bond spreads 0.032nnn 0.027nnn �0.005nnn 0.022nnn

(0.002) (0.006) (0.001) (0.006)Industry percentage bonds450 �0.099nnn 0.014 0.014nnn 0.027

(0.014) (0.016) (0.005) (0.017)Bond spreads n Industry percentage bonds450 0.012nnn �0.015nn �0.002 �0.016nn

(0.004) (0.007) (0.002) (0.007)Number of observations 28,948 28,854 28,946 28,814R-squared 0.332 0.278 0.090 0.226

Panel D: Hot bond markets and industry percentage of firms with at least 50% of debt in bonds

Hot bond market �0.030nnn �0.048nnn 0.001 �0.047nnn

(0.003) (0.009) (0.002) (0.009)Industry percentage bonds450 �0.070nnn �0.024nn 0.010nnn �0.014

(0.010) (0.009) (0.003) (0.010)Hot bond market n Industry percentage bonds450 �0.018nnn 0.023nn 0.001 0.025nn

(0.005) (0.010) (0.002) (0.010)Number of observations 28,948 28,854 28,946 28,814R-squared 0.329 0.278 0.089 0.226

Panel E: Bond spreads and industry percentage of firms that are rated

Bond spreads 0.030nnn 0.029nnn �0.004nnn 0.025nnn

(0.002) (0.006) (0.001) (0.006)Industry percentage of rated �0.124nnn 0.026 0.019nnn 0.046nnn

(0.014) (0.016) (0.005) (0.017)Bond spreads n Industry percentage of rated 0.015nnn �0.021nnn �0.002 �0.024nnn

(0.004) (0.006) (0.002) (0.007)Number of observations 28,948 28,854 28,946 28,814R-squared 0.336 0.278 0.091 0.226

Panel F: Hot bond markets and industry percentage of firms that are rated

Hot bond market �0.029nnn �0.051nnn 0.001 �0.050nnn

(0.003) (0.008) (0.002) (0.009)Industry percentage of rated �0.087nnn �0.028nnn 0.014nnn �0.014

(0.010) (0.009) (0.003) (0.009)

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319 313

Table A3 (continued )

(1) (2) (3) (4)Cash/(cash þ Cash/net Undrawn Total liquiditycredit lines) assets credit lines/net /net assets

asset

Hot bond market n Industry percentage of rated �0.021nnn 0.032nnn 0.002 0.034nnn

(0.005) (0.010) (0.003) (0.010)Number of observations 28,948 28,854 28,946 28,814R-squared 0.333 0.278 0.090 0.226

Panel G: GDP growth and average industry cash flow volatility

GDP growth �0.521nnn �0.947nnn 0.008 �0.942nnn

(0.059) (0.115) (0.027) (0.118)R-squared 28,272 28,180 28,271 28,142Number of observations 0.327 0.282 0.089 0.229

Panel H: Equity markets returns and average industry cash flow volatility

Equity returns �0.090nnn �0.005 0.028nnn 0.023nn

(0.006) (0.010) (0.003) (0.011)R- squared 28,272 28,180 28,271 28,142Number of observations 0.328 0.280 0.091 0.227

Table A4Was there a reversal of the cash accumulation after the General Motors (GM)–Ford event?

This table replicates the analysis of Table 2. The treatment group is constructed as in the main analysis. The crisis dummy is replaced by a dummy (Postdummy) for the twelve months following the GM–Ford event (July 2005–June 2006). We include observations only in Compustat fiscal years 2004 and2005 (June 2004–June 2006). nnn, nn, and n denote statistical significance at the 1%, 5%, and 10% level, respectively.

