Rising Intangible Capital, Shrinking Debt Capacity, and the US Corporate Savings Glut

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. Rising Intangible Capital, Shrinking Debt Capacity, and the US Corporate Savings Glut Antonio Falato, Dalida Kadyrzhanova, and Jae W. Sim 2013-67 NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Rising Intangible Capital, Shrinking Debt Capacity, and the US Corporate Savings Glut AntonioFalato DalidaKadyrzhanova JaeW.Sim (cid:3) FederalReserveBoard UniversityofMaryland FederalReserveBoard November2012. Thisversion: September2013 Abstract Thispaperexploresthehypothesisthattheriseinintangiblecapitalisafundamentaldriverof theseculartrendinUScorporatecashholdingsoverthelastdecades. Usinganewmeasure, weshowthatintangiblecapitalisthemostimportantfirm-leveldeterminantofcorporatecash holdings. Our measure accounts for almost as much of the secular increase in cash since the 1980sasallotherdeterminantstogether. Wethendevelopanewdynamicdynamicmodelof corporate cash holdings with two types of productive assets, tangible and intangible capital. Since only tangible capital can be pledged as collateral, a shift toward greater reliance on intangiblecapitalshrinksthedebtcapacityoffirmsandleadsthemtooptimallyholdmorecash inordertopreservefinancialflexibility. Inthemodel,firmswithgrowthoptionstendtohold more cash in anticipation of (S,s)-type adjustments in physical capital because they want to avoid raising costly external finance. We show that this mechanism is quantitatively important, as our model generates cash holdings that are up to an order of magnitude higher than the standard benchmark and in line with their empirical averages for the last two decades. Overall, our results suggest that technological change has contributed significantly to recent changesincorporateliquiditymanagement. JELClassification: E22,E44,G31,G32 Keywords: Asset Intangibility; Debt Capacity; Risk Management; Corporate Cash Hoarding; (S,s)Adjustment (cid:3)Forhelpfulcommentsandsuggestions, wethankHengjieAi, GianLucaClementi, MaxCroce, AndreaEisfeldt, Mike Faulkender, Xavier Gabaix, Vito Gala, Joao Gomes, John Graham, Cam Harvey, Urban Jermann, Kose John, MichaelKiley,DavidLebow,AlbertoManconi,PaigeOiumet,MitchellPetersen,ManjuPuri,AdrianoRampini,David Robinson,DanSichel,PerStromberg,LukeTaylor,VishViswanathan,ToniWhited,andseminarparticipantsatDuke University,GeorgeMasonUniversity,StockholmSchoolofEconomics,UniversityofMaryland,UniversityofMassachusetts-Amherst, UniversityofBritishColumbia, UniversityofNorthCarolina, CFEA2012Meeting, FinancialIntermediationandResearchSociety2013Meeting,WesternFinanceAssociation2013meeting,NBERSummerInstitute 2013meeting,CEPR2013SummerSymposiuminFinancialMarkets,MinnesotaWorkshoponMacroeconomicTheory 2013meeting,andSocietyforEconomicDynamics2013meeting. TheBestPaperonCorporateFinanceAwardfrom WFA-CharlesRiverAssociatesisgratefullyacknowledged. Allremainingerrorsareours. Theviewsexpressedinthis paperarethoseoftheauthorsanddonotnecessarilyreflecttheviewsoftheFederalReserveBoardofGovernorsorthe FederalReserveSystem.Correspondingauthor:JaeW.Sim,Phone:(202)452-2680.Email:jae.w.sim@frb.gov. 1

1 Introduction Public corporations in the US have steadily increased their cash holdings over the last decades. This dramatic trend in corporate liquidity management is an actively debated issue that has attracted wide attention in the popular press, with commentators dubbing it the "corporate saving glut," expressing concerns it might hamper growth of the US economy, and even raising calls to heavily tax corporate savings. Yet, understanding which fundamental economic determinants drive the secular trend in corporate cash holdings and why corporations now hold almost three times as much cash as they used to in the 1970s1 represents a big outstanding challenge for both empiricalandtheoreticalresearchincorporatefinance. Inparticular,ontheempiricalside,existingevidenceonthedeterminantsoftheseculartrend in corporate cash holdings is at best mixed. Several explanations have been put forth such as, for example, agency conflicts between managers and shareholders, or precautionary motives in the face of uncertainty (Bates, Kahle, and Stulz (2006)). However, these standard cross-sectional determinantsofcorporatecashholdingshavebeenrelativelystableovertimeand,thus,canoffer at best only a partial explanation of why cash holdings have risen so much over time. On the theoryside,thecashtoassetratiospredictedbystandardcalibrationsofexistingmodelsaremuch smaller than their empirical counterparts (Riddick and Whited (2009)). Thus, the current high levelsofcashrepresentaquantitativepuzzleforexistingtheories. In this paper, we explore whether firms’ growing reliance on intangible capital in their productiontechnologycanhelpaddressboththeempiricalandthetheoreticalchallenges. Intangible capitalcannotbeeasilyverifiedorliquidatedand,assuch,cannotbepledgedascollateraltoraise debt financing. Under frictional capital markets where external funds command substantial premiums, we argue that its rising importance as an input of production may have boosted firms’ precautionary demand for cash in order to insure that they have sufficient liquidity to weather adverse shocks and to exploit investment opportunities. We document a large number of new stylized facts which support this "collateral channel" and broadly suggest that there is a strong empiricallinkbetweentheriseinintangiblecapitalandtheseculartrendincorporatecashholdings. Astructuralmodelthatcansuccessfullyreplicatethefactssuggeststhatthecollateralchan- 1SurveyevidencefromCFOsconfirmsthatliquiditymanagementtoolssuchascashareessentialcomponentsofa firm’sfinancialpolicy(Lins,Servaes,andTufano(2010),Campello,Giambona,Graham,andHarvey(2011)). 2

nel is also quantitatively important, thus offering a potential resolution of the puzzle for existing theories. Overall, our results indicate that intangible capital is crucial to providing a satisfactory analyticaccountofthemainstylizedfactsofcorporatecashholdings. Ourfocusonintangiblecapitalbuildsonavastbodyofevidencespanningseveralliteratures, which shows that over the last decades there has been a shift away from physical capital investments. ThereissolidevidenceattheaggregatelevelthatinvestmentsinintangiblecapitalbyUS firms have picked up substantially since the 1980s (Corrado, Hulten, and Sichel (2009) and Corrado and Hulten (2010)), especially in computerized information and private R&D. There is also evidence that organizational capital is becoming increasingly important (Lev (2001)). This welldocumentedshiftinfirms’modeofproductionisaneconomy-widephenomenon,somethingthat theliteraturehasdubbedageneralpurposetechnology(GPT)shock,orthethirdindustrialrevolution(JovanovicandRousseau(2005)). Thisbodyofevidencebroadlysuggeststhatfundamental technologicalchangesinthe1980sand1990shavehadapervasiveeffectonpubliccorporations. Westartbydocumentingthestylizedfactsofthelinkbetweentheriseinintangiblecapitaland theseculartrendincorporatecashholdings. Tothatend,weconstructanewfirm-levelmeasureof intangible capital, which is challenging since intangible assets are not reported on firms’ balance sheets. Existingattemptsatmeasuringintangiblecapitalhavebeenmostlyattheaggregatelevel. For example, one approach is to construct a proxy using aggregate stock market or accounting data (Hall (2001), McGrattan and Prescott (2007)). While these approaches measure intangibles as unexplained (by physical capital) residuals of stock market value or firm productivity, a more directapproachistoconstructaggregatemeasuresofthedifferentcomponentsofintangiblecapital,whichincludethestockofassetscreatedbyR&Dexpenditures,brandequity,andhumanand organizationalcapitalusingNIPAaccounts(Corrado,Hulten,andSichel(2009)andCorradoand Hulten(2010)). Webuildonthislatterapproachandusestandardaccountingdatatoconstructanewcomprehensivefirm-levelmeasureofintangiblecapitalforallnon-financialfirmsinCompustatbetween 1970 and 2010. Our measure is defined as the sum of three main components: the stock of informationtechnology(IT)capital;thestockofinnovative(R&D)capital;andthestockofhumanand organizationalcapital. The stockof innovativecapitalis constructedbycapitalizing R&Dexpendituresusingastandardperpetualinventorymethod(e.g.,Hall(2001)),whilethestockofhuman 3

and organizational capital capitalizes SG&A expenditures.2 IT capital is constructed capitalizing expendituresincomputersoftwarefromBEA. Usingthisfirm-levelmeasure,wedocumentseveralnewempiricalregularitiesonthelinkbetweenintangiblecapital,firmfinancing,andcorporateinvestment. Inparticular,thereisastrong positive relation between intangible capital and corporate cash both in the cross-section and in thetime-series,withintangiblecapitalemergingasthemostimportantfirm-leveldeterminantof cash holdings robustly across different specifications. Results of a simple out-of-sample forecastingexercisethatfollowstheapproachofBates,Kahle,andStulz(2006)showthataneconomically significant part of the overall increase in cash holdings over the last decades can be attributed to increases inintangible capital. Intangible capital doesnot affectonly cash levels, but alsothe adjustmentdynamicsofcashandtherelationbetweencorporateinvestmentandcashholdings. Finally,thelinkbetweencashandintangiblecapitalisespeciallystrongforfirmsthatarefinancially constrainedandthosethatbelongtoindustrieswithgreaterinvestmentinflexibility. Overall,our empiricalresultssuggestthatintangiblecapitalhasapervasiveimpactoncorporatecashholding decisionsandthattheimpactisduetobothfinancialandrealfrictions. Inordertobetterunderstandtheeconomicforcesthatdrivetheempiricallinkbetweenintangiblecapital,cashholdings,andcorporateinvestment,wenextdevelopanewdynamicmodelof thejointdeterminationoffirms’financing,riskmanagement,andcapitalaccumulationdecisions. The model is cast in an infinite-horizon, discrete-time stochastic environment, where firms make valuemaximizinginvestmentdecisionsinreal(tangibleandintangible)andfinancial(cashholdings,debt,andequity)assets. Therearetwokeyfrictions: first,financialfrictionsarisesincedebt financingissubjecttoacollateralconstraint,intangiblecapitalcannotbepledgedascollateral,and equityfinancinginvolvesissuancecosts;second,therearerealfrictionsthatarisesinceinvestment inrealassetsissubjecttonon-convexadjustmentcosts,whichmakeitinfrequentandlumpy(see AbelandEberly(1994)forastandardtreatment). In this setting, we show that the interplay of external financing costs and non-convex adjustmentcostsofinvestmentgeneratesaquantitativelylargeprecautionarydemandforcashhoarding becauseitmakesdifficultforfirmstogeneratefundsbyeitherdivestingrealassetsorraisingex- 2LevandRadhakrishnan(2005)andEisfeldtandPapanikolaou(2013)userelatedmeasuresoforganizationalcapital. BloomandReenen(2007)showevidencethatonetypeoforganizationalcapital,managerialpractices,mattersforfirm performance. 4

ternal finance (see Bolton, Chen, and Wang (2009), and Riddick and Whited (2009) for related settings,whichabstractfromdebtandcapitalheterogeneity;seealsoFroot,Scharfstein,andStein (1993) for a seminal model of corporate liquidity management). Intangible assets further boost thisprecautionarydemandforcashbyshrinkingdebtcapacitythroughatighteningofthecollateraldebtconstraint. Asaresult,firmswithgrowthoptionstendtoholdmorecashinanticipation of(S,s)-typeadjustmentsinphysicalcapitaltoavoidraisingcostlyexternalfinance. Toassessthe quantitative importance of this "collateral channel" through which firms’ debt capacity and asset tangibility are linked in our model, we develop a stationary general equilibrium analysis of net savings of corporations and households. All else equal, a technological transformation that increasesfirms’relianceonintangibleassetstolevelsthatmatchtheirempiricalcounterpartsinthe USeconomyforthelastdecadeleadstolevelsofcorporatecashholdingsthatareroughlyinline withtheirUSaveragesforthesameperiod. Inaddition,itendogenouslygeneratesalargeshiftof corporations’andhouseholds’balancesheetstowardswitchingtheirrolesasnet-savers,whichis alsoinlinewiththedata. Overall,theseresultssuggestthattechnologicalchangehascontributed significantlytorecentchangesinUScorporations’andhouseholds’privateliquiditydecisions. In addition to offering a potential resolution to important empirical and quantitative puzzles of corporate cash holdings, our results make two other important contributions. On the finance side, while it is well-established that more tangible capital assets support more debt (see Shleifer and Vishny (1992), and Hart and Moore (1994) for seminal contributions, and Rampini andViswanathan(2010)forrecentwork),ourcontributionistoexaminethequantitativeimplicationsofassettangibilityforliquiditymanagement,whicharerelativelylesswell-understood. Our proposedrationalefortheseculartrendincorporatecashholdingsbasedontechnologicalchange iscomplementarytorecentworkbyKarabarbounisandNeiman(2012)whoexplorethelinkwith the rise in the labor share.3 There are also recent studies by Bolton, Chen, and Wang (2013), Eisfeldt and Muir (2012) and Warusawitharana and Whited (2012), which focus on the short-run fluctuationofcashhoardingassociatedwithequitymarkettiming. Ourcontributionwithrespect 3Inourmodel,anincreaseincapitalsharealsoleadstoanincreaseincashhoardingasinKarabarbounisandNeiman (2012).However,wedidnotfollowthisroutesincewehavefoundthatthetheelasticityofprofitfunctionwithrespect tocapitalofU.S.Compustatfirms,whichisanincreasingfunctionofcapitalincomeshare,butadecreasingfunctionof marketpower,hasbeenstableovertimesince1970s. Thissuggestsapossibilitythattheremighthavebeenasecular trendinthemarketshare,whichoffsetstheinfluenceofincreasingcapitalincomeshareontheelasticity,leavingthe incentivetoholdcashthroughthischannelintactonaverage. 5

