The Cash Flow Sensitivity of Cash

The Cash Flow Sensitivity of CashHeitor Almeida, Murillo Campello, and Michael S. Weisbach*Almeida is at New York University. Campello and Weisbach are at the University of Illinois. Wethank an anonymous referee, Rick Green (the editor), Viral Acharya, Dan Bernhardt, Matt Bil-lett, Long Chen, Ted Fee, Jon Garnkel, John Graham, Charlie Hadlock, Jarrad Harford, HarrisonHong, George Pennacchi, Eric Rasmussen, Ren Stulz, Toni Whited, and Jerey Wurgler for theirvery helpful suggestions. Comments from seminar participants at the November 2002 NBER Cor-porate Finance Meeting, Duke University, Georgia State University, Indiana University, LouisianaState University, New York University, New York Federal Reserve Bank, University of Illinois, andUniversity of Iowa are also appreciated.abstractWe model a rms demand for liquidity to develop a new test of the eect of nancial constraintson corporate policies. The eect of nancial constraints is captured by the rms propensity to savecash out of cash ows (the cash ow sensitivity of cash). We hypothesize that constrained rmsshould have a positive cash ow sensitivity of cash, while unconstrained rms cash savings shouldnot be systematically related to cash ows. We empirically estimate the cash ow sensitivity ofcash using a large sample of manufacturing rms over the 1971 to 2000 period and nd robustsupport for our theory.Two important areas of research in corporate nance are the eects of nancial constraints onrm behavior and the manner in which rms perform nancial management. These two issues,although often studied separately, are fundamentally linked. As originally proposed by Keynes(1936), a major advantage of a liquid balance sheet is that it allows rms to undertake valuableprojects when they arise. However, Keynes also argued that the importance of balance sheetliquidity is inuenced by the extent to which rms have access to external capital markets (p. 196).If a rm has unrestricted access to external capital that is, if a rm is nancially unconstrained there is no need to safeguard against future investment needs and corporate liquidity becomesirrelevant. In contrast, when the rm faces nancing frictions, liquidity management may becomea key issue for corporate policy.Despite the link between nancial constraints and corporate liquidity demand, the literaturethat examines the eects of nancial constraints on rm behavior has traditionally focused on corpo-rate investment demand.1 In an inuential paper, Fazzari, Hubbard, and Petersen (1988) proposethat when rms face nancing constraints, investment spending will vary with the availability ofinternal funds, rather than just with the availability of positive net present value (NPV) projects.Accordingly, one should be able to examine the inuence of nancing frictions on corporate in-vestment by comparing the empirical sensitivity of investment to cash ow across groups of rmssorted according to a proxy for nancial constraints. Recent research, however, has identied sev-eral problems with that strategy. The robustness of the implications proposed by Fazzari, Hubbard,and Petersen has been challenged on theoretical grounds by Kaplan and Zingales (1997), Povel andRaith (2001), and Almeida and Campello (2002), while the robustness of cross-sectional patternspresented in their empirical work (and in the subsequent literature) has been questioned by Kaplanand Zingales, Cleary (1999), and Erickson and Whited (2000). Alti (2003) further demonstratesthat because cash ows contain valuable information about a rms investment opportunities, thecross-sectional patterns reported by Fazzari, Hubbard, and Petersen can be consistent with a modelwith no nancing frictions (see also Gomes (2001). This argument casts doubt on the very meaningof the empirical cash ow sensitivities of investment reported in the literature.1In this paper, we argue that the link between nancial constraints and a rms demand forliquidity can help us identify whether nancial constraints are an important determinant of rmbehavior. We rst present a model of a rms liquidity demand that formalizes Keynes intuition.In it, rms anticipating nancing constraints in the future respond to those potential constraints byhoarding cash today. Holding cash is costly, nonetheless, since higher cash savings require reductionsin current, valuable investments. Constrained rms thus choose their optimal cash policy to balancethe protability of current and future investments. This policy is in contrast to that of rms thatare able to fund all of their positive NPV investments: Financially unconstrained rms have no usefor cash, but also face no cost of holding cash (i.e., their cash policies are indeterminate).The stark dierence in the implied cash policies of constrained and unconstrained rms allowsus to formulate an empirical prediction about the eect of nancial constraints on rms nancialpolicies. Our model suggests that nancial constraints should be related to a rms propensity tosave cash out of cash inows, which we refer to as the cash ow sensitivity of cash. In particular,nancially unconstrained rms should not display a systematic propensity to save cash, whilerms that are constrained should have a positive cash ow sensitivity of cash. As such, the cashow sensitivity of cash provides a theoretically justied, empirically implementable measure of theimportance of nancial constraints.The use of cash ow sensitivities of cash to test for nancial constraints avoids some of theproblems associated with the investmentcash