Cash/ Cash/ Undrawn credit Total liquidity(cash þ credit lines) net assets lines/net asset /net assets

(1) (2) (3) (4)

Panel A: Sort by industry average bonds to assets ratioPost dummy 0.004 0.006n 0.000 0.006

(0.007) (0.003) (0.003) (0.004)Treatment dummy �0.052nnn �0.012nn 0.004 �0.007

(0.014) (0.005) (0.005) (0.007)Treatment n Post �0.005 �0.002 0.001 �0.001

(0.010) (0.005) (0.004) (0.006)R-squared 0.307 0.335 0.105 0.159

Panel B: Sort by industry percentage of firms with at least 50% of debt in bondsPost dummy 0.009 0.008nn 0.000 0.008n

(0.007) (0.004) (0.003) (0.005)Treatment dummy �0.050nnn �0.010n 0.010nn 0.001

(0.014) (0.006) (0.005) (0.007)Treatment n Post �0.015 �0.006 0.001 �0.005

(0.010) (0.005) (0.004) (0.006)R-squared 0.308 0.335 0.107 0.158

Panel C: Sort by industry percentage of firms that are ratedPost dummy 0.011 0.010nnn �0.000 0.009nn

(0.007) (0.004) (0.003) (0.005)Treatment dummy �0.110nnn �0.016nnn 0.024nnn 0.009

(0.015) (0.005) (0.005) (0.007)Treatment n Post �0.020n �0.011nn 0.002 �0.009

(0.010) (0.005) (0.005) (0.006)R- squared 0.322 0.337 0.113 0.159Number of observations 4,802 4,768 4,804 4,768

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319314

Table A5External finance dependence.

This table examines the relationship between external finance dependence and liquidity management in relation to the General Motors (GM)–Ford crisisand the placebo crisis. Panel A is constructed as Table 2 (GM–Ford Crisis), and Panel B is constructed as Table 3 (placebo crisis).

Cash/ Cash/ Undrawn credit Total liquidity(cash þ credit lines) net assets lines/net assets /net assets

(1) (2) (3) (4)

Panel A: Sort by industry average external finance ratioCrisis dummy 0.022nn �0.001 �0.010nn �0.011nn

(0.011) (0.004) (0.004) (0.005)Treatment dummy 0.095nnn 0.020nnn �0.018nnn 0.001

(0.015) (0.005) (0.005) (0.007)Treatment n Crisis �0.043nnn �0.023nnn 0.010n �0.012

(0.013) (0.006) (0.005) (0.008)R-squared 0.393 0.405 0.180 0.197Number of observations 4,705 4,689 4,707 4,689

Panel B: Sort by industry average external finance ratioCrisis dummy �0.062nnn 0.010nn 0.013nnn 0.023nnn

(0.010) (0.004) (0.004) (0.005)Treatment dummy 0.057nnn 0.018nnn �0.012nn 0.007

(0.013) (0.006) (0.005) (0.007)Treatment n Crisis 0.013 �0.006 �0.007 �0.014n

(0.013) (0.006) (0.004) (0.007)R-squared 0.311 0.410 0.138 0.204Number of observations 4,775 4,759 4,777 4,759

Table A6Changes in the tightness of covenants.

This table replicates Table 2 using the index of covenant slackness as dependent variable. To construct the index, we collect the list of covenants and theassociated ratios on all the credit lines issued by the firms in our sample over the period December 2004–May 2005. We adjust all ratios by their standarddeviation and construct an index of covenant slackness based on the standardized ratios. A higher value of the index means less tight covenants. Theregressions are executed at the facility-firm-year level, rather than at the firm-year level, as each facility carries different covenants.

Index of covenant slackness

Panel A: Sort by industry average bonds to assets ratioCrisis dummy 0.120

(0.093)Treatment dummy 0.231nnn

(0.088)Treatment n Crisis �0.229n

(0.131)R-squared 0.105

Panel B: Sort by industry percentage of firms with at least 50% of debt in bondsCrisis dummy 0.184nn

(0.094)Treatment dummy 0.279nnn

(0.092)Treatment n Crisis �0.326nn

(0.128)R-squared 0.107

Panel C: Sort by industry percentage of firms that are ratedCrisis dummy �0.012

(0.101)Treatment dummy 0.139

(0.091)Treatment n Crisis �0.006

(0.130)R-squared 0.103Firm controls YESLoan type controls YESNumber of observations 1,495

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319 315

Table A7Credit line usage across top and bottom deciles and terciles of hedging needs.

This table provides two replicas of Panel A of Table 6, respectively, using top and bottom deciles (Panel A) and top and bottom terciles (Panel B).