to these studies is to offer a quantitative explanation of the long-run secular trend in corporate savings, which remains challenging even in a setting with equity market timing.4 On the macroeconomics side, we contribute to the financial accelerator literature (Kiyotaki and Moore (1997), Bernanke,Gertler,andGilchrist(1999)andCooleyandQuadrini(2001))byextendingtheanalysis totheissueofdynamicriskmanagementunderfinancialmarketfrictionsingeneralequilibrium. 2 Intangible Capital and the Rise in Cash Holdings: The Evidence We begin our analysis by summarizing the stylized facts of the evolution of intangible capital and corporate cash holdings over time, as well as a number of related empirical regularities on intangible capital, firm financing, and corporate investment. To that end, we retrieve standard accounting data from Compustat to assemble a large panel of 18,535 US corporations over the 1970to2010period(176,877firm-yearobservations). 2.1 MeasuringIntangibleCapital Weconstructameasureofintangiblecapitalforeachfirm-year,5 whichisourempiricalproxyfor theamountofcapitalaccumulatedbypastinvestmentsinintangibleassets. Themainhurdlewith measuringintangiblecapitalisthat,sinceinvestmentsinintangibleassetsareexpensedintheyear inwhichtheyareincurred,thecapitalthatiscreatedbysuchinvestmentsisnotreportedonfirm balancesheets.6 Weuseannualdataonexpensesinthreebroadcategoriesofintangibleinvestment whoseimportancehasbeenemphasizedintheliteratureontheeconomicsofinnovation(Corrado, Hulten, and Sichel (2009) and Corrado and Hulten (2010)): knowledge capital, organizational capabilities,andcomputerizedinformationandsoftware. Intangiblecapitalisdefinedasthesum ofthestocksofinvestmentsinthesethreecategoriesdividedbynetbookassets. First,weconstructthestockofknowledgecapitalfrompastR&Dexpensesusingtheperpetual 4Our study also contributes to the small but growing theory literature on dynamic models of corporate savings (e.g.,Bolton,Chen,andWang(2009),RiddickandWhited(2009),GambaandTriantis(2008),AndersonandCarverhill (2012),andHugonnier,Malamud,andMorellec(2012))byshowingthataricherproduction-sideiskeytoimprovethe quantitativeperformanceofthisclassofmodels. 5As is standard in the literature, we exclude financial firms (SIC codes 6000-6999), regulated utilities (SIC codes 4900-4999),andfirmswithmissingornon-positivebookvalueofassetsandsalesinagivenyear. 6Corrado,Hulten,andSichel(2005)estimatethatroughly$1trillionofintangibleinvestmentisexcludedfromNIPAs annuallyovertheperiod2000to2003. 6

inventorymethod: G = (1 δ )G +R&D (1) it R&D it 1 it (cid:0) (cid:0) where G is the end-of-period stock of knowledge capital, R&D is the ($1990 real) expenditures t it onR&Dduringtheyear,andδ = 15%(Hall,Jaffe,andTrajtenberg(2000)).7 R&D Second,weconstructthestockoforganizationalcapitalfrompastsales,general,andadministrative(SG&A)expensesalsousingthesamemethodwithδ = 20%LevandRadhakrishnan SG&A (2005), Eisfeldt and Papanikolaou (2013)). These investments enhance the value of brand names andotherknowledgeembeddedinfirm-specifichumanandstructuralresourcesandincludeemployee training costs, payments to management and strategy consultants, and distribution systems. SinceSG&Aexpendituresincludeotherexpensesunrelatedtoinvestmentsinorganizational capabilities, we follow Corrado, Hulten, and Sichel (2009) and only weigh the stock of organizationalcapitalby0.2.8 Third, we construct the stock of computerized information and software by applying again perpetual inventory method with a depreciation rate of 31% as in the BEA data. Since these expenses are not reported at the firm level, we use the annual (2-SIC) industry level BEA Fixed ReproducibleTangibleWealth(FRTW)data. Wethenconstructamultipleofthisstocktotangible capitalstockattheindustrylevelandapplythemultipletoeachfirm’stangiblecapitalstock(PPE) toderiveafirm-levelstock.9 Ourresultingestimateforaverageintangibletotangiblecapitalover the last decade is close to 1, which is comparable to the estimate in Corrado, Hulten, and Sichel (2009)basedonaggregateNIPAaccounts. 2.2 StylizedFacts Webeginouranalysisbyconsideringthestylizedfactsoftheevolutionofcorporatecashholdings over time. Figure 1 helps to visualize the main hypothesis of this paper: that the rising share of intangiblecapitalinproductionisafundamentaldriveroftheseculardownwardtrendinleverage 7IfR&Dexpendituresareconstant(inrealterms),thestockofknowledgecapitalisGt =∑∞ s=0 (1 (cid:0) δ)sR&Dt (cid:0) s = R δ . WesettheinitialstocktobeequaltotheR&Dexpendituresinthefirstyeardividedbythedepreciationrateδ . In R&D addition, we interpolate missing values of R&D following Hall (1993) who shows that this results in an unbiased measureofR&Dcapital.ForfirmsthatdonotreportR&D,wesetR&Dtozero. 8Inrobustnessanalysiswehaveexploredalternativeweightsinawide(+/-50%)range, whichleaveourresults qualitativelyunchanged. 9Ourresultsarelittlechangedifwedonotincludethisstockinourmeasureofintangiblecapital. 7

andupwardtrendincorporatesavings. Thefigureplotsannualaveragesacrossfirmsoftheratio of intangible capital to book value of assets (left panel), of cash holdings to book assets (middle panel), and of net debt to book assets (net leverage ratio, right panel) over the last four decades. The intangible ratio rose tenfold from about 5% of net book assets in 1970 to about 60% in 2010. TheshareofliquidfinancialassetsinthebalancesheetsofU.S.corporationshasalsogrownfrom 8% to 22% over the same period, with cash holdings displaying a pronounced secular upward trend which was not concentrated in any particular decade, but rather has been steady; third, despitethecyclesovermedium-run,thenetleverageratioofU.S.corporationshastrendeddown from 20% to 6% over the same period, and appears to be cointegrated with the cash ratio with a coefficient -1. Panels A and C of Table 2 show averages (and medians) by decades of the cash ratio and net leverage for the entire sample and for the sub-sample of firms that invest in R&D. The increase in cash holdings is even stronger for these firms, suggesting that the secular trend inaveragecashisnotanartifactofthesefirmsbecomingmoreheavilyrepresented. Forinstance, mean(median)cashholdingsratioforthesefirmshaveincreasedfrom9%(5%)to27(19%)forthe sameperiod,whichis6ppt(8ppt)morethanthetotalsample. Next,wenowturntocross-industryandcross-firmunivariateevidence. TheleftpanelofFigure 2 plots the distribution of average industry cash and intangible ratios by decades. Intangible capital ratios have steadily risen in all broad industry categories (12-Fama and French) over the last four decades, consistent with an economy-wide shift in firms’ mode of production that affected firms well beyond just high-tech sectors.10 While the increase has been more dramatic in someindustries(e.g.,byafactorofalmost40,from0.13to5.07,inHealthcare),theintangibleratio went up by a factor of 10 (from 0.01 to 0.13) even in traditional industries such as retail (Shops). Thereisastrongcorrelationbetweenintangiblecapitalandcashratiosovertimebyindustry,with aregressioncoefficientofabout0.13andanR2 ofmorethan75%. Movingontocross-firmvariation,wecomputeforeachfirmthechangeinaverageintangible capital ratio before and after 1990 and divide the sample into deciles according to these firmlevel changes in intangible capital. The right panel of Figure 2 plots the corresponding average changeincashratiosbeforeandafter1990foreachdecileofthedistributionoffirm-levelchanges 10Inadditionalgraphicalanalysis,wehavedividedthesampleintotercileseachyearbysizeandage,high-techand othersectors,andincumbentandentrantfirms.Thisanalysisshowsthattheseculartrendincashhasnotbeenconfined toanyparticularsubsetoffirmsand,thus,hasbeenaneconomy-widedevelopment. 8

in intangible capital. Firms in the lower deciles have declines in intangible capital, while firms in the top deciles correspond to the largest increases. Changes in cash line up quite well along the diagonal, with firms that experienced a decline in intangible capital also seeing their cash ratiosdecline,whilefirmsforwhichintangiblecapitalrosethemostalsoexperiencingthegreatest increasesincash. PanelsBandDofTable1showadditionalunivariateevidenceoncross-firmvariationbystratifying the sample into four subsamples, based on quartiles of the empirical distribution of intangiblecapitalandshowingtheaverage(andmedian)cashandnetleverageratiosforeachofthese quartiles for the entire sample and for the sub-sample of firms that invest in R&D, respectively. Mean (median) cash ratios strongly and monotonically increase from about 8% (4%) in the bottom quartile of intangible capital to about 23% (12%) in the top quartile. The univariate relation betweencashandintangiblecapitalisevenstrongerwhenwerestrictthesampletoexcludefirms thatdonotreportR&D,withmean(median)cashratiosnowgoinguptoabout31%(23%)inthe topquartile. Finally,thecolumnstotherightshowthatfirmsinthetopquartileofintangiblecapital are also those for which cash matters the most to finance growth opportunities, as especially in this top quartile cash rich firms have on average been investing relatively more and growing faster. 2.3 PanelEvidence In the remainder of this section, we corroborate the stylized facts using panel data analysis. To that end, we regress cash holdings and net leverage ratios on our measure of intangible capital, whilecontrollingforasetofstandarddeterminantsofcashholdings(e.g.,Opler,Pinkowitz,Stulz, and Williamson (1999) and Bates, Kahle, and Stulz (2006)). In particular, we consider two specifications, OLS and fixed effects11, with firm-level controls such as industry cash flow volatility, market-to-book ratio, firm size, cash flow, capital expenditures, (cash) acquisitions expenditures, and a dummy for whether the firm pays dividend in any given year, as well as year effects to controlfortimevariationincashholdings.12 Weevaluatestatisticalsignificanceusingrobustclus- 11Although the time dimension of our sample is long (40 years), the panel is unbalanced. In order to reduce the “withingroupsbias”onexplanatoryvariables,weexcludefirmswithlessthanfiveyearsofdata. Forthefixed-effects specificationwereportthewithin-groupR2. 12SeeAppendixCfordetailedvariabledefinitions. 9

teredstandarderrorsadjustedfornon-independenceofobservationswithinfirms. TheresultingestimatesarereportedinPanelAofTable3fortheoverallsample(Columns(1)- (4))andforthesubsetoffirmsthatreportpositiveR&D(Columns(5)-(8)),toaddresstheconcern that the overall sample may reflect spurious differences in average cash holdings between noninnovative vs. innovative firms.13 The coefficient on intangible capital is robustly positive and statisticallysignificantacrossthetwosamplesandbothspecifications.14 Intangiblecapitalisalsoeconomicallysignificant. Forexample,forthebaselineOLSspecificationin Column (1), one standarddeviation increasein intangiblecapital isassociated withabout 8and1/2%increaseinthecashratio, whichisequaltoabouthalfthesamplemeanvalueofthe cash ratio of 15%. In the specification with firm fixed effects (Column (2)), the estimates decline by only about 2 and 1/2%, suggesting that intangible capital is also an economically significant determinantofthewithin-firmtime-seriesevolutionofcashholdings. Finally,estimatesforfirms withpositiveR&Dareevenlargerthanthosefortheentiresample,whichsuggeststhatourbaseline OLS result is not spurious and that intangible capital is an even more important economic determinantofcashholdingsforinnovativefirms. When we replicate our tests for net leverage, which is the ratio of total debt net of cash to book assets, the coefficient on intangible capital is robustly negative and statistically significant across both samples and specifications. It is also economically significant. For example, for the baseline OLS specification in Column (3), one standard deviation increase in intangible capital is associated with about 11 % decrease in net leverage ratio, which is almost as large as the sample mean value of net leverage of 14%. These results suggest that intangible capital is not only an importantdeterminantoffirms’cashholdingsdecisions,butalsooftheircapitalstructureandnet indebtedness.15 13The results are robust to using median regressions that address the concern that outliers firm-year observations withveryhighlevelsofcashmaybedrivingtheOLSestimates,aswellasOLSestimatesforaspecificationinchanges, ratherthenlevels.DetailedcoefficientestimatesforthecontrolvariablesarereportedinAppendixD. 14Signsandstatisticalsignificanceofcoefficientsoncontrolvariablesarealsounchangedacrossspecificationsand areinlinewiththefindingsofthepreviousliterature,withlargefirmsandfirmsthatpaydividendsholdinglesscash, andfirmswithhighercashflowvolatilityandmarket-to-bookholdingmore. Thecoefficientsoncapitalexpenditures and acquisitions are negative and significant, consistent with firms using their cash holdings to pursue investment opportunities.SeeAppendixDforacompletesummaryofestimationresults. 15Inadditionalchecks,wehaveverifiedthattheseresultsarerobusttocontrollingforR&Dexpenditures(flow),to usingdifferentdefinitionsofcashratios(cashasaratiotomarketvalueofassetsornetbookassets),andtoexcluding entrants(i.e.,firmsthatarenotpresentineveryyearofthesampleperiod)andfirmsinhightechnologysectors. 10