ow literature. In particular, because cash is anancial (as opposed to a real) variable, it is dicult to argue that the explanatory power ofcash ows over cash policies could be ascribed to its ability to forecast future business conditions(investment demand). For unconstrained rms, changes in cash holdings should depend neither oncurrent cash ows nor on future investment opportunities, so in the absence of nancial constraints,one should expect no systematic patterns in cash policies. Evidence that the sensitivity of cashholdings to cash ow varies systematically with proxies for nancing frictions is therefore morepowerful and less ambiguous evidence of the role of nancial constraints than what investmentcash ow sensitivities can provide.2We evaluate the extent to which the cash ow sensitivity of cash provides an empirically usefulmeasure of nancial constraints using a large sample of manufacturing rms between 1971 and 2000.We estimate that sensitivity for various rm subsamples, partitioned on the basis of the likelihoodthat rms have constrained access to external capital. In doing so, we use ve alternative approachessuggested by the literature to partition the sample into unconstrained and constrained subsamples:Payout policy, asset size, bond ratings, commercial paper ratings, and an index measure derivedfrom results in Kaplan and Zingales (1997) (the “KZ index”). We nd that under each of the rstfour classication schemes, the cash ow sensitivity of cash is close to and not statistically dierentfrom zero for the unconstrained rms, but positive and signicantly dierent from zero for theconstrained rms. The KZ index generates constrained/unconstrained rm assignments that aremostly negatively correlated with those of the other four classication criteria. Not surprisingly, weobtain the very opposite results for our estimates of the cash ow sensitivity of cash that use the KZindex. All of the patterns we observe in our basic tests remain after we subject our estimations tovarious robustness checks involving changes in empirical specications, sampling restrictions, andeconometric methodologies. Our ndings are fully consistent with the implications of our model ofcorporate liquidity.We further test the intuition of our argument by investigating rms propensity to save cash outof cash inows over the business cycle. Our model implies changes in corporate liquidity demandover the business cycle, because aggregate demand uctuations work as exogenous shocks aectingboth the size of current cash ows as well as the relative attractiveness of current investments vis-vis future ones. In a recession, nancially constrained rms should save a greater proportion oftheir cash ows, while unconstrained rms cash policies should not show any systematic changes.We nd that for constrained rms, cashcash ow sensitivities appear to be negatively associatedwith shocks to aggregate demand (i.e., on the margin, constrained rms save more in recessions),while unconstrained rms display no change in their cashcash ow sensitivities in response tomacroeconomic shocks. Once again, these results hold for four of our proxies for nancial constraints(payout policy, size, bond ratings, and commercial paper ratings), but not for the KZ index. The3macro-level tests provide additional support for our argument for two dierent reasons. Theyconrm that a natural extension of our model is consistent with the data, and they help sidestepthe usual concerns with estimation biases that arise in standard regression analysis involving rm-level data.While the literature has examined the eects of nancial constraints on corporate policies such asxed investment (e.g., Fazzari, Hubbard, and Petersen (1988) and Almeida and Campello (2003),working capital (Fazzari and Petersen (1993) and Calomiris, Himmelberg, and Wachtel (1995),and inventory demand (Carpenter, Fazzari, and Petersen (1994) and Kashyap, Lamont, and Stein(1994), it has not explicitly considered the relationship between nancial constraints and a rmsliquidity demand.2 A number of recent empirical studies do examine the cross-section of cash re-serves and the factors that appear to be associated with higher holdings of cash.3 Among otherresults, these papers nd that the levels of cash tend to be positively associated with future invest-ment opportunities, business risk, and negatively associated with proxies for the level of protectionof outside investors. However, while these studies focus on dierences in the level of cash acrossrms, our paper examines dierences in the sensitivity of cash holdings to cash ow and the extentto which they are aected by nancial constraints. We do so because our theory has much clearerpredictions about rms marginal propensity to save/disburse funds out of cash ow innovationsthan about the amount of cash in their balance sheets. To our knowledge, our paper is the rst topursue this approach in dealing with the issue of corporate liquidity.The remainder of the paper proceeds as follows. Section I introduces a theory of corporateliquidity demand and derives our main empirical implications. Section II presents the empiricaltests of these implications. Section III concludes.I. A Model of Liquidity DemandThe rst step of our analysis is to model corporate demand for liquid assets as a means ofensuring the rms ability to invest in an imperfect capital market. Our basic model is a simplerepresentation of a dynamic problem in which the rm has both present and future investment4opportunities, and in which cash ows from assets in place