Variable: Mean (median)

Presence of a credit Undrawn credit/ Undrawn credit/ Revolving credit/line (dummy) net assets total liquidity net assets

Panel A: DecilesIndustry median investmentactivities

High hedging needs 0.621 0.088 0.305 0.029N¼3,037 (1.000) (0.049) (0.198) (0.000)Low hedging needs 0.858 0.135 0.532 0.037N¼3,104 (1.000) (0.110) (0.567) (0.000)

T-test 21.930 15.433 27.042 4.172(P-value) (0.000) (0.000) (0.000) (0.000)

Industry median Tobin's QHigh hedging needs 0.765 0.124 0.465 0.039N¼2,944 (1.000) (0.091) (0.490) (0.000)Low hedging needs 0.839 0.140 0.538 0.048N¼3,058 (1.000) (0.118) (0.595) (0.000)

T-test 7.226 4.558 8.162 3.780(P-value) (0.000) (0.000) (0.000) (0.000)

Panel B: TercilesIndustry median Investmentactivities

High hedging needs 0.533 0.080 0.256 0.025N¼11,563 (1.000) (0.014) (0.049) (0.000)Low hedging needs 0.799 0.126 0.513 0.047N¼11,798 (1.000) (0.101) (0.568) (0.000)

T-test 44.987 27.938 57.774 21.629(P-value) (0.000) (0.000) (0.000) (0.000)

Industry median Tobin's QHigh hedging needs 0.625 0.097 0.316 0.029N¼12,701 (1.000) (0.051) (0.206) (0.000)Low hedging needs 0.691 0.111 0.426 0.046N¼12,362 (1.000) (0.080) (0.437) (0.000)

T-test 10.946 9.122 24.565 15.804(P-value) (0.000) (0.000) (0.000) (0.000)

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319316

A.5. Proof of Proposition 4

The choice between cash and credit lines is driven bythe difference between UC and Un

L . For firms of type L,

ðUC�Un

L ÞL ¼ ρ01�λ0Lρ�ρ1þλLρþðυ 0 �υÞðρτ�τÞ

þλL½cþμðρ1�ρþυLðρτ�τÞÞ�

¼ ρ01�ρ1�tρ�tðυH�υLÞðρτ�τÞ

þλL½cþμðρ1�ρþυLðρτ�τÞÞ� ð30Þ

Similarly, for firms of type H,

ðUC�Un

L ÞH ¼ ρ01�ρ1�tρ�tðυH�υLÞðρτ�τÞ

þλH ½cþμðρ1�ρþυLðρτ�τÞÞ�: ð31Þ

Thus, we have that

ðUC�Un

L ÞH�ðUC�Un

L ÞL ¼ ðλH�λLÞcþμ½ρ1�ρþυLðρτ�τÞ�40:

ð32Þ

Thus, the difference in payoffs is larger for firms of type Hwhich are then more likely to choose cash. □

A.6. Proof of Proposition 5

The difference in payoffs for the firm with high hedgingneeds can be written as

ðUC�Un

L ÞHHN ¼ Kþλυðρτ�τÞ; ð33Þwhere K ¼ ρ01�ρ1�ðλ0 �λÞρþλ½cþμðρ1�ρÞ� is a term thatdoes not depend on υ. Similarly, the difference in payoffsfor the firm with low hedging needs can be written as

ðUC�Un

L ÞLHN ¼ KþλυLðρτ�τÞþðυ 0 �υÞðρτ�τÞ: ð34ÞBecause υLoυ and υ 0 �υ40, it follows that ðUC�Un

L ÞLHNoðUC�Un

L ÞHHN . □

A.7. Model extension of Section 2.4.4

In this version of the model, illiquidity transformationcan be modeled as a shift in the distribution function bF ð�Þ,which also affects the expected project payoff cρ1 . Specifi-cally, we assume that the liquid project is represented by afunction Fð�Þ and an expected payoff ρ1, while the illiquidproject is represented by a function F 0ð�ÞrFð�Þ for all ρ and anexpected payoff ρ014ρ1. The former condition means thatthe probability that ρ is below any given value ρX is greaterunder the liquid project choice. That is, the illiquid project

Table A8Description of variables.