Quantifying the Contribution of Intangible Capital to the Rise in Cash Figure 1 shows that cashholdingsincreasedbyabout10%overthelastdecades,fromabout8%in1970toabout20% by 2010. To quantify the importance of intangible capital in explaining the secular trend in cash holdings, we investigate how changes in firm characteristics over time affect cash ratios.16 The intuitionforthisexerciseisasfollows:sampleaverageintangiblecapitalwas0.42in1980and0.75 in 2000. If the (unscaled) coefficient on intangible capital is 0.087, then we infer that, holding all other variables constant, the average cash ratio increased by 2.8 percentage points from 1980 to 2000 because of the increase in intangible capital, going from 3.7 percentage points (=0.087*0.42) in1980to6.5percentagepoints(=0.087*0.75)in2000. Panel A of Table 3 shows the results of this analysis that quantifies the contribution to the overallincreaseinthepredictedcashratioofchangesinthefirm-leveldeterminantsofthatratio. We first estimate the augmented OLS regression specification of Column (1) in the first half of the sample, i.e. the pre-1990 period. Using these coefficient estimates, we construct measures of the contributions of each of the explanatory variables in explaining changes in cash holdings between the 2000s and the pre-1990 period. Changes in intangible capital stand out as the most important driving factor of the rise in cash, with an increase in cash of about 3 (5) percentage pointsattributabletotheincreaseinintangiblecapitalintheoverallsample(inthesub-sampleof positive R&D firms). Changes in all other standard determinants have quite limited explanatory powerfortheriseincash. IntangibleCapitalandCashDynamics Wefurtherprobetheroleofintangiblecapitalindriving the time-series dynamics of corporate cash management by adding a lagged dependent variable toourbaselinespecification. Thisnewingredientallowsustodotwothings:first,thoughwedo not report the coefficient estimates for brevity, we check that our results are robust to allowing forimperfectionsincashrebalancingorpartialadjustmentincashratios(Lemmon, Roberts, and Zender(2008));second,wegatheradditionalevidenceontheroleoffinancingfrictions. Inparticular, we examine the hypothesis that intangible capital lowers the speed of adjustment (SOA) of cash: ifintangiblesmakeitmoredifficulttoraiseexternalfinance,thentheyshouldbeexpectedto 16To ease comparison and gauge the relative contribution of intangible capital compared to other standard determinants,weusetheapproachinBates,Kahle,andStulz(2006)andaugmenttheOLSspecificationwithnetdebtand equityissuance. 11

increaseadjustmentcostsofcash,thusleadingtolowerSOA(seeFaulkender,Flannery,Hankins, andSmith(2012)formoredetailsonthisintuition). Because there is an ongoing debate in the literature about the proper estimation procedure of SOA, Panel B of Table 3 reports results for a wide battery of SOA estimators. The annual SOA of cash ranges between 0.27 and 0.54 (not shown), suggesting that cash is imperfectly adjusted towarditstarget. Toprovideeconomicintuition,wetranslatetheseSOAsintohalf-lives,thetime thatittakesafirmtoadjustone-halfthedistancetoitstargetcashafteraoneunitshocktotheerror term. The half-life ranges from about 1 to about 2 years. Robustlyacross the different estimation techniques, SOAs decline monotonically with intangible capital. For example, OLS estimates in Column (1) imply that the half-life of 3 years for firms in the top quartile of the distribution of intangible capital is almost three time larger than for firms in the bottom quartile. These results areconsistentwiththehypothesisthatintangiblecapitalincreasesadjustmentcostsofcash. Corporate Investment and Firm Dynamics Next, we detail the empirical regularities that pertain to the real side decisions of firms. We ask whether cash holdings are an important source of financing for firm investment and growth and, if so, whether their importance varies systematically with intangible capital. To that end, we regress total investment (the ratio of the sum of capex and R&D to net book assets) and sales growth on lagged cash holdings, while controlling forasetofstandarddeterminantsofinvestment(e.g.,Gomes(2001)). WeconsiderbothOLSand firm fixed effects versions of this baseline model, with firm-level controls that include industry cashflowvolatility,market-to-bookratio,firmsize,cashflow,andadummyforwhetherthefirm paysdividendinanygivenyear,aswellasyeareffects. Weevaluatestatisticalsignificanceusing robustclusteredstandarderrorsadjustedfornon-independenceofobservationswithinfirms. The resulting estimates are reported in Panel A of Table 4 for the overall sample (Columns (1)-(4)) and for the subset of firms that report positive R&D (Columns (5)-(8)). The coefficient on lagged cash holdings is robustly positive and statistically significant across the two samples and both specifications, and is also economically significant. For example, for the baseline OLS specificationinColumn(1),anincreaseinlaggedcashholdingsfromtheirlowesttotheirhighest levelsleadstoabout7%increaseininvestment,whichisalmostaslargeasthesamplemeanvalue of investment of 10%. The estimate declines just a bit in the specification with firm fixed effects, 12

and is even larger for firms with positive R&D, suggesting that cash holdings are an even more importantsourceoffinancinggrowthopportunitiesforinnovativefirms. Panel B of Table 4 shows that intangible capital strengthens the relation between cash holdings and firm investment and growth. In fact, when we run our investment and sales growth regressions separately for each of four bins of our sample depending on quartiles of the empirical distribution of intangible capital, the size of the coefficient on lagged cash holdings increases monotonicallyandaboutdoublesaswemovefromthebottomtothetopquartile. Thesefindings holdrobustlyforbothinvestmentandsalesgrowth,aswellasforboththeentiresampleandthe sub-sampleoffirmsthatareactiveinR&D. The Role of Financial and Real Frictions In our last set of panel results, we use sample-split analysistobetterunderstandwhyintangiblecapitalisaneconomicallyimportantdeterminantof corporate cash holdings. In particular, we examine both financial and real investment frictions, whicharethekeyingredientsofourmodel. Iffirmswithmoreintangiblecapitalholdmorecash because of financing frictions, we would expect that the relation between intangible and cash shouldbestrongeramongfirmsforwhichfinancingfrictionsaremoresevere. Asforinvestment frictions,thebasicinsightofthevastliteratureonrealoptions(e.g.,AbelandEberly(1994),Bertola andCaballero(1994))isthatfixedadjustmentcostsleadfirmstomakelarge,lumpyinvestments. Thus, if intangible capital makes it more difficult to raise external finance, these real frictions may lead firms with more intangible capital to accumulate even more cash to finance their large investments. Panel A of Table 5 shows evidence supporting the role of financial frictions. We follow the standard approach in the literature (e.g., Hennessy and Whited (2007)) and in every year over thesampleperiodwerankfirmsbasedonfiveex-anteindicatorsoftheirfinancialconstraintstatus, which include firm size, dividend payer status, the WW-Index by Whited and Wu (2006), a measure of asset liquidation value by Berger, Ofek, and Swary (1996), and an index of industry assetredeployabilitybyBalasubramanianandSivadasan(2009). Weassigntothefinanciallyconstrained(unconstrained)groupsthosefirmsinthebottom(top)quartileoftheannualdistribution of each of these measures in turn, except for the financial constraints index, for which the ordering is reversed. Consistently across specifications and irrespective of which indicator of ex-ante 13

financing status is chosen, we find that the economic significance of the coefficient on intangible capitalismuchstrongerinthesub-samplesoffirmsthataremorelikelytofacefinancialfrictions. Forexample,theOLScoefficientinColumn(1)morethantripleswhenwegofromthetoptothe bottomquartileofthefirmsizedistribution(Rows[1]and[2]). PanelBsplitsthesamplebetweenbottomandtopquartilesofthefollowingfive(time-invariant) proxies of investment frictions: (4-SIC) industry frequency of investment inaction and an indicator for whether there are investment spikes in the industry, which are both defined following Cooper and Haltiwanger (2006); time-series skewness and kurtosis of annual aggregate industry investment, both based on Caballero (1999); and the time-series standard deviation of aggregate industryoperatingcosts. Theintuitionunderlingtheseproxiesisthat,duetotechnologicaldifferences, the extent to which firms face fixed costs varies across industries. Thus, industries where fixedcostarehigherarethosewherefirmsaremorelikelytoadjustinvestmentinfrequently,and, conditionalonadjusting,byaproportionallylargeramount. Inaddition,intheseindustriesfixed costs lead to a time-series distribution of aggregate investment that is sharply right-skewed and fat-tailed. Thus, we assign to the high (low) investment friction groups those firms in the top (bottom) quartile of the distribution of each of these measures in turn, except for the variability ofoperatingcosts,forwhichtheorderingisreversed. Consistentlyacrossspecificationsandirrespective of the indicator chosen, the economic significance of the coefficient on intangible capital ismuchstrongerinthesub-samplesoffirmsthataremorelikelytofaceinvestmentfrictions. For example, the OLS coefficient in Column (1) about doubles for firms that are in industries with investmentspikescomparedtothosewithoutsuchspikes(Rows[3]and[4]). In summary, the empirical regularities on intangible capital, firm financing, and corporate investmentare: Firm cash holdings (net indebtedness) have increased (decreased) over time, and are pos- (cid:15) itively (negatively) related with intangible capital both in the time series and in the cross section. Investment and growth are positively related to cash holdings, especially for firms with (cid:15) greaterintangiblecapital. Theadjustmentdynamicsofcashismoresluggishforfirmswithmoreintangiblecapital. (cid:15) The link between cash and intangible capital is stronger for the firms that are financially (cid:15) constrainedandbelongtoindustrieswithgreaterskewnessandkurtosisininvestmentrates. 14

3 A Structural Model of Corporate Cash Management This section develops a stationary general equilibrium model in which illiquidity of productive assets and financial market frictions interact with each other to determine firms’ optimal liquidity management policies. Our general equilibrium approach has a number of advantages: it allowsustoderivetheergodicjointdistributionofcapitalstock,financialassetsandliabilitiesand idiosyncratic technology of heterogeneous firms; it allows us to construct the exact moments of endogenous quantities in a simulation-free environment; it incorporates endogenous feedbacks of market prices in general equilibrium. To save space however, we focus on the description of firm problem in this section. A complete definition of stationary equilibrium and the numerical methodcanbefoundintheAppendix. 3.1 Technology Firms combine labor and capital to produce and sell output in a competitive market. They use two different types of capital: tangible (K ) and intangible (K ). In particular, the production T N technologytakesthefollowingDecreasing-Returns-to-Scale(DRS)Cobb-Douglasform: αξ/ρ Y = Z1 (1 α)ξN(1 α)ξ θ K T (cid:0) ρ +(1 θ) K N (cid:0) ρ (cid:0) F , 0 < ξ < 1 (2) (cid:0) (cid:0) (cid:0) O θ (cid:0) 1 θ (cid:0) " # (cid:18) (cid:19) (cid:18) (cid:0) (cid:19) where α is value added share of capital, ξ is the return-to-scale parameter, Z is an idiosyncratic technology shock, and N is labor hours. Production is subject to fixed operation costs, denoted by F , which make it possible for the firm to incur operating losses ex post. The idiosyncratic O technology shock follows a N state Markov Chain process with a transition function denoted z (cid:0) by Q(Z,dZ ).17 With the technological assumption, a static optimization over labor yields the 0 followingprofitbeforefixedoperationcost: γ/ρ Π(K ,K ;θ,ρ) = η(w)Z θ K T (cid:0) ρ +(1 θ) K N (cid:0) ρ (cid:0) (3) T N θ (cid:0) 1 θ " # (cid:18) (cid:19) (cid:18) (cid:0) (cid:19) (1 α)ξ (1 α)ξ 1 ((cid:0) 1 α)ξ αξ η(w) [1 (1 α)ξ] (cid:0) (cid:0) (cid:0) , γ . (cid:17) (cid:0) (cid:0) w (cid:17) 1 (1 α)ξ (cid:20) (cid:21) (cid:0) (cid:0) 17TheMarkovChainprocessisadoptedtofacilitateouranalysisoftheergodicdistributionoffirms’balancesheets ingeneralequilibrium.Later,wegeneralizethisprocesstoacontinuousMarkovprocess. 15

wherewismarketclearingwage. NotethatweassumethatcapitalserviceisgivenbyaCRS-CESformwiththeelasticityofsubstitutiongivenby1/(1+ρ). Forthebaselinespecification,wespecify ρ = ∞ ,implyingLeontief, min K /θ,K /(1 θ) . Inthisspecialcase, K /θ = K /(1 θ) K. Itisthenstraightforward T N T N f (cid:0) g (cid:0) (cid:17) to verify that the profit function is simply given by Π(K;θ,ρ) = η(w)ZKγ. Since the two types of capital are held in a fixed proportion, the assumption preserves tractability by allowing us to essentially eliminate one state variable. This reduces the size of problem considerably, especially ingeneralequilibrium,thusallowingustodevelopintuitionaboutthemainforcesatworkinthe model. Asarobustnesscheck,welaterconsiderthegeneralCEScaseρ < ∞ . Tomotivatefirms’cashholdings,itisnecessary,althoughnotsufficient,tointroduceilliquidity of long-lived capital assets. For this reason, we assume that all capital expenditures are only partially reversible (Abel and Eberly (1996)). We denote initial purchase prices and liquidation values by p+ and p for i = T,N. The partial irreversibility can be formally expressed as 0 i i(cid:0) (cid:20) p p+ fori = T,N.Sincethedesiretoholdcasharisefromtheimperfectresalevalue,wemake i(cid:0) (cid:20) i asimplifyingassumption, p+ = p+ fori = T,N: theinitialpurchasepricesareidentical. i In addition, we also assume that adjustment of capital is costly in either direction because it involves fixed adjustment costs (interchangeably, non-convex adjustment). We assume that both typesofcapitalaresubjecttothesameamountoffixedcosts,denotedby F . Combiningthetwo K assumptionsabouttheilliquidityofcapital,wecanexpresstheadjustmentcostas Γ(K ,K ) = p (K ,K )[K (1 δ)K ]+ν F (4) i0 i i i0 i i0 (cid:0) (cid:0) i K K whereν 1(K = (1 δ)K )andthepriceofcapital p (K ,K )isgivenby K (cid:17) i0 6 (cid:0) i i i0 i p+ if K (1 δ)K p (K ,K ) = i0 (cid:21) (cid:0) i (5) i i0 i p if K < (1 δ)K (cid:26) i(cid:0) i0 (cid:0) i for i = T,N with 0 p p p+. Note that, for simplicity, we assume that the two types of (cid:20) (cid:0)N (cid:20) (cid:0)T (cid:20) capitalhaveanidenticaldepreciationrateδ.18 18Inprinciple,itwouldbedesirabletospecifycapital-specificdepreciationrateand/orfixedcosttomatchthedynamicpatternsofadjustmentofthetwotypesofcapitalinthemicro-leveldata. Inourbaseline,thiswouldcreatean additionalstatevariable. However,thealternativespecificationthatweconsiderlater,thegeneralCEScase,isflexible enoughtoincorporatethisextension. 16