might not be sucient to fund allpositive NPV projects. Depending on the rms capacity for external nance, hoarding cash mayfacilitate future investments. Another way the rm can plan for the funding of future investmentsis by hedging against future earnings. In all, our framework considers four components of nancialpolicy: Cash management, hedging, dividend payouts, and borrowing.A. StructureThe model has three dates, 0, 1, and 2. At time 0, the rm is an ongoing concern whose cashow from current operations is c0.4 At that date, the rm has the option to invest in a long-termproject that requires I0 today and pays o F (I0) at time 2. Additionally, the rm expects to haveaccess to another investment opportunity at time 1. If the rm invests I1 at time 1, the technologyproduces G(I1) at time 2. The production functions F () and G() have standard properties; i.e.,they are increasing, concave, and continuously dierentiable. The rm has existing assets thatproduce a cash ow equal to c1 at time 1. With probability p, the time 1 cash ow is high, equal tocH1 , and with probability (1 p), equal to cL1 0. Dene total cash ows from investments as f (I0) F (I0) + qI0, andg(I1) G(I1) + qI1.We suppose that the cash ows F (I0) and G(I1) are not veriable and thus cannot be contractedupon. While the rm cannot pledge those cash ows to outside investors, it can raise external nanceby pledging the underlying productive assets as collateral. Following Hart and Moore (1994), theidea is that the liquidation value of “hard” assets is veriable by a court, and if the rm reneges onits debt, creditors will seize those assets. We assume that the liquidation value of assets that can becaptured by creditors is given by (1 )qI. The parameter (0, 1) is a function of factors suchas the tangibility of a rms assets and of the legal environment that dictates relations betweendebtors and creditors (see Myers and Rajan (1998). For a high enough , the rm may pass up5positive NPV projects for lack of external nancing and may thus become nancially constrained.In our setup, the rm is concerned only about whether or not to store cash from time 0 untiltime 1; there are no new investment opportunities to fund at time 2. We denote by C the amountof cash the rm chooses to carry from time 0 until time 1. We also assume that the rm can fullyhedge future earnings at a fair cost. As argued by Froot, Scharfstein, and Stein (1993), a bindingnancial constraint creates a motive for hedging future cash ows.6Two comments are in order before we analyze demand for liquidity in our proposed setup.First, we note that although the optimal contract in our Hart-Moore-type framework is mosteasily interpreted as collateralized debt, none of our conclusions hinge on this strict interpretation.The crucial feature for our theory is that some rms have limitations in their capacity to raiseexternal nance, and that such limitations may cause those rms to invest below rst-best levels.In particular, the models intuition is unchanged if we allow for uncollateralized debt or equity issues,so long as constrained rms have to pay a premium over and above the fair cost of uncollateralizeddebt and/or equity, or so long as there is a maximum amount of funds that constrained rms canraise using those instruments in the capital markets.7 Second, notice that the model does notimply a one-to-one correspondence between asset liquidity ( ) and nancial constraints. Whilesome degree of imperfection is necessary for a rm to be constrained i.e., we need some degree ofilliquidity, or a high enough whether or not a particular rm is constrained also depends on thesize of cash ows from existing assets relative to the magnitude of capital expenditures associatedwith the new investment opportunities.B. AnalysisThe rms objective is to maximize the expected lifetime sum of all dividends subject to variousbudget and nancial constraints. This problem can be written as: maxd0 + pdH1 + (1 p)dL1 + pdH2 + (1 p)dL2 s.t.C,h,I6(1)d0 = c0 + B0 I0 C 0dS1 = cS1 + hS + B1S I 1S + C 0, for S = H, LdS2 = f (I0) + g(I 1S) B0 B1S, for S = H, LB0 (1 )qI0B1S (1 )qI1S, for S = H, LphH + (1 p)hL = 0.The rst two constraints restrict dividends (d) to be non-negative in times 0 and 1. The terms B0and B1 are the borrowing amounts, which have to be lower than the collateral value generated bythe new investments. Debt obligations are repaid at the time when the assets they help nancegenerate cash ows. Hedging payments in states H and L are denoted by hH and hL, respectively.The hedging strategies we focus on typically give hH 0. If the rm uses futurescontracts, for example, we should think of cS1 + hS as the futures payo in state S. The rm sellsfutures at a price equal to the expected future spot value, and thus increases cash ows in stateL at the expense of reducing cash ows in state H. Finally, note that the fair hedging constraintdenes hH as a function of hL (hH = (1p p) hL).B.1. First-Best SolutionThe rm is nancially unconstrained if it is able to invest at the rst-best levels at times 0 and1, which are dened asf 0(I 0F B) = 1g0(I 1F B,S) = 1, for S = H, L.Since the productivity of investment does not vary across states, we have I 1F B,H = I 1F B,L I 1F B.When the rm is unconstrained, its investment policy satises all the dividend, hedging, andborrowing constraints above for some nancial policy (B0, B1S, C, hH ). More explicitly, the condition7for the rm to be unconstrained is that there exists a nancial policy (B0, B1S, C, hH ) such thatI 0F B c0 + B0 C (2)I 1F B cH1 1 p L Hh + B +