Variable Construction

Beta KMV Firm's asset (unlevered) beta, calculated from equity (levered) betas and a Merton-KMV formula as in Acharya,Almeida, and Campello (2012)

Book Leverage Total Debt/Total Assets (COMPUSTAT Item 6)BV Equity Total Assets (COMPUSTAT Item 6) � Total Liabilities (COMPUSTAT Item 181) � Deferred Taxes and Investment

Tax Credit (COMPUSTAT Item 35) � Preferred StockCapex/Assets Capital Expenditures (COMPUSTAT Item 128)/Total Assets (COMPUSTAT Item 6)Cash Flow/Assets [Operating Income Before Depreciation (COMPUSTAT Item 13) � Interest Expense (COMPUSTAT Item 15) �

Income Taxes (COMPUSTAT Item 16) � Dividends (COMPUSTAT Item 21)]/Total Assets (COMPUSTAT Item 6)Cash Ratio Cash and Short-Term Investments (COMPUSTAT Item 1)/[Cash and Short-Term Investments (COMPUSTAT Item

1) þ(Undrawn) Credit Lines (CapitalIQ)]Cash/Net Assets Cash and Short-Term Investments (COMPUSTAT Item 1)/Net AssetsCash/Net Assets (market value) Cash and Short-Term Investments (COMPUSTAT Item 1)/[Total Assets (COMPUSTAT Item 6) � BV Equity þ

Market Value of Equity � Cash and Short-Term Investments (COMPUSTAT Item 1)]Change in Working Capital Yearly Change in Working Capital (COMPUSTAT Item 179) � Yearly Change in Cash and Short-Term Investments

(COMPUSTAT Item 1)CF Volatility Standard Deviation of Operating Income Before Depreciation (13) over Previous 12 Quarters Scaled

by Total Assets (6)Covenant Index (Demiroglu and

James, 2010)Sum of the following covenants: Collateral Release, Dividend Restrictions, Dummy Financial Covenants, AssetSales Sweep, Equity Issuance Sweep and Debt Issuance SweepDummy Financial Covenants equals one when at least two of the following covenants are included: Debt/Tangible Assets, Max Capex, Max Debt/Assets, Max Debt/Ebitda, Max Debt/Equity, Max Leverage, Max SeniorDebt/Ebitda, Max Senior Leverage, Min Change Interest Coverage, Min Current Ratio, Min Debt Coverage, MinEbitda, Min Equity/Asset, Min Fixed Charge, Min Interest Coverage, Min Net Worth/Assets, Min Quick Ratio, NetWorth, Other Ratio, Other, Tangible Net Worth

Covenant Index(Drucker and Puri, 2010)

Sum of following covenants: % of Excess CF, % of Net Income, Asset Sales Sweep, Collateral Release, Debt IssuanceSweep, Dividend Restrictions, Equity Issuance Sweep, Excess CF Sweep, Insurance Proceeds Sweep, Max Capex,Max Debt/Assets, Max Debt/Ebitda, Max Debt/Equity, Max Debt/Tangible Assets, Max Leverage, Max SeniorDebt/Ebitda, Max Senior Leverage, Min Change Interest Coverage, Min Current Ratio, Min Debt Coverage, MinEbitda, Min Equity/Asset, Min Fixed Charge, Min Interest Coverage, Min Net Worth/Assets, Min Quick Ratio, NetWorth, Other, Other Ratio, Tangible Net Worth

Covenant Index(Only CFCovenants)

Sum of following covenants: % of Excess CF, % of Net Income, Excess CF Sweep, Max Capex, Max Debt/Ebitda,Max Senior Debt/Ebitda, Min Change Interest Coverage, Min Ebitda

Covenant Index (Only Sweeps) Sum of following covenants: Asset Sales Sweep, Debt Issuance Sweep, Equity Issuance Sweep, Excess CF Sweep,Insurance Proceeds Sweep

Credit Lines/Net Assets (Undrawn) Credit Lines (CapitalIQ)/Net AssetsCredit Lines/Net Assets

(market value)(Undrawn) Credit Lines (CapitalIQ)/[Total Assets (COMPUSTAT Item 6) � BV Equity þ Market Value of Equity �Cash and Short-Term Investments (COMPUSTAT Item 1)]