3.2 FinancingFrictions Firmshavethreefinancingoptions: (i)internalfunds,includingoperatingincomeandcashholdings; (ii) debt financing; and (iii) equity issuance. We consider capital market frictions that make the capital structure of the firm deviate from the Modigliani-Miller theorem. Next, we detail the debtandequitymarketfrictionsinturn. 3.2.1 DebtMarketFriction It is well-established in the literature that more tangible capital assets support more debt (see Shleifer and Vishny (1992), Hart and Moore (1994) and Rampini and Viswanathan (2010) for theoreticalarguments,andSibilkov(2009)forempiricalevidence). Thisisbecauseintangiblecapital, by its very own nature, is difficult to verify in quality or quantity. In fact, it often embodies the humancapitalofdevelopers,whichcannotbeeasilytransferredtoathirdentityinitsentirety. As a consequence, intangible capital is rarely pledged as collateral in debt contracts. To capture this feature,weassumethatthefirmcannotcommittotransferthetechnologyembodiedintheintangiblecapitalstocktocreditorsupondefault. Sinceembodiedhumancapitalcannotbetransferred, intangiblecapitalcannotbeliquidatedforapositivevaluebyathirdparty.19 Furthermore, in the spirit of Hart and Moore (1994) we assume that the firm’s output is observable, but not verifiable by a court. Hence, no debt contract can be written on the outcome of the firm’s output. Under this circumstance, as shown by Kiyotaki and Moore (1997), the only possibleformofdebtcontractisarisk-freedebtcontractcollateralizedbycapitalassets. Wediffer from Kiyotaki and Moore (1997) however in that only tangible capital assets constitute eligible collateral.20 Theresultingrisk-freedebtcontractissubjecttothefollowingborrowingconstraint: (1 δ)K (1 δ)θK B 0 (cid:20) Bmax(K T0 ;p (cid:0)T ) (cid:17) p (cid:0)T 1+r (cid:0) (1 τ T0 ) = p (cid:0)T 1+ (cid:0) r(1 τ 0 ) (6) i i (cid:0) (cid:0) 19However,weallowthefirmtohavedownsizingoption,i.e.,thefirmcanpartiallyliquidateintangiblecapitalstock byincurringtheliquidationcost1 (cid:0) p(cid:0)N . Animplicitassumptionisthatthefirm,aslongasitoperatesasanongoing concern, commits itself to deliver the human capital to the entity that is obtaining the liquidated part of intangible capital. 20Rampini and Viswanathan (2013) recently developed a model of investment, capital structure and risk managementinasimilarsettinginwhichonlytangiblecapitalworksascollateralasset. However, wedifferfromRampini andViswanathan(2013)inthattheyconsiderastatecontingentcontractbasedontherealizedcashflow. Inoursetting,weassumethatthedebtcontractcannotbecontingentupontherealizedcashflowforbothrealismandgreater precautionarysavings. 17

where r(1 τ ) is after-tax interest rate. The assumption that intangible assets cannot be used i (cid:0) as collateral is also broadly factual.21 Using a large sample of syndicated loans to US corporations for which a detailed breakdown of type of collateral used is available,22 we have verified thatcontractualloantermsstatethatonlyassetsthatcanbeeasilyvaluedrepresenteligiblecollateral. Consistentwiththelegaldefinitionofeligiblecollateral,onlyanextremelysmallminorityof securedsyndicatedloans(about3%oftotalloanvalue)havepatentsorbrandsusedascollateral. For later reference, we define the financial slack of the firm as Bmax(K ;p ) B . A natural T0 (cid:0)T (cid:0) 0 interpretationofBmax(K ;p )iscollateralizedlineofcredit arrangement. Notethattheconstraint T0 (cid:0)T isanoccasionallybindingone. Financialslackcanbedecomposedintotwopartsasfollows: Bmax(K ;p ) B = Bmax(K ;p ) max 0,B + [ min 0,B ]. (7) T0 (cid:0)T (cid:0) 0 T0 (cid:0)T (cid:0) f 0 g (cid:0) f 0 g Debt Cash Unusedlineof|credi{tz } | {z } | {z } Thesecondtermontheright-handsidecanbeinterpretedasdebtwhilethelasttermascash. The differencebetweenthefirstandthesecondtermisequivalenttotheunusedlineofcredit. In our stylized setting, firms never hold debt and cash at the same time. In the case when a firm finds it optimal to have strictly positive cash balances, the firm’s financial (liquidity) facility is composed of two terms: option to borrow up to the debt capacity given by Bmax(K ;p ) and T0 (cid:0)T the cash holdings given by min 0,B . When the firm finds it optimal to carry debt, the firm’s 0 (cid:0) f g remainingliquidityfacilityisgivenby Bmax(K ;p ) max 0,B ,theunusedlineofcredit. T0 (cid:0)T (cid:0) f 0 g 3.2.2 EquityMarketFriction Iftherewerenoequitymarketfrictions,thedebtmarketfrictionwouldplaynorolesincethefirm could undo it at no cost by issuing new equity. Thus, to create scope for active risk management policies, we assume that equity finance is costly in that raising outside equity reduces the value 21Usinganadditionaldatasource,CapitalIQ,whichcoversasmallercross-sectionoffirms(about1,000peryear)and ashortertime-series(2002-2010)buthasdetailedinformationonfirmdebtstructure,wehaveverifiedthatthemedian ratioofsecuredtototaldebtvalueisabout80%. 22Our loan information comes from a 2011 extract of Loan Pricing Corporation’s (LPC) Dealscan database. The dataconsistofdollar-denominatedprivateloansmadebybank(e.g.,commercialandinvestment)andnonbank(e.g., insurancecompaniesandpensionfunds)lenderstoU.S.corporationsduringtheperiod1981to2010,whichincludes about90,000loans. Mostoftheloansinthisdatasetareseniorsecuredclaims,featurescommontocommercialloans. However,adetailedbrakedownofcollateraltypesisavailableforonly20,000loans. 18

ofexistingshareholdersmorethanthenotionalamountofequityissuance(SeeMyersandMajluf (1984) and Cooley and Quadrini (2001)). We capture the loss to existing shareholders using a "dilution"function, ϕ(E): ϕ(E) ϕ ν K+ϕ max 0,E (8) (cid:17) 0 E 1(cid:1) f g where ν 1(E 0). In words, the firm incurs fixed costs when issuing new equity, which E (cid:17) (cid:21) are proportional to its size. In addition, the firm also incurs linear costs that are proportional to the amount issued. This specification is standard in the literature and facilitates comparison with the results of Bolton, Chen, and Wang (2009), who show that fixed costs of equity issuance significantlystrengthenfirms’precautionarydemandforcash. 3.3 ValueMaximizationProblem Theflowoffundsconstraintfacingthefirmcanbeexpressedas D = (1 τ )[Π(K;θ,ρ) F ] ∑ [Γ(K ,K ) τ δK ] [1+r(B)]B+B +E ϕ(E) (9) (cid:0) c (cid:0) 0 (cid:0) i0 i (cid:0) c i (cid:0) 0 (cid:0) i=T,N where D denotesthedividendspayout, τ isaflatratecorporateincometax.23 Weallow B tobe c 0 negative,inwhichcase B isinvestmentinliquidassets(cashaccumulation). Weassumethatthe 0 interestincometaxrate,τ ,islowerthanthecorporateincometaxrate,τ ,whichcreatesscopefor i c thefirmtoaccumulatedebt. Wealsoassumethatwhenthefirminvestinliquidassets–i.e.,when itaccumulatescash,itearnsareturnthatisstrictlylessthanrisk-freeaftertaxreturn,r(1 τ ) κ. i (cid:0) (cid:0) Weinterpretκ asagencycostofcashholdings.24 Hence,after-taxinterestratecanbeexpressedas r(1 τ ) if B 0 r(B) = (cid:0) c (cid:21) (10) r(1 τ ) κ if B < 0 (cid:26) (cid:0) i (cid:0) Despitethetaxadvantageofdebt,thefirmmayoptimallychoosetoholdcash. Inordertopreserve tractability,wedonotintroducefrictions,suchastransactioncosts,thatmakefrequentrefinancing ofdebtcostly. Thesefrictionsmayleadthefirmtosimultaneously holddebtandliquidassets,an 23Forsimplicity,weassumenodividendtaxation. 24ThesameassumptionismadebyBolton,Chen,andWang(2009). 19

issuethatisnotcentraltothetaskofexplainingthelowfrequencymovementofcash.25 Thefirmproblemcanbedefinedrecursivelyasthemaximizationofthevalueofequity, p (1 δ)θ V(K,B,Z) = min max (1+λ)D E+ψ (cid:0)T (cid:0) K B 0 0 λ,ψ K 0 ,B 0 ,D,E (cid:26) (cid:0) (cid:20) 1+r(1 (cid:0) τ i ) (cid:0) (cid:21) 1 + V(K ,B ,Z )Q(Z,dZ ) (11) 0 0 0 0 1+r(1 τ ) (cid:0) i Z (cid:27) s.t. (4),(8),(9)and(10) where λ and ψ are the Lagrangian multipliers associated with the nonnegativity constraint for dividends and the collateralized borrowing constraint, respectively. In particular, the former can beinterpretedastheshadowvalueofinternalfunds. Notethatthefirmdiscountsthecontinuation valuewithafter-taxriskfreerate. Whileweadoptageneralequilibriumframework,wesolvefor a stationary equilibrium without aggregate shocks. For this reason, we do not use the stochastic discountingfactorofthehousehold. The non-convexity of the value function with respect to capital due to the presence of fixed costofadjustmentmakestheanalysisoftheefficiencyconditionforcapitalrathercomplicated. To detourthiscomplexity,weredefinethefirmproblemasadiscretechoiceproblem, V(K,B,Z) = max V(K,B,Z ν = 1),V(K,B,Z ν = 0) K K f j j g Following Abel and Eberly (1994), we proceed in two steps: first, find the optimal investment under the assumption that action (ν = 1) is optimal; second, find the condition that ν = 1 is K K optimal. ToderivetheFOCforcapital,wefirstdefinethemarginalvalueofcapitalas 1 qM(K ,B ,Z) V (K ,B ,Z )Q(Z,dZ ) (12) 0 0 (cid:17) 1+r(1 τ ) K 0 0 0 0 (cid:0) i Z Note that Γ(K ,K) is everywhere differentiable with respect to K except at K = (1 δ)K. We 0 0 0 (cid:0) denoteleftandrightsidederivativesoftheadjustmentcostfunctionatK 0 = (1 (cid:0) δ)Kby Γ( K (cid:0) 0 ) ((1 (cid:0) δ)K,K)and Γ(+) ((1 δ)K,K). TheFOCforcapitalstocktomorrowcanthenbedescribedas26 K 0 (cid:0) 25Technically,toconsiderthiscasewewouldneedtointroduceanadditionalstatevariable. Aspointedoutinthe earliersection, cashandnet leveragearecointegrated withacoefficient closeto-1 despiteshortrun deviationfrom eachother. 26The condition (13) describes the optimal level of capital stock when action is optimal. However, the condition 20

1. IfqM(K ,B ,Z) / (qM ,qM+),K satisfies 0 0 (cid:0) 0 2 p (1 δ)θ K 0 : (1+λ)Γ K 0 (K 0 ,K) = qM(K 0 ,B 0 ,Z)+ψ 1+ (cid:0) r(1 (cid:0) τ ) (13) i (cid:0) 2. IfqM(K ,B ,Z) (qM ,qM+),K satisfies 0 0 (cid:0) 0 2 K : 0 = K (1 δ)K (14) 0 0 (cid:0) (cid:0) where qM (cid:0) (cid:17) (1+λ)Γ( K (cid:0) 0 ) ((1 (cid:0) δ)K,K) (cid:0) ψ 1 p + (cid:0) ( r 1 (1 (cid:0) δ) τ θ i ) (cid:0) and qM+ (1+λ)Γ(+) ((1 δ)K,K) ψ p (cid:0) (1 (cid:0) δ)θ . (cid:17) K 0 (cid:0) (cid:0) 1+r(1 τ i ) (cid:0) To build intution, consider the neoclassical case without financial frictions, i.e., λ = ψ = 0 always. Inthiscase,theoptimalityconditiononaction(13)reducesto Γ = qM,simplyequalizing K 0 themarginalcostandbenefitofinvestment. Thedifferencebroughtaboutbythefinancialfriction is that the marginal cost and benefit of adjustment are no longer measured by Γ and qM, but K 0 by (1+λ)Γ and qM +ψp (1 δ)θ/[1+r(1 τ )] as shown by (13).27 Since the shadow value K (cid:0) i 0 (cid:0) (cid:0) of internal funds can go above 1 with financial frictions, we can see that these frictions increase the marginal cost of investment to (1+λ)Γ Γ . However, they also increase the marginal K K 0 (cid:21) 0 benefit of investment, which is reflected in the fact that ψp (1 δ)θ/[1+r(1 τ )] 0. This (cid:0) i (cid:0) (cid:0) (cid:21) last term measures the shadow value of the borrowing constraint: as the firm chooses a greater productioncapacity,itexpandsitsdebtcapacityaswell. Thesecondtermofthemarginalbenefit ofinvestmentmeasuresthemarginalvalueofexpandeddebtcapacity. The FOC (13) shows that the target capital level chosen by the firm when action is optimal is afunctionofthefinancialconditionsofthefirm. Thisisincontrasttotheneoclassicalbenchmark in which the target capital stock is a function of investment fundamentals, i.e., the current technologylevel. Inparticular,thetangibilityofcapitalassetsaffectsinvestmentdirectlythroughthe presence of θ in the efficiency condition: only to the extent that the capital assets of the firm are itself does not establish the optimality of action. This latter optimality is provided by a value matching condition, V(K,B,Z ν = 1) V(K,B,Z ν = 0). Intheappendix,weprovideafurthercharacterizationofthevaluematching K K j (cid:21) j condition. Thisadditionalconditionisnecessaryduetothepresenceofthefixedcostofadjustment. Iftheadjustment frictionwereonlyduetothepartialirreversibility,thevaluematchingconditionwouldnotberequired(seeAbeland Eberly(1996)andAbelandEberly(1994)). 27Thefactthattheaction/inactionboundariesareaffectedbythefinancialconditionisalsoduetothepresenceof financialfriction. 21

pledgeableagainstdebtfinancing,capacityexpansionleadstotheexpansionofdebtcapacity. The seculardownwardtrendinassettangibilityθ affectsthetrade-offinvolvedinoptimalinvestment decisions by making the marginal cost of investment weigh more than the marginal benefit of investment. One margin firms have to avoid this de facto strengthening of financial market frictions is to reduce their dependence on debt financing by holding outright cash assets. Using a quantitativeexercisewewillprovideatestofthestrengthofthis"collateralchannel." The efficiency condition for equity issuance can be derived in the same way. We start with the observation that ϕ(E) is everywhere differentiable except at E = 0. Let ϕ (0)+ denote the 0 right hand side derivatie of ϕ(E) at E = 0.28 As in the case of investment problem, we derive an optimalityconditionbasedonν = 1asfollows: E 1. If(1+λ)(1 ϕ (0)+) 1,equityissuancesatisfies 0 (cid:0) (cid:21) 1 E : 1+λ = > 1 (15) 1 ϕ (E) 0 (cid:0) 2. If(1+λ)(1 ϕ (0)+) < 1,equityissuancesatisfies 0 (cid:0) E : E = 0 (16) Notethat(8)implies ϕ (0)+ = ϕ . Hence,theconditionforequityissuancecanalsobestated 0 1 as λ λ + ϕ /(1 ϕ ). A few remarks are useful to characterize the equity issuance policy. (cid:21) (cid:17) 1 (cid:0) 1 First, (15) implies that equity issuance is associated with 1+λ > 1. Given the complementary slackness condition λD = 0, the condition implies that D = 0 when equity issuance is strictly positive, i.e., the firm never issues new shares and pays out dividends simultaneously. Second, ϕ (E) = ϕ in our parametric assumption of equity issuance cost. This means that (15) does not 0 1 pindownthelevelofequityissuance. However,sinceD = 0whenE > 0,Eshouldbedetermined byafinancinggap, 1 ϕ +v Γ(K ,K)+(1+r )B [(1 τ )Π(K,Z;w)+τ δK+B ] , 1 ϕ 0 K 0 B (cid:0) (cid:0) c c 0 (cid:0) 1 (cid:8) (cid:9) which is the amount of equity issuance required to satisfy the flow of funds constraint with D = 28Anegativeissuance,i.e.,sharerepurchaseisidenticaltodividendpayoutinourmodel,andhencewedonotneed toconsiderthelefthandsidederivativeofϕ 0 (0) (cid:0). 22

0. Third, (15) and (16) imply that it may be optimal neither to issue new shares nor to pay out dividends. Evenwhentheshadowvalueofinternalfundsisstrictlygreaterthan1,whichrequires zerodividendpayout,itmaystillbeoptimaltodelayequityfinancinguntiltheliquidityproblem issufficientlyaggravatedinthesenseofλ ϕ /(1 ϕ ). Fourth,asinthecaseofrealinvestment (cid:21) 1 (cid:0) 1 problem,condition,(15)itselfdoesnotestablishtheoptimalityofactioninequityfinance. Owingto thefixedcomponentofequityissuancecost,theoptimalityofactionshouldbeinsuredbyavalue matchingcondition,i.e.,V(K,B,Z ν = 1) V(K,B,Z ν = 0). E E j (cid:21) j ToanalyzetheefficiencyconditionforB ,wedefine,inananalogytoTobin’sQ,financialQ as 0 1 qF(K ,B ,Z) V (K ,B ,Z )Q(Z,dZ ). (17) 0 0 (cid:17) (cid:0)1+r(1 τ ) B 0 0 0 0 (cid:0) i Z qF can be interpreted as the marginal value of corporate savings.29 The efficiency condition for net-debtdecisioncanthenbestatedas B : 1+λ ψ = qF(K ,B ,Z) (18) 0 0 (cid:0) When B < 0,theleftsideof(18)measuresthemarginalcostofinvestinginliquidfinancialasset 0 as the firm forgoes the current cash flow. In this case, the borrowing constraint becomes slack, theefficiencyconditionsimplybecomes1+λ = qF(K ,B ,Z). Therighthandsideof(18)should 0 0 then be interpreted as the marginal value of financial investment. When B > 0, the left hand 0 sideof(18)measuresthemarginalbenefitofadditionalcashflowfromborrowing. Therighthand sideof(18)inthiscaseisinterpretedasthemarginalcostofborrowingastheborrowingreduces equityvalue. Dependingontheutilizationrateofthelineofcredit,theborrowingconstraintmay become binding, in which case, the marginal benefit of additional borrowing is reduced by the shadowcostψ. In the particular case of non-binding collateral constraint, one can merge (13) and (18) into a singleexpression, qM(K ,B ,Z) qF(K ,B ,Z) = 0 0 . (19) 0 0 Γ (K ,K) K 0 0 29GilchristandHimmelberg(1995)coinedtheterm,“financialQ”.Ourdefinitionisnotidentical,butcertainlyconsistentwiththeirs.qFissimilartowhatBolton,ChenandWang(2012)called‘equityvalueofcash’.Inourframework,the marginalvalueismodifiedsothatitcancapturetheinteractionwiththecollateralconstraint,andhenceqF =1+λ ψ. Whenandonlywhenthecollateralconstraintbecomesslack,qFcanbeinterpretedasthemarginalvalueofcash. (cid:0) 23

This condition says that at the margin, the firm should be indifferent between physical investment and financial investment (either by investing in financial assets or by reducing debt). This suggests that the information contents of Tobin’s marginal Q and financial Q (or corporate cash hoarding) are correlated along any equilibrium path, and more so if the collateral constraint is binding less frequently. In our quantitative exercise, we show that the secular downward trend in asset tangibility leads to a lower frequency of collateral constraint binding as the firms want to hold greater portions of line of credits unused. As a result, we predict that the decline in the asset tangibility leads to a greater correlation between Tobin’s Q and cash hoarding, which is a newtestablepredictionofourtheory. 3.4 PropertiesofOptimalRealandFinancialPolicies In this section, before turning to the quantitative analysis we characterize the properties of the optimal policies of the individual firm in partial equilibrium by showing the marginal impact of eachstatevariable. Wedenoteafirm’srealinvestmentpolicybyK = g (K,B,Z;θ),andnetdebt 0 K policyby B = g (K,B,Z;θ). 0 B 3.4.1 InvestmentOptions,FinancialFrictionsandRiskManagement The top panels of Figure 3 show how the net leverage policy (B /K ) responds to the current 0 0 capacity level (K, on the horizontal axis). If the ratio is negative (positive), it means that the firm desires to hold net financial assets (liabilities) for the next period. The left panel shows the case whenfirmsfacebothnon-convexadjustmentfrictionandpartialirreversibility,whereastheright panelshowsthecasewiththeirreversibilityfrictiononly(wefollowthesameorderinsubsequent figures). Thefiguresassumethatthefirm’scurrentfinancialcondition(B)isheldconstantatitssteady statelevel. Thethreelinesfornetleverageratiocorrespondtothecaseswithanormaltechnology level(blue,solidline),30%above(black,dash-dottedline)and30%below(red,dashedline)the normal level. To highlight the properties of liquidity demand, we set θ = 0.4, a relatively low level of tangible capital ratio, which implies a low degree of pledgeability of production assets in borrowing contracts. In fact, as will be shown later, this value of θ is associated with a large 24

accumulationofnetfinancialassetsbyfirmsonaverage. Considerthecasewiththenon-convexadjustmentcostandirreversibility. Theoverallcontour of the net leverage policy can be described in the following way: first, the liquidity demand is associatedwithrelativelysmallfirmswhiledebtaccumulationisassociatedwithlargefirmsonthe scaleofthehorizontalaxis;second,firmstendtoaccumulateliquidfinancialassetsinanticipation ofalargeadjustmentofproductioncapacity,andthentodisburseallliquidfinancialassetsatthe momentofexercisinginvestmentoption(seethebottomleftpaneltogetherwiththetopleftpanel toseethejointdynamicsofcapacityexpansionandliquidityhoarding);third,thedurationofcash hoarding tends to be longer with lower technology as the firms desire to keep their investment optionsalivegiventhemeanrevertingprocessoftechnology. The general contour of the leverage policy for the pure irreversibility case is similar, but the overall level of cash hoarding is substantially smaller. The reason for this can be found in the difference in the nature of adjustments of physical capacity. In the bottom panels of the figure, onecanseethatthesizeofinvestmentinactionregionisgreatlyreducedwithoutthenon-convex adjustment.30 Forexample,whenthetechnologylevelisatitssteadystate(thecaseofblue,solid line), the inaction region without the nonconvex adjustment cost is more than 80% smaller than withthenon-convexcost,whichexplainstheshorterdurationofcashhoardingwithoutthenonconvexadjustmentfriction. Withthenon-convexadjustmentfriction,thesizeofadjustmentismuchlargerandlumpy. In the figure, the jumps up and down in capacity levels are on the order of 30% of existing capacity,whichiswhatCooperandHaltiwanger(2006)classifyas(dis)investment“spikes”or“lumpy investment.” Lumpy investment is what the firm wants to insure itself against by holding extra liquidity in frictional financial markets. Absent this friction, the optimal amount of liquidity insurance is substantially reduced in the top right panel. In fact, in this case, the financial policy cantakeanextremeformofnever-to-investinliquidfinancialassets,buildupleveragetoexploit taxbenefitsandtopayouttheproceedsofdebtissuanceasdividendsespeciallywhenthefirm’s currenttechnologylevelisunusuallyhighorthefirmhasasubstantialdegreeofovercapacityas shownbyblack,dash-dottedlineontheupperrightpanel. AsmentionedearlierinthediscussionofFOCs,thefinancialfrictionmodifiesthepropertiesof 30InactionregioniswhereK 0isequalizedto(1 δ)Kalongthethin,dashed,purplelineinthebottompanels. (cid:0) 25

the(S,s)adjustmentruleforproductioncapacity. However,asshowninthebottompanels,under the financial friction, the investment targets are not independent of currently installed capital because the amount of financing that can be raised through internal funds and borrowing may not be sufficient to cover investment expenditures if the gap between the target and currently installed capital is too large. As a result, K = g (K,B,Z;θ) may positively respond to K even 0 K whenfirm’sinvestmentmodeisactivedespitetheabsenceofconvexadjustmentcost. 3.4.2 FinancialFrictionsandDebt/CashDynamics Inthissubsection,weshowhowpastfinancialconditionsaffectthedynamicsofriskmanagement. To that end, we consider hypothetical firms with an identical level of technology (at its steady state), but with different net-debt position (B). In the upper panels of Figure 4, we show how firms’liquiditydemandchangesastheirinitialfinancialconditionchanges. Morespecifically,we considerthreelevelsof B thatwouldimplynetdebtratios-0.3(blue, solidline), 0.0(black, dashdottedline),and0.30(red,dashedline),relativetothesteadystateleveloftotalassets(bookvalue ofcapitalpluscash, B/A¯),respectively. The first can be considered as the case of so called cash cow firm, a special case with ample financial slack that allows the firm to behave like a financially frictionless firm in most of the statespace. Thethirdcanbethoughtofasthefirmwithweakbalancesheetcondition,whichare potentially subject to financial distress. Roughly, this case matches the net-debt structure of the firms in the 75 percentile of net-leverage distribution in 2012Q4 of Compustat. The second is an intermediate case. For each of these cases, we consider an identical level of asset tangibility at θ = 0.4. The key takeaway here is that there is a tremendous amount of inertia in firm’s financial position: firms that have net financial assets (positive net debt) today are more likely to hold net financial assets (positive net debt) tomorrow. In a frictionless world with no adjustment costs of financial assets/liabilities, there is no reason to expect such inertia since equity markets should provide perfect shock absorption. Equity frictions are responsible for making financial variables slow moving. Thus, the inertial dynamics of firms’ balance sheet is an indicator of financial frictions. 26

This is also related to the phenomenon known as the sensitivity of cash to cash flow in the literature (see Almeida, Campello, and Weisbach (2004)). A poor balance sheet condition today is likely to lead to a poor cash flow today, which, in turn, is likely to lead to a poor balance sheet conditiontomorrowunderfrictionalfinancialmarkets. Whatissurprisinginourresultsisthatit is not only the firms with positive net debt position that exhibit sensitivity to cash flow in their cash/debt policy, but it is also the firms with ample financial slacks that show sensitivity to current financial conditions. This is because the seemingly unconstrained firms have obtained such financialslacktoinsurethemselvesagainstfuturefinancialconstraintsinaforwardlookingmanner. Thelowerpanelsshowthesamephenomenonfromadifferentanglein(B,B )diagrams,where 0 apurple,dashedlineshows45degreeline,which,whencrossedbythepolicyfunctions,indicates fixedpoints g (K,B,Z) = B. Toshowtherelationshipbetween B and B ,weconsiderthreefixed B 0 levelsofproductioncapacity. Inallcases,wesetthetechnologylevelequaltothesteadystate,1. Consider the case with fixed cost and irreversibility, the lower left panel. The contour of the policyfunctionscanbedescribedas‘S-shape’withtwoflatregionsandanupwardslopingregion between them. Among the two flat regions, only one of them is the ‘target’ level of net-debt position,asonlythelowerplateaucrossesthe45degreelinefromtheabove. Theupwardsloping region is responsible for the sluggish dynamics of cash. It indicates a positive serial correlation betweentoday’sandtomorrow’sfinancialcondition,which,inturn,isentirelyduetothefinancial frictioninourframeworkratherthantofrictionsinadjustingbalancesheetcondition. Notethattheupwardslopingregionismuchsmallerwhenthereal adjustment friction isentirelyduetoirreversibility(thebottomrightpanel).31 Theinflexibilityintheadjustmentofphysical capital interacts with financial friction to create a greater serial correlation in cash balances. Thestrengthoftheinteractioncruciallydependsonthestateofthefirm,K and Z. 31Theirreversibilitycasealsoshowsthattheupwardslopingregionmaylieaboveorbelowthe45degreeline. Ifit isabovethe45degreeline,thefirmwantstoincrease(reduce)debt(financialassets).Theexactlocationoftheupward slopingpartcruciallydependsonthecurrenttechnologyandovercapacityasindicatedbythethreedifferentpolicy functions. 27