Dividend Payer Dummy A dummy variable that takes the value of one if common stock dividends (COMPUSTAT Item 21) are positive,and zero otherwise

External Finance Ratio Following Rajan and Zingales (1995) external financing as the fraction of total financing is the ratio of netexternal financing to the sum of cash flow from operations (net income plus depreciation) and net externalfinancing. Net external financing is the difference between net debt financing (debt issuances less debtreductions) and net equity financing (equity issuances less equity reductions taking minus dividends)

Full (Partial) Revocation of CreditLine (Dummy)

Take value of one in period t if a firm has available undrawn credit in T � 1 and no available undrawn credit in T,and drawn credit has not increased between T � 1 and T. Partial revocations are those in which undrawn creditavailability decreases by 30% or more

Hedging based on investmentActivities

Correlation between three-digit annual mean industry investment activities adjusted for research anddevelopment (R&D) expenses (COMPUSTAT Item 311 þ COMPUSTAT Item 46) and the annual mean industrycash flows measured as in Acharya, Almeida, and Campello (2007). Three-digit mean industry investmentactivities are computed on the sample of unconstrained firms, defined as firms that pay dividends, have assetsabove 500 m and rating above Bþ

Hedging based on Tobin's Q Correlation between three-digit median industry market-to-book and the firm-year cash flows measured as inAcharya, Almeida, and Campello (2007). The three-digit mean industry market-to-book is computed on thesample of unconstrained firms, defined as firms that pay dividends, have assets above $500 million and ratingabove Bþ

Industry Mean Cash Ratio Mean cash ratio at the two digit SIC codeM/B (Market Value of Equity þ Total Debt þ Preferred Stock Liquidating Value (COMPUSTAT Item 10) � Deferred

Taxes and Investment Tax Credit (COMPUSTAT Item 35))/Total Assets (COMPUSTAT Item 6)Market Value of Equity Stock Price (COMPUSTAT Item 199)�Common Shares Used to Calculate earnings per share (EPS) (COMPUSTAT

Item 54)Net Assets Assets (COMPUSTAT Item 6) � Cash and Short-Term Investments (COMPUSTAT Item 1))NWC/Assets (Working Capital (COMPUSTAT Item 179) � Cash and Short-Term Investments(COMPUSTAT Item 1))/Total

Assets (COMPUSTAT Item 6)Preferred Stock Max[Preferred Stock Liquidating Value (COMPUSTAT Item 10), Preferred Stock Redemption Value (COMPUSTAT

Item 56), Preferred Stock Carrying Value (COMPUSTAT Item 130)]Profitability Operating Income Before Depreciation (13)/Total Assets (6)R&D/Sales Research and Development Expenses (46)/Sales (12)Rated A dummy variable that takes the value of one if the firm is rated by the Standard and Poor, and zero otherwiseRating

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319 317

Table A8 (continued )

Variable Construction

Monthly S&P ratings (280). Takes 23 values: 1¼AAA, 2¼AAþ , 3¼AA, 4¼AA� , 5¼Aþ , 6¼A, 7¼A� , 8¼BBBþ ,9¼BBB, 10¼BBB� , 11¼BBþ , 12¼BB, 13¼BB� , 14¼Bþ , 15¼B, 16¼B� , 17¼CCCþ , 18¼CCC, 19¼CCC� ,20¼CC, 21¼SD, 22¼D and 23¼Unrated

Revolving Credit/Net Assets Revolving Credit (CapitalIQ)/Net AssetsSize Logarithm of Revenues (COMPUSTAT Item 12)Tangibility Net Property, Plant, and Equipment (COMPUSTAT Item 8)/Total Assets (COMPUSTAT Item 6)Total Debt Debt in Current Liabilities (COMPUSTAT Item 34) þ Long-Term Debt (COMPUSTAT Item 9)

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319318

shifts mass toward high levels of the liquidity shock. As inthe model above, we assume that illiquidity transformationdoes not affect Date 1 pledgeable income ρ0.

In this case, the optimal liquidity policy is to establish acutoff ρn below which the firm is allowed to continue theproject in Date 1 (see Tirole, 2006, pp. 201–204). Thecutoff ρn is always greater than or equal to the pledgeableincome ρ0, for ρoρ0 continuation benefits both outsideinvestors and the firm. To be able to pay for liquidityshocks above ρ0, the firm must have access to pre-committed liquidity sources such as cash or credit lines.Increasing the cutoff up to ρn ¼cρ1 maximizes the NPV ofthe project, but it might not be feasible. In what follows weassume that the firm can finance the NPV-maximizingcutoffs for both the liquid and the illiquid project:

IþZ ρ01

0ρf 0ðρÞ dρrF 0ðρ01Þρ0: ð35Þ

This condition ensures that the firm can fund the NPV-maximizing cutoff if it chooses the liquid project as well(in this case the value-maximizing cutoff is ρn ¼ ρ1). Theliquid project has greater Date 0 expected pledgeableincome because ρ014ρ1 and Fð�ÞZF 0ð�Þ for all ρ and becauseDate 1 pledgeable income (ρ0) is the same for bothprojects. In other words, both projects produce the samepledgeable income contingent on continuation (ρ0), butthe illiquid project requires liquidity infusions with greaterprobability at Date 1.

The monitoring technology is identical to the modelabove. For this version of the model we have

Probðs¼ s0=bF ¼ FÞ ¼ μo1;

Probðs¼ s0=bF ¼ F 0Þ ¼ 1: ð36ÞAs in the model above, the bank monitors after the firmreports a given liquidity shock ρ and attempts to draw onthe credit line if ρ4ρ0.

We also make the following assumptions:

F 0ðρ1Þðρ01�ρ0Þ4Fðρ1Þðρ1�ρ0Þ; ð37Þ

μðρ1�ρ0ÞZc; ð38Þand

F ρ0þcμ

� �þ 1�μð Þ F ρ1

� ��F ρ0þcμ

� �� �� �ρ1�ρ0� �

4F 0 ρ0þcμ

� �ρ01�ρ0� �

: ð39Þ

The first assumption [condition (37)] creates a role formonitoring. To see this, suppose that the firm wishes to

implement the liquid project. Because ρn ¼ ρ1 is feasible,the firm would open a credit line of size ρ1�ρ0 and use itto fund liquidity shocks above ρ0 and up to ρ1. Conditionalon having started the project and secured a fully com-mitted credit line, the firm's payoff is then Fðρ1Þðρ1�ρ0Þ,while the payoff from deviating to the illiquid project isF 0ðρ1Þðρ01�ρ0Þ. Thus, as long as condition (37) holds, thefirm cannot access fully committed liquidity insurance.The firm will be liquidated for ρ4ρ1 even when it deviatesto the illiquid project, because its credit line is of sizeρ1�ρ0.

The second assumption [condition (38)] assures thatthe bank has incentives to monitor. In particular, the bankwill monitor if and only if μðρ�ρ0Þ4c. Thus, there is aminimum level of the liquidity shock ρ that induces bankmonitoring. This is such that

μðρmin�ρ0Þ ¼ c: ð40ÞAssumption (38) assures that ρminrρ1. Firm and bankstrategies are as follows (as a function of ρ): If ρoρ0, thefirm can continue the project without drawing on thecredit line. If ρ0rρrρmin, the firm draws on the credit lineand the bank does not monitor. If ρminoρrρ1, the bankmonitors in equilibrium and could revoke the credit line.Finally, if ρ1oρ, the firm has no incentive to draw onthe credit line because it is not sufficient to financecontinuation.

The third assumption [condition (39)] assures that bankmonitoring is sufficient to induce the firm not to deviate tothe illiquid project once the credit line is in place (see theproof to Proposition 6). The firm's ex ante payoff under themonitored credit line is

ULC ¼ Fðρ1Þρ1� I�Z ρ1

0ρf ðρÞ dρ�

Z ρ1

ρmin

½cþμðρ1�ρÞ�f ðρÞ dρ:

ð41ÞThe expression Fðρ1Þρ1� I� R ρ10 ρf ðρÞ dρ represents thefirm's liquid project payoff under fully committed liquidityinsurance. The term

R ρ1ρmin

½cþμðρ1�ρÞ�f ðρÞ dρ represents themonitoring cost. If ρminoρoρ1, the bank monitors (at costc) and the credit line is revoked with probability μ,implying a value loss ρ1�ρ for each ρ.