3.4.3 InvestmentPredictability: Cashvs. Tobin’sQ Non-convex adjustment frictions imply that firms adjust their capital infrequently, and investmentislumpywhenexecuted. Whenafirmanticipatesaninvestmentopportunity,butkeepsthe growth option alive, the value of equity grows due to the option value, but the current capacity shrinksataconstantrateofdepreciation. Thisimpliesthatafirm’sTobin’sQshouldbehighwhen it nears the exercise of growth option. In contrast, Tobin’s Q should be unusually low right after a large expansion of capacity. This pattern can be seen in Figure 5, which shows how Tobin’s Q respondstochangesincapitalaccumulation.32 Ifexternalfinancingiscostly,thesamedynamicpatterncanbeexpectedfortheaccumulation of liquid assets: a firm accumulates financial resources in anticipation of its investment opportunities,butdecumulatesitsfinancialassetsratherquicklyorevenaccumulatedebtafterexercising the options. This suggests that the information contents of Tobin’s Q and liquid asset holdings maybehighlycorrelatedunderfinancialmarketfrictionandnonconvexadjustmentfriction. Toverifythispossibility,thebottompanelsreproducethenetdebt-to-totalcapitalratiosshown inFigure3. Bycomparingtheupperandlowerpanelstogether,onecaneasilyseethatTobin’sQ and cash accumulation are perfectly correlated when technology level changes. For instance, in theleftpanelsforthecasewiththefixedadjustmentfriction,black,dashedverticallinesindicate the places where the orderings of Tobin’s Qs associated with different technology levels change. Tobin’sQforthecaseoflowtechnologyishigher(lower)thantheoneofhightechnologyexactly when net-debt policy g (K,B,Z ) is placed lower (higher) than g (K,B,Z ), suggesting that B low B high the ordering of Tobin’s Q might be driven by the same mechanism behind the opposite ordering of net-debt policy. To put it more simply, Tobin’s Q is higher when the firm’s liquidity demand, including unused line of credit, is higher. This confirms our conjecture that cash holdings and Tobin’sQmayconveythesameinformationregardingfutureinvestments. [1 3 + 2T r o (B bi ) n ]B ’s s Q ho is ul f d or b m e a in ll t y er d p e r fi e n te e d d a a s s th qA em (cid:17) ar 1 k + e r t ( 1 1 v (cid:0) a τ lu i ) eof V n (K et 0,B d 0 e , b Z t 0) p K + 0 o [1 s + it r i ( o B n 0) , ]B i. 0 e Q ., ( t Z he ,d m Z a 0 ) rk . e In tv a a n lu e e m o p f i ‘ r t i o c t a a l l c d o e n b t t e m xt in th u e s t c e a r s m h R andcashequivalents’.Thismeansthatforsomefirms,thetotalvalueofthefirm(V+[1+r(B)]B)maybelessthanthe valueofequity(V)asthefirmhasbecomenetcreditor. 28

3.4.4 AssetTangibilityandCashAccumulation We now show the effect of changes in asset tangibility on the liquidity management strategy. Figure 6 considers three different values of the tangible capital ratio, θ = 0.8, 0.5 and 0.3 for whichweshowthecorrespondingoptimal B policies(upperpanels). θ = 0.8ischosentomatch 0 the intangible-to-total tangible assets ratio, 0.25 (= (1 θ)/θ) of early 1970s in the data, shown (cid:0) in the first panel of Figure 1. θ = 0.5 can be interpreted as implying the intangible ratio 1.0 (= (1 θ)/θ) around early 2000s in the data. We then consider one more hypothetical value (cid:0) θ = 0.3, (or 2.3 = (1 θ)/θ in terms of the intangible capital ratio) for a ‘projection’. This case (cid:0) alsoservesasanillustrationoftheliquiditymanagementstrategyforfirmsoperatinginintangible intensiveindustries.33 Thedramaticincreaseinliquiditydemandinreactiontothedeclineofassettangibilityisevidentintheupperpanels. Forinstance, whentherealadjustmentfrictionconsistsoffixedadjustmentcostandirreversibility(theupperleftpanel),themaximumliquiditydemanddoublesfrom 0.25 to 0.5 relative to the book value of capital chosen for tomorrow when θ declines from 0.8 to 0.5. Theliquiditydemandsalsodoublesto1.0relativetothebookvalueofcapitalwhenθ further declinesto0.3. Thepureirreversibilitycaseexhibitsthesamepattern,althoughinthiscasethemagnitudeof cash holdings becomes much smaller relative to the case with non-convex adjustment friction. It is not the fixed cost per se that increases the demand for liquidity. The lumpy investment friction makes it optimal to adjust physical capital only infrequently, increasing the average size of adjustment,whichthenrequiresabulkofliquidfinancialresources. Underthefrictionalfinancialmarket,thefirmcannotpledgethecapitalagainstborrowingdue to the reduced debt capacity with the decline in θ. As the growing reliance on intangible capital in production reduces a firm’s financial buffer, the firm compensates for the forgone financial flexibility by holding more liquid assets. The bottom panels of the figure show that the marginal 33Careneedstobetakenwhensettingthecurrentfinancialcondition. Astraightforwardthingtodoissettingthe initialfinancialconditionatthesamelevelforallcases. However,doingsoissomewhatmisleading. Thisisbecausea particularfinancialconditionthatisclosetoitssteadystateforagivenparametervalue(θinthiscase)maybefaroff ofanothersteadystateunderadifferentparametervalue. Forthisreason,wesettheinitialfinancialconditionatthe stochasticsteadystatesobtainedinaseriesofpartialequilibriumsimulations.Forthisexercise,wesimulatethepartial equilibriumeconomyfor250periodswith10,000firmswithidenticalsetofidiosyncraticshocks. Wedeletetheinitial 50yearsofobservationsbeforewecomputethemeans. 29

gainsfromholdingliquidassetscanbequitesubstantial. Togaugethecontributionoftheliquidity managementstrategytoequityvalue,were-solvetheequityproblemafterimposingaconstraint that the firm cannot invest in liquid financial assets, i.e. 0 B Bmax(K ;p ). We denote (cid:20) 0 (cid:20) T0 (cid:0)T this equity value by VO. The difference between V and VO then measures the value of option to invest in liquid assets. This is equivalent to the willingness to pay to obtain the option to invest in liquidassets. Thebottompanelsshowthattheoptioncanincreasethevalueofequitymorethan 5percent,whichcanbeconsideredasignificantimprovement. 4 Results 4.1 Calibration Our key comparative statics consists in detailing the general equilibrium outcomes of our model for levels of asset tangibility that match their historical counterparts in the US economy over the last decades. The goal of the exercise is to assess whether the model can generate a good match forthehighaveragelevelofcashholdingsweobserveinUSdataforthelastdecade. Theexercise reliesonthefollowingcalibration. Theelasticityoftheprofitfunctionwithrespecttocapital γ is setequalto0.6asinpreviousstudies(forinstance,seeHennessyandWhited(2007)forastructural estimate). Wesettheannualdepreciationrateequalto0.10. Wechoosetheresalevalueofcapital p = p = p = 0.95. Thisisanupperboundwithrespecttotheavailableevidenceontheresale (cid:0) (cid:0)T (cid:0)N valueofcapital(see,forinstance,RameyandShapiro(2001)),whichwechoosetoshowthateven a relatively small discount in resale value is sufficient to generate substantial amount of saving in liquid asset holdings.34 To parametrize the fixed cost of investment, we follow Cooper and Haltiwanger (2006), who estimate a fixed cost of investment of about 1% of installed capital. We setthisvaluetobeproportionaltothesteadystatelevelofcapitalaccumulationinthefrictionless benchmark. Thefixedcostofoperation(F )issettobeequalto5%ofsteadystateprofitsfollowing O Gilchrist,Sim,andZakrajsek(2010). Thisvaluehelpstomatchdividendpayoutratiointhedata. Tocalibratetheidiosyncratictechnologyshockprocess,wesetρ = 0.8andσ = 0.3(equivalentto z z 0.15inquarterlyfrequency),whichisroughlyinlinewiththeestimatesofGourio(2008)regarding 34Wehavetriedarangeofresalevaluesbetween0.60and0.95. Whilealowerresalevalueofcapitalgeneratesmore cashholdings,wehavefoundthatthisisnotafirstordereffect. Infact,wechoose0.95toshutdownthisadditional channel. 30

the transitory part of idiosyncratic shock process based on Compustat data. As for the tangibleto-intangible capital ratio θ, this is the key parameter for which we provide an extensive set of comparativestatistics. The risk free rate is calibrated as 0.06, such that the after tax annual interest rate is about 0.04. We choose the fixed cost of equity issuance to be 1.5 percent of the steady state level of the frictionless capital stock. This is slightly higher than in Bolton, Chen, and Wang (2009), for example. For the linear cost of equity issuance, estimates and calibration choices in the literature range from 0.06 (Gomes (2001)) to 0.30 (Cooley and Quadrini (2001)). We choose a relatively conservative value of 0.15, which is roughly in the middle range. Finally, we set the corporate income tax and interest income tax rates as 0.35 and 0.30, respectively. As our results show, this differenceislargeenoughtocreateasubstantialincentivetoaccumulatedebtwithouttheneedto makeadditionalassumptionsonfirms’discountingfactorordeathprobability. Finally,wespecify averysmallagencycostofcashholdings,5bps. 4.2 ComparativeStatisticsinGeneralEquilibrium Tables7 9summarizeourquantitativeresultsontheeffectofassettangibilityonrealandfinan- (cid:0) cial decisions of individual firms and macroeconomic aggregates in stationary equilibrium. We consider3differentlevelsofassettangibility(tangible-to-totalcapitalratio),startingfromθ = 0.8 and going down to θ = 0.5, a range that roughly covers its empirical counterpart for the US economy since 1970s shown in Figure 1 and Table 2. For the sake of comparison and to further highlight the quantitative implications of the model, we also report model-implied moments for an evenlower hypotheticallevel of θ = 0.3. Wereport the resultsfor thenon-convex adjustment cost and irreversibility case in the first three columns and the results for the irreversibility only caseinthelastthreecolumnsofeachtable. First,considertheeconomywithrealfrictionsinvolvingbothnon-convexadjustmentandirreversibilityinTable7. Thefirstrowshowsthatasthetangiblecapitalratiodeclinesfrom0.8to0.5, thecash-to-tangibleassetratio(bookvalueoftangiblecapital,profitsandcashholdings)increases by 16ppt from 2% to 18%. This result shows that, in line with our reduced-form estimates, an increase in firms’ precautionary demand for liquid assets in response to a decline in the pledge- 31

abilityofproductionassetsgoesalongwaytowardexplainingasubstantialpartoftheobserved seculartrendinUScashholdings. Infact,overthe1970-2010period,thecashratiointhedataincreasedby12pptfrom9%to21% (the second row), suggesting that our model with non-convex adjustment frictions not only can resolvethequantitativepuzzleintheliterature,butifanythingitactuallyovershootsthedata. The third column of the table suggests that if asset tangibility had declined to 0.3, the model would have predicted that almost more than 40% of corporate balance sheets would consist of liquid financial assets, which represents a radical departure form the conventional notion of a public corporation which relies mostly on external finance. While this result may be surprising and the parameterchoiceisoutofrangewhenthebenchmarkistheentireUSeconomy,itdoeshavesome relevanceforfirmsoperatingintechnologyintensiveindustries. In contrast, the right panel of Table 7 shows that the irreversibility friction by itself is not sufficient to match the magnitudes of liquid assets seen in the data: as the tangible capital ratio goes down from 0.8 to 0.5, cash-to-tangible asset ratio only goes up 5ppt from zero to 5% (the first row). While the hypothetical decline in the tangible ratio to 0.3 can generate 20% cash ratio, theseresultsindicatethattheinterplayofbothfinancialandrealfrictionsiskeytoexplainingthe magnitudeofliquidityhoardingsweseeinthedata.35 Thefourthrowofthetableshowsthatassettangibilityhasadramaticimpactonnet-leverage, a result that mirrors the rise in precautionary savings and further corroborates our "shrinking debtcapacity"mechanism. Thelastrowofthetableshowsthatthedeclineinassettangibilityalso greatly reduces the usage of debt capacity: the limited pledgeability not only reduces the debt capacity of firms, but leads them to hold greater slack in borrowing capacity. As we mentioned earlier, the model penalizes cash holding in two ways: tax disadvantage and agency costs. The precautionarysavingmotiveinthefaceofrisingintangiblecapitalispotentenoughtoovercome these disadvantages. For instance, the economy with both non-convex cost and irreversible frictionsischaracterizedbyareductioninnet-leverageby30pptfrom47%to17%asassettangibility 35Increasing the degree of resale discount certaintly helps with generating greater cash hoardings in irreversible investmentmodels. Wehaveexperimentedwithsubstantiallylowerresalevaluesuptop (cid:0) =0.5. However,wecould increasethecashratioonlyby3pptallelseequal,thusconfirmingafundamentallimitationtogeneratingprecautionary savingthroughthischannel. Onereasonforthisfailureisthatoncetheresalevalueofcapitalgoesdownbelow0.85, firmsinthemodelneverdisinvestcapitalforliquidityreason(seeVeracierto(2002)),makingnomaterialdifferencefor liquidityhoarding. 32

declinesfrom0.8to0.5. Figure 7 shows how the joint distribution of capital and net-debt positions also responds to the decline in the tangible capital ratio from θ = 0.8 (blue bars) to 0.5 (red bars) to 0.3 (light blue bars). As the tangible capital ratio declines, the distribution moves to the left in such a way that more probability mass is allocated to the negative portion of the support on the net-debt dimension. In fact, with θ = 0.3, the stationary distribution allocates more probability mass to the negative support than to the positive support, which means that the corporate sector moves toward becoming a net creditor sector. The third column of the fourth row confirms this result. Theflipsideofthistransformationisthathouseholdsturnintoanetborrowingsector,supplying liquidfinancialassetstothecorporatesector. Table8showsseveraladditionalmodel-impliedmoments,includinghowassettangibilityaffectsthesensitivityofinvestmentandcashholdingstocashflows(Rows[1]-[2]),andthecomovementbetweeninvestment,cashandTobin’sQ(Rows[3]-[4]),andthepropertiesofdistributionof investment rates (Rows [5]-[6]). As predicted, asset tangibility generally increases the sensitivity of investment and cash to cash flow. This is another manifestation of the shrinking debt capacity. AsshowninRow[3], thecontemporaneouscorrelationbetweeninvestmentandTobin’sQis negativeduetothenatureofthe(S,s)adjustmentincapacity. Notablyandinlinewithourearlier discussion of investment predictability, the contemporaneous correlation between cash and Tobin’sQispositive,andmoreimportantly,increasingwiththedegreeofintangibility. Forinstance, considerthecasewithirreversibilityfrictiononly,showninRow[3]ofrightpanel. Whenθ = 0.8, the correlation is essentially zero. However as θ drops to 0.5, the correlation jumps to 0.83. The increase in the contemporaneous correlation is less dramatic when capacity adjustment involves non-convexadjustmentcostasitjumpsfrom0.6to0.88. Theseresultsconfirmourpredictionthat cash and Tobin’s Q become more highly correlated as the share of tangible assets declines and firmsareledtorelyincreasinglyoninternalsourcesoffunding. Finally, Table 9 shows that the decline in asset tangibility can have sizable, adverse consequences for the overall level of capital accumulation, output and consumption. Depending on the specification of the adjustment cost friction, a transition from θ = 0.8 to θ = 0.3 can have an impact on capital accumulation and consumption of an order of magnitude up to -2.3% and -1.2%, respectively. As the production technology shifts toward intangible assets, capital assets 33