The firm's alternative is to use cash and to invest in theilliquid project. In this case, the payoff is

UC ¼ F 0ðρ01Þρ01� I�Z ρ01

0ρf 0ðρÞ dρ ð42Þ

Condition (35) ensures that both payoffs are achievable.As in the previous model, the firm chooses cash when

V. Acharya et al. / Journal of Financial Economics 112 (2014) 287–319 319

UCZULC and the credit line otherwise. The main result ofthis extension is as follows:

Proposition 6. Under assumptions (35), (37)–(39), a decreasein pledgeable income ρ0 makes it more likely that the firmwill choose cash instead of the credit line. That is, ULC�UC isincreasing in ρ0.

Given the bank's monitoring strategy, the firm's payoffconditional on having started a project and secured acredit line of size ρ1�ρ0 is

½FðρminÞþð1�μÞðFðρ1Þ�FðρminÞÞ�ðρ1�ρ0Þ ð43ÞIf ρoρmin, the firm can continue the project with

probability one because the bank does not monitor.If ρminoρoρ1, then the bank monitors and revokes thecredit line with probability μ.

If the firm deviates to the illiquid project, the payoff is

F 0ðρminÞðρ01�ρ0Þ: ð44ÞIn this case the bank revokes the credit line with prob-ability one if ρ4ρmin.

As long as ½FðρminÞþð1�μÞðFðρ1Þ�FðρminÞÞ�ðρ1�ρ0Þ4F 0ðρminÞðρ01�ρ0Þ, the firm can implement the liquid projectusing a monitored credit line.

The monitored credit line is feasible if

IþZ ρ1

0ρf ðρÞ dρrFðρ1Þρ0þ

Z ρ1

ρmin

½μðρ�ρ0Þ�c�f ðρÞ dρ ð45Þ

This condition is implied by Eq. (35) above, because theexpression

R ρ1ρmin

½μðρ�ρ0Þ�c�f ðρÞ dρ is positive and because

ρ014ρ1 and Fð�ÞZF 0ð�Þ for all ρ. Thus, the payoff under thecredit line is given by Eq. (41).

The feasibility condition for the cash solution is

Iþ R ρ010 ρf 0ðρÞ dρrF 0ðρ01Þρ0, which is exactly condition (35).In this case the firm holds an amount of cash equal toρ01�ρ0 and uses it to fund liquidity shocks up to ρ01 atDate 1. Thus, the payoff is given by Eq. (42).

To derive the comparative statics, notice that theexpression UC does not depend on ρ0. In contrast, ULC isincreasing in ρmin, which is increasing in ρ0 by Eq. (40).Thus, the difference UC�ULC decreases in ρ0.

References

Acharya, V., Almeida, H., Campello, M., 2007. Is cash negative debt?A hedging perspective on corporate financial policies. J. Financ.Intermed. 16, 515–554.

Acharya, V., Almeida, H., Campello, M., 2013. Aggregate risk and thechoice between cash and credit lines. J. Finance 68, 2059–2116, http://dx.doi.org/10.1111/jofi.12056.

Acharya, V., Schaefer, S., Zhang, Y., 2008. Liquidity risk and correlationrisk: a clinical study of the General Motors and Ford downgrade ofMay 2005. Unpublished working paper.

Arnott, R., Stiglitz, J., 1988. The basic analytics of moral hazard. Scand.J. Econ. 90, 383–413.

Barakova, I., Parthasarathy, H., 2012. How committed are bank corporateline commitments? Unpublished working paper. Office of the Comp-troller of the Currency.

Bates, T., Kahle, K., Stulz, R., 2009. Why do U.S. firms hold so much morecash than they used to? J. Finance 64, 1985–2021.

Berrospide, J., Meisenzahl, R., Sullivan, B., 2012. Credit line use andavailability in the financial crisis: the importance of hedging. Unpub-lished working paper. Federal Reserve Board.