loseasubstantialpartoftheireligibilityascollateral,whichleadstoatighteningofthedebtconstraint. Surprisingly, however, there is virtually no impact on total firm value, as shown in the lastrow. Thisresultimpliesthatfirmsareabletoalmostperfectlyneutralizetheimpactofdeclining asset tangibility by switching to liquidity management strategies that rely more heavily on cash. Nevertheless, the results on real economic activity point to an inefficiency in that the limitedpledgeabilityofcapitalassetshindersefficientcapitalallocation,aresultthatcouldhavebeen avoidedifthedebtcontractshadbeenfullyenforceable. 4.3 AlternativeSpecification: AGeneralCESCase OurbaselinemodelusesthespecialcaseofCEScapitalaggregatorwhenρ = ∞ , whichimpliesno substitutabilitybetweentangibleandintangiblecapitalinputs. Itisimportanttoassesstheextent towhichourresultsowetothisspecificassumption. Toaddressthisquestion,weconsiderthecase ofageneralCEScapitalaggregator. However,wehavetofaceaclassicalcurseofdimensionality issuethatarisesinthiscaseduetothefactthatwiththismoregeneraltechnologyweneedtotrack thetwotypesofcapitalasseparatestatevariables. A final contribution of the paper is to show that one can overcome this problem and recover tractability by adopting the following setup: (i) the idiosyncratic technology follows a geometric random walk, logZ = logZ +(cid:101), (cid:101) N( 0.5σ2,σ2); (ii) the profit function is homogeneous of (cid:0) 1 (cid:24) (cid:0) (cid:101) (cid:101) degree 1 in the technology and capital service, i.e., Π(K ,K ;θ,ρ) = η(w)Z1 γΦ(K ,K ;θ,ρ)γ; T N (cid:0) T N (iii)thefixedcostsofoperation, thefixedcostsofcapitaladjustment, andthefixedcostofequity issuance are proportional to the current technology level. Under these assumptions, the equity value function is homogeneous of degree 1 in (K ,K ,B,K ,K ,B ,Z) as the capital aggregator T N T0 0N 0 is homogeneous of degree 1 in (K ,K ,Z) and so is the profit function in Z and Φ(K ,K ;θ,ρ). T N T N Using this property, we can normalize the value function by the current technology level Z, and 34

solvethefollowing’normalized’valuemaximizationproblem: (1 δ )k˜ v(k ,k ,b) = min max (1+λ)d e+ψ p (cid:0) T 0T b˜ (20) T N λ,ψ k˜ 0T ,k˜ 0N ,b˜ 0 ,d,e( (cid:0) " (cid:0)T 1+r(b˜ 0 ) (cid:0) 0 # 1 k˜ k˜ b˜ + exp((cid:101) )v 0T , 0N , 0 dF((cid:101) ) 0 0 1+r(1 τ ) exp((cid:101) ) exp((cid:101) ) exp((cid:101) ) (cid:0) i Z 0 0 0 ! ) s.t. d = (1 τ )[π(k ,k ) F ] ∑ [Γ(k˜ ,k ) τ δ k ] (21) (cid:0) c T N (cid:0) o (cid:0) 0i i (cid:0) c i i i=T,N [1+r(b)]b+b˜ +e ϕ(e) 0 (cid:0) (cid:0) wherethenormalizedprofitfunctionisgivenby γ/ρ π(k ,k ) = η(w) (1 θ) k T (cid:0) ρ +θ k N (cid:0) ρ (cid:0) T N (cid:0) 1 θ θ " # (cid:18) (cid:0) (cid:19) (cid:18) (cid:19) andk K /Z,k K /Z,b B/Z,k˜ K /Z,k˜ K /Z,andb˜ B /Z. T (cid:17) T N (cid:17) N (cid:17) 0T (cid:17) T0 0N (cid:17) 0N 0 (cid:17) 0 Evenwiththenormalization,solvingforv(k ,k ,b)requirestremendouscomputingresources. T N Forthisreason,wedonotsolveforastationarygeneralequilibrium,butsolvefortheindividual firm problem, and then simulate the model economy with a set random draws for idiosyncratic technologyshocksinpartialequilibriumwith250periodsand10,000firms.36 Whilenotideal,our resultsindicatethatwedonotlosemuchinsightbyadoptingthispartialequilibriumsetting. Table 10 reports moments from this simulation exercise. In interpreting these results, it is importanttokeepinmindthatthevalueofθ isnolongeridenticaltothetangible-to-totalcapital ratio since firms can now freely deviate from this value. The idea of the exercise is to see if firms still react to changes in θ in the same way they did in our baseline results. With more flexible technology,firmsmaynothaveasstronganincentivetohoardliquidassets,sincetheynowgain anextramarginfrombeingabletosubstitutebetweenthetwotypesofrealassets. The first and second rows show the impact of technological change on the ratios of cash-totangible assets and net debt-to-tangible assets. Firms do not hold liquid assets when θ = 0.8. Relative to the Leontief case firms can now can use the more flexible technology as another risk management tool: when facing investment opportunities, firms can choose a higher ratio of 36Whencomputingmoments,wedeletetheinitial50periods. 35

tangible-to-intangiblecapitalratiotoexpandtheirborrowingcapacitydespitelowertechnological efficiency. ρ = 0.3 means that the marginal cost of deviating from the technologically efficient (cid:0) ratio is not big and, as a result, it reduces the amount of liquidity demand. However, as θ decreasesto0.5andto0.3,liquiditydemandpicksupsignificantlysincetheefficiencylossinvolved in the adoption of higher tangible capital ratios becomes higher and, as a result, firms are led to usetheirfinancialsideforriskmanagementandholdmoreliquidassets. Asinthebaselinecase, thenetleverageratiodeclinesmonotonicallyas θ goesdown. Thisresult, togetherwiththoseon cashratio,confirmsthatourbaselineconclusionsarenotdrivenbytheLeontiefassumption. The extendedmodelalsoshowsthattheinformationcontentofTobin’sQandcashholdingsisalmost identicalasthesevariablesarenearlyperfectlycorrelated(row[3]). Thelastrowofthetablereportsthecorrelationbetweennetleverageandtangibleassetratios. This correlation increases as θ goes down. This result shows that when firms find it optimal to choose higher leverage ratios they tend to choose higher levels of tangible capital ratios. In the model, causation runs in both directions. Firms can borrow more because they have more tangible capital as collateral. However, at the same time, more vulnerable balance sheet conditions also lead firms to hold more pledgeable assets in their balance sheet since these assets expand borrowingcapacityforalongperiodoftimeandsinceitiscostlytostrengthentheirbalancesheet conditionsbyholdingliquidassetsintheshortrun. Finally,anadvantageoftheCESextensionisthatitoffersasettingthatcanbeusedtoexplore additionalchannels,overandabovethecollateralonethatconstitutesthemainfocusofourpaper. Forexample,itwouldbeinterestingtoexplorewhetherlowerassettangibilityalsoleadstohigher cashholdingswhenintangibleassetshavelowerdepreciationratesoradjustmentcosts. However, these additional channels are beyond the scope of the present paper and constitute a potentially interestingavenueforfuturework. 5 Conclusion We have presented new evidence and theory which support the hypothesis that the rise in intangible capital can explain the secular increase in US corporate cash holdings over the last four decades. Our empirical evidence shows that intangible capital is a key empirical determinant of 36

cash holdings. In addition, the evidence suggests that both financial and real frictions contribute to explain why intangible capital matters so much. Next, we built a structural dynamic model whereintangiblecapitalmattersforfirms’cashmanagementdecisionsthrougha"collateralchannel"becauseoftheinterplaybetweenfinancialandinvestmentfrictions. Allelseequal,ourmodel generates an outsized increase in the demand for corporate cash in response to an increase in intangiblecapital. Weconcludethatintangiblecapitalisacrucialingredienttoprovidingasatisfactory analytic account of key stylized facts in corporate finance and macroeconomics, which to datehadeludedstandardexplanations. References ABEL, A. B., AND J. C. EBERLY (1994): “A Unified Model of Investment Under Uncertainty,” The AmericanEconomicReview,84(5),pp.1369–1384. (1996): “Optimal Investment with Costly Reversibility,” The Review of Economic Studies, 63(4),pp.581–593. ALMEIDA,H.,M.CAMPELLO,ANDM.S.WEISBACH(2004): “TheCashFlowSensitivityofCash,” JournalofFinance,59(4),1777–1804. ANDERSON, R. W., AND A. CARVERHILL (2012): “Corporate Liquidity and Capital Structure,” ReviewofFinancialStudies,25(3),797–837. BALASUBRAMANIAN, N., AND J. SIVADASAN (2009): “Capital Resalability, Productivity Dispersion,andMarketStructure,”TheReviewofEconomicsandStatistics,91(3),547–557. BATES, T. W., K. M. KAHLE, AND R. M. STULZ (2006): “WhyDoU.S.FirmsHoldSoMuchMore Cash Than They Used To?,” NBER Working Papers 12534, National Bureau of Economic Research,Inc. BERGER, P. G., E. OFEK, AND I. SWARY (1996): “Investor valuation of the abandonment option,” JournalofFinancialEconomics,42(2),257–287. BERNANKE, B. S., M. GERTLER, AND S. GILCHRIST (1999): “Chapter 21 The financial accelerator in a quantitative business cycle framework,” vol. 1, Part 3 of Handbook of Macroeconomics, pp. 1341–1393.Elsevier. BERTOLA, G., AND R. J. CABALLERO (1994): “Irreversibility and Aggregate Investment,” The ReviewofEconomicStudies,61(2),pp.223–246. BLOOM, N., AND J. V. REENEN(2007): “MeasuringandExplainingManagementPracticesAcross FirmsandCountries,”TheQuarterlyJournalofEconomics,122(4),1351–1408. BOLTON, P., H. CHEN, AND N. WANG (2009): “A Unified Theory of Tobin’s q, Corporate Investment, Financing, and Risk Management,” NBER Working Papers 14845, National Bureau of EconomicResearch,Inc. 37

(2013): “Markettiming,investment,andriskmanagement,”JournalofFinancialEconomics, 109(1),40–62. CABALLERO, R. J. (1999): “Aggregate investment,” in Handbook of Macroeconomics, ed. by J. B. Taylor,andM.Woodford,vol.1ofHandbookofMacroeconomics,chap.12,pp.813–862.Elsevier. CAMPELLO, M., E. GIAMBONA, J. R. GRAHAM, AND C. R. HARVEY (2011): “Liquidity Management and Corporate Investment During a Financial Crisis,” Review of Financial Studies, 24(6), 1944–1979. COOLEY, T. F., AND V. QUADRINI (2001): “Financial Markets and Firm Dynamics,” The American EconomicReview,91(5),pp.1286–1310. COOPER, R. W., AND J. C. HALTIWANGER (2006): “On the Nature of Capital Adjustment Costs,” TheReviewofEconomicStudies,73(3),pp.611–633. CORRADO, C., C. HULTEN, AND D. SICHEL (2009): “Intangible Capital And U.S. Economic Growth,”ReviewofIncomeandWealth,55(3),661–685. CORRADO, C. A., AND C. R. HULTEN(2010): “HowDoYouMeasurea&quot;TechnologicalRevolution&quot;?,”AmericanEconomicReview,100(2),99–104. EISFELDT, A., AND T. MUIR (2012): “The Joint Dynamics of Internal and External Finance,” Discussionpaper. EISFELDT, A. L., AND D. PAPANIKOLAOU (2013): “OrganizationCapitalandtheCross-Sectionof ExpectedReturns,”TheJournalofFinance,pp.n/a–n/a. FAULKENDER, M., M. J. FLANNERY, K. W. HANKINS, AND J. M. SMITH (2012): “Cash flowsand leverageadjustments,”JournalofFinancialEconomics,103(3),632–646. FROOT, K. A., D. S. SCHARFSTEIN, AND J. C. STEIN (1993): “Risk Management: Coordinating CorporateInvestmentandFinancingPolicies,”TheJournalofFinance,48(5),pp.1629–1658. GAMBA, A., ANDA.TRIANTIS(2008): “TheValueofFinancialFlexibility,”JournalofFinance,63(5), 2263–2296. GILCHRIST,S.,ANDC.P.HIMMELBERG(1995): “Evidenceontheroleofcashflowforinvestment,” JournalofMonetaryEconomics,36(3),541–572. GILCHRIST, S., J. SIM, AND E. ZAKRAJSEK (2010): “Uncertainty, Financial Frictions, and InvestmentDynamics,”Discussionpaper. GOMES, J. F. (2001): “FinancingInvestment,”TheAmericanEconomicReview,91(5),pp.1263–1285. GOURIO, F. (2008): “EstimatingFirm-LevelRisk,”mimeo,BostonUniversity. GOURIO, F., AND J. MIAO(2010): “FirmHeterogeneityandtheLong-RunEffectsofDividendTax Reform,”AmericanEconomicJournal: Macroeconomics,2(1),131–68. GREENWOOD, J., Z. HERCOWITZ, AND G. W. HUFFMAN (1988): “Investment, Capacity Utilization,andtheRealBusinessCycle,”AmericanEconomicReview,78(3),402–17. HALL, B. H. (1993): “TheStockMarket’sValuationofR&DInvestmentduringthe1980’s,”AmericanEconomicReview,83(2),259–64. 38