Bradley, M., Roberts, M., 2004. The structure and pricing of corporate debtcovenants. Unpublished working paper. Duke University.

Campello, M., Giambona, E., Graham, J., Harvey, C., 2010. Liquiditymanagement and corporate investment during a financial crisis.Unpublished working paper. University of Illinois, University ofAmsterdam, Duke University.

Demiroglu, C., James, C., 2010. The information content of bank loancovenants. Rev. Financ. Stud. 23, 3700–3737.

Disatnik, D., Duchin, R., Schmidt, B., 2010. Cash flow hedging and liquiditychoices. Unpublished working paper. Tel Aviv University, MichiganUniversity, Emory University.

Drucker, S., Puri, M., 2009. On loan sales, loan contracting, and lendingrelations. Rev. Financ. Stud. 22, 2635–2672.

Duchin, R., 2010. Cash holdings and corporate diversification. J. Finance65, 955–992.

Duchin, R., Ozbas, O., Sensoy, B., 2010. Costly external finance, corporateinvestment, and the subprime mortgage credit crisis. J. Financ. Econ.97, 418–435.

Edmans, A., Goldstein, I., Jiang, W., 2012. The real effects of financialmarkets: the impact of prices on takeovers. J. Finance 67, 933–971.

Fama, E., 1985. What's different about banks? J. Monet. Econ. 15, 29–39.Gatev, E., Strahan, P., 2006. Banks' advantage in hedging liquidity risk:

theory and evidence from the commercial paper market. J. Finance61, 867–892.

Gatev, E., Schuermann, T., Strahan, P., 2009. Managing bank liquidity risk:how deposit-loan synergies vary with market conditions. Review ofFinancial Studies 22, 995–1020.

Holmstrom, B., Tirole, J., 1997. Financial intermediation, loanable funds,and the real sector. Q. J. Econ. 112, 663–691.

Holmstrom, B., Tirole, J., 1998. Private and public supply of liquidity.J. Polit. Econ. 106, 1–40.

Houston, J., James, C., 1996. Bank information monopolies and thedeterminants of the mix of private and public debt claims. J. Finance51, 1863–1889.

Jensen, M., Meckling, W., 1976. Theory of the firm: managerial behavior,agency costs and ownership structure. J. Financ. Econ. 3, 305–360.

Kashyap, A., Rajan, R., Stein, J., 2002. Banks as liquidity providers: anexplanation for the co-existence of lending and deposit-taking.J. Finance 57, 33–73.

Lemmon, M., Roberts, M., Zender, J., 2008. Back to the beginning:persistence and the cross-section of corporate capital structure.J. Finance 63, 1575–1608.

Lins, K., Servaes, H., Tufano, P., 2010. What drives corporate liquidity? Aninternational survey of strategic cash and lines of credit. J. Financ.Econ. 98 (1), 160–176, http://dx.doi.org/10.1016/j.jfineco.2010.04.006.

Nini, G., Smith, D., Sufi, A., 2009. Creditor control rights and firminvestment policy. J. Financ. Econ. 92, 400–420.

Rajan, Raghuram G., Zingales, Luigi, 1995. What Do We Know aboutCapital Structure? Some Evidence from International Data. J. Finance,American Finance Association 50 (5), 1421–1460.

Rajan, R., Zingales, L., 1998. Financial dependence and growth. Am. Econ.Rev. 88, 559–573.

Rothschild, M., Stiglitz, J., 1976. Equilibrium in competitive insurancemarkets: an essay on the economics of imperfect information. Q. J.Econ. 90, 626–649.

Shockley, R., Thakor, A., 1997. Bank loan commitment contracts: data,theory and tests. J. Money Credit Bank. 29, 517–534.

Sufi, A., 2009. Bank lines of credit in corporate finance: an empiricalanalysis. Rev. Financ. Stud. 22, 1057–1088.

Tirole, J., 2006. The Theory of Corporate Finance. Princeton UniversityPress, Princeton.

Yun, H., 2009. The choice of corporate liquidity and corporate govern-ance. Rev. Financ. Stud. 22 (4), 1447–1475, http://dx.doi.org/10.1093/rfs/hhn041.