HALL, B. H., A. B. JAFFE, AND M. TRAJTENBERG (2000): “Market Value and Patent Citations: A FirstLook,”NBERWorkingPapers7741,NationalBureauofEconomicResearch,Inc. HALL, R. E. (2001): “The Stock Market and Capital Accumulation,” American Economic Review, 91(5),1185–1202. HART, O., AND J. MOORE (1994): “ATheoryofDebtBasedontheInalienabilityofHumanCapital,”TheQuarterlyJournalofEconomics,109(4),841–879. HENNESSY, C. A., AND T. M. WHITED(2007): “HowCostlyIsExternalFinancing? Evidencefrom aStructuralEstimation,”TheJournalofFinance,62(4),pp.1705–1745. HUGONNIER, J., S. MALAMUD, AND E. MORELLEC (2012): “Capital Supply Uncertainty, Cash Holdings,andInvestment,”SwissFinanceInstituteResearchPaperSeries11-44,SwissFinance Institute. JOVANOVIC, B., AND P. L. ROUSSEAU(2005): “GeneralPurposeTechnologies,”inHandbookofEconomicGrowth,ed.byP.Aghion,andS.Durlauf,vol.1ofHandbookofEconomicGrowth,chap.18, pp.1181–1224.Elsevier. KARABARBOUNIS, L., AND B. NEIMAN (2012): “Declining Labor Shares and the Global Rise of CorporateSaving,”NBERWorkingPapers18154,NationalBureauofEconomicResearch,Inc. KIYOTAKI, N., AND J. MOORE(1997): “CreditCycles,”JournalofPoliticalEconomy,105(2),211–48. LEMMON, M. L., M. R. ROBERTS, AND J. F. ZENDER (2008): “Back to the Beginning: Persistence andtheCross-SectionofCorporateCapitalStructure,”JournalofFinance,63(4),1575–1608. LEV, B. (2001): Intangibles: Management, measurement, and reporting. Brookings Institution Press, Washington,DC. LEV, B., AND S. RADHAKRISHNAN (2005): “TheValuationofOrganizationCapital,”inMeasuring CapitalintheNewEconomy,NBERChapters,pp.73–110.NationalBureauofEconomicResearch, Inc. LINS, K. V., H. SERVAES, AND P. TUFANO (2010): “What drives corporate liquidity? An internationalsurveyofcashholdingsandlinesofcredit,”JournalofFinancialEconomics,98(1),160–176. MCGRATTAN, E. R., AND E. C. PRESCOTT (2007): “Technology capital and the U.S. current account,”Discussionpaper. MYERS, S. C., AND N. S. MAJLUF (1984): “Corporate financing and investment decisions when firms have information that investors do not have,” Journal of Financial Economics, 13(2), 187 – 221. OPLER, T., L. PINKOWITZ, R. STULZ, AND R. WILLIAMSON(1999): “Thedeterminantsandimplicationsofcorporatecashholdings,”JournalofFinancialEconomics,52(1),3–46. RAMEY, V. A., AND M. D. SHAPIRO(2001): “DisplacedCapital: AStudyofAerospacePlantClosings,”JournalofPoliticalEconomy,109(5),958–992. RAMPINI, A. A., AND S. VISWANATHAN (2010): “Collateral,RiskManagement,andtheDistributionofDebtCapacity,”TheJournalofFinance,65(6),2293–2322. 39

(2013): “Collateralandcapitalstructure,”JournalofFinancialEconomics,109(2),466–492. RIDDICK,L.A.,ANDT.M.WHITED(2009): “TheCorporatePropensitytoSave,”JournalofFinance, 64(4),1729–1766. SHLEIFER, A., AND R. W. VISHNY (1992): “Liquidation Values and Debt Capacity: A Market EquilibriumApproach,”JournalofFinance,47(4),1343–66. SIBILKOV, V. (2009): “Asset Liquidity and Capital Structure,” Journal of Financial and Quantitative Analysis,44(05),1173–1196. STOKEY, N. L., E. C. PRESCOTT, AND R. E. LUCAS (1989): Recursive methods in economic dynamics / Nancy L. Stokey and Robert E. Lucas, Jr., with Edward C. Prescott. Harvard University Press Cambridge,Mass. VERACIERTO, M. L. (2002): “Plant-Level Irreversible Investment and Equilibrium Business Cycles,”AmericanEconomicReview,92(1),181–197. WARUSAWITHARANA, M., AND T. M. WHITED (2012): “Equity Market Misvaluation, Financing, andInvestment,”Workingpaper,SSRN. WHITED, T.M., ANDG.WU(2006): “FinancialConstraintsRisk,”ReviewofFinancialStudies,19(2), 531–559. 40

Figure1: IntangibleCapital,CashHoardingsandLeverage (a) Intangible−to−Net Total Asset (b) Cash−to−Total Asset (c) Net Leverage 1 0.24 0.22 0.9 0.22 0.2 0.8 0.2 0.18 0.7 0.18 0.16 0.6 0.16 0.14 0.5 0.14 0.12 0.4 0.12 0.1 0.3 0.1 0.08 0.2 0.08 0.06 1970 1980 1990 2000 2010 1970 1980 1990 2000 2010 1970 1980 1990 2000 2010 Note: Panel(a), (b)and(c)showintangiblecapitalratiorelativetototal(tangible)assets, cash-to-total (tangible)assetsandnet-debt-to-total(tangible)assets,respectively. ThesampleincludesallCompustat firm-yearobservationsfrom1970to2010withpositivevaluesforthebookvalueoftotalassetsandsales revenueforfirmsincorporatedintheUnitedStates. Financialfirms(SICcode6000-6999)andutilities (SICcodes4900-4999)areexcludedfromthesample,yieldingapanelof176,877observationsfor18,535 uniquefirms.VariabledefinitionsareprovidedintheAppendix. Figure2: ChangesinIntangibleCapitalandCash: Cross-Industry,Cross-FirmVariation 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 0 0.5 1 1.5 Intangible Ratio oitaR hsaC 30 1970s 1980s 1990s 20 2000s 10 0 y=0.13x + 0.09 R2=0.77 −10 −20 −30 −50 0 50 100 Changes in Intagible Ratio, p.p. p.p ,oitaR hsaC ni segnahC Decile 10 Decile 1 Note:ThesampleincludesallCompustatfirm-yearobservationsfrom1970to2010withpositivevalues forthebookvalueoftotalassetsandsalesrevenueforfirmsincorporatedintheUnitedStates.Financial firms(SICcode6000-6999)andutilities(SICcodes4900-4999)areexcludedfromthesample,yieldinga panelof176,877observationsfor18,535uniquefirms.VariabledefinitionsareprovidedintheAppendix Table 1 also shows the levels of each combination of cash ratio and intangible ratio across time and industry. 41

Figure3: InvestmentOpportunities,OptimalCashandCapacity (a) Bnew/Knew: IRR+FIX (c) Bnew/Knew: IRR 0.5 0.5 0 0 −0.5 −0.5 z = 1.0 z = 1.0 −1 z = 1.3 −1 z = 1.3 z = 0.7 z = 0.7 −1.5 −1.5 0 1 2 3 0 1 2 3 capital, K capital, K (b) Knew: IRR+FIX (d) Knew: IRR 3 3 2.5 z = 1.0 2.5 z = 1.0 z = 1.3 z = 1.3 2 z = 0.7 2 z = 0.7 1.5 1.5 1 1 0.5 0.5 0 0 0 1 2 3 0 1 2 3 capital, K capital, K Figure4: FinancialConditionandtheDynamicsofCash (a) Bnew/Knew: IRR+FIX (c) Bnew/Knew: IRR 0.5 0.5 0 0 −0.5 −0.5 b/E(k) = −0.3 b/E(k) = −0.3 −1 b/E(k) = 0.0 −1 b/E(k) = 0.0 b/E(k) = 0.3 b/E(k) = 0.3 −1.5 −1.5 0 1 2 3 0 1 2 3 capital, K capital, K (b) Bnew: IRR+FIX (d) Bnew: IRR 1 1 k/E(k) = 0.58 0.5 k/E(k) = 0.72 0.5 k/E(k) = 0.86 0 0 k/E(k) = 0.58 −0.5 −0.5 k/E(k) = 0.72 k/E(k) = 0.86 −1 −1 −1.0 −0.5 0.0 0.5 1.0 −1.0 −0.5 0.0 0.5 1.0 capital, B capital, B 42

Figure5: Tobin’s(Average)QandCashHoldings (a) Average Q: FIX+IRR (c) Average Q: IRR 3.5 3.5 z = 1.3 z = 1.3 3 z = 0.7 3 z = 0.7 2.5 2.5 2 2 1.5 1.5 1 1 0.5 0.5 0 1 2 3 0 1 2 3 capital, K capital, K (b) Bnew/Knew: FIX+IRR (d) Bnew/Knew: IRR 0.5 0.5 0 0 −0.5 −0.5 −1 −1 z = 1.3 z = 1.3 z = 0.7 z = 0.7 −1.5 −1.5 0 1 2 3 0 1 2 3 capital, K capital, K Figure6: AssetTangibility,ValueofLiquidityandCashHoldings (a) Bnew/Knew, FIX+IRR (c) Bnew/Knew, IRR 1 1 0.5 0.5 0 0 −0.5 −0.5 θ = 0.8 θ = 0.8 θ = 0.5 θ = 0.5 −1 θ = 0.3 −1 θ = 0.3 0 1 2 3 0 1 2 3 capital, K capital, K (b) 100 × Δlog(V/VO), FIX+IRR (d) 100 × Δlog(V/VO), IRR 6 6 θ = 0.8 θ = 0.8 5 θ = 0.5 5 θ = 0.5 θ = 0.3 θ = 0.3 4 4 3 3 2 2 1 1 0 0 −1 −1 0 1 2 3 0 1 2 3 capital, K capital, K 43

Figure7: AssetTangibilityandStationaryDistributionofCapitalandNet-Debt Note:Blue,redandcyanbarsarethecaseswithθ=0.8,0.5and0.3,respectively.Ez [µ]isthemarginal distributionof(K,B). Table1: EvolutionofIndustry-levelCashHoardingandIntangibleCapital 1970s 1980s 1990s 2000s Cash IntgblK Cash IntgblK Cash IntgblK Cash IntgblK Hlth 0.10 0.13 0.25 0.71 0.34 2.33 0.41 5.07 BusEq 0.10 0.18 0.18 0.47 0.27 0.89 0.31 1.75 Durbl 0.08 0.05 0.11 0.14 0.10 0.25 0.15 0.78 Chems 0.10 0.10 0.12 0.22 0.12 0.42 0.13 0.77 Other 0.11 0.02 0.15 0.13 0.15 0.35 0.21 0.59 Manuf 0.08 0.05 0.10 0.13 0.10 0.21 0.12 0.44 Telcm 0.06 0.02 0.08 0.07 0.13 0.13 0.14 0.35 NonDur 0.08 0.01 0.11 0.07 0.10 0.14 0.11 0.24 Enrgy 0.11 0.01 0.12 0.06 0.10 0.13 0.12 0.17 Shops 0.09 0.01 0.11 0.06 0.10 0.07 0.11 0.13 Note: Cashratioisdefinedascashandequivalentsrelativetototal(tangible)assetsandintangible ratio(IntgblK)isdefinedasintangible-to-tangibleassetratio. 44

Table2: StylizedFactsonIntangibleCapital,FirmFinancing,andCorporateInvestment Investment&FirmDynamics: FirmFinancing DifferenceBetweenCashRich andCashStrappedFirms PanelA:Time-SeriesStylizedFacts,byDecade Cash NetDebt Investment SalesGrowth Mean Median Mean Median Mean Median Mean Median 1970s 0.09 0.05 0.18 0.19 0.04 0.02 0.03 0.02 1980s 0.13 0.06 0.15 0.17 0.07 0.04 0.07 0.06 1990s 0.17 0.07 0.11 0.14 0.08 0.05 0.08 0.08 2000s 0.21 0.11 0.05 0.06 0.06 0.03 0.05 0.05 PanelB:Cross-sectionalStylizedFacts,byQuartileofIntangibleCapital Cash NetDebt Investment SalesGrowth Mean Median Mean Median Mean Median Mean Median IntangibleCapital,Q1 0.08 0.04 0.28 0.29 0.04 0.03 0.05 0.05 IntangibleCapital,Q2 0.10 0.05 0.20 0.21 0.04 0.03 0.07 0.06 IntangibleCapital,Q3 0.13 0.08 0.10 0.11 0.05 0.03 0.09 0.07 IntangibleCapital,Q4 0.23 0.12 -0.07 -0.10 0.13 0.06 0.11 0.08 PanelC:Time-SeriesStylizedFactsforInnovativeFirms(R&D>0) Cash NetDebt Investment SalesGrowth Mean Median Mean Median Mean Median Mean Median 1970s 0.09 0.05 0.16 0.18 0.04 0.02 0.04 0.03 1980s 0.15 0.07 0.10 0.12 0.08 0.04 0.08 0.07 1990s 0.21 0.11 0.01 0.04 0.09 0.05 0.10 0.09 2000s 0.27 0.19 -0.06 -0.06 0.07 0.03 0.06 0.05 PanelD:Cross-sectionalStylizedFactsforInnovativeFirms(R&D>0) Cash NetDebt Investment SalesGrowth Mean Median Mean Median Mean Median Mean Median IntangibleCapital,Q1 0.08 0.04 0.26 0.27 0.03 0.02 0.06 0.06 IntangibleCapital,Q2 0.10 0.05 0.18 0.20 0.04 0.02 0.08 0.07 IntangibleCapital,Q3 0.14 0.08 0.08 0.10 0.05 0.03 0.10 0.08 IntangibleCapital,Q4 0.31 0.23 -0.10 -0.14 0.14 0.06 0.12 0.09 Note: a. Thetablereportsmeansandmediansforvarioussub-samplesofallUSnonfinancialfirms(excluding Utilities)inCompustatfrom1970to2010[176,877observationsfor18,535uniquefirms]. b. Firmfinancingfacts refer to cash (ratio of the sum of cash and short-term marketable securities to book assets) and net debt (ratio oftotaldebtnetofcashholdingtobookassets). Investmentandfirmdynamicsfactsrefertototalinvestment (theratioofthesumofcapitalexpendituresandR&Dtonetbookassets)andsalesgrowth(annualchangein log sales). The reported figures are mean and median differences between cash rich and cash strapped firms, whicharedefinedasthosefirmsinthetopandbottomquartilesofthedistributionofyear-priorcashholdings, respectively. c. PanelsAandCreporttime-seriesevidencebydecadesfortheentiresampleandthesub-sample offirmsthatreportpositiveR&D,respectively. PanelsBandDreportcross-sectionalsortsbasedonintangible capital, which is defined as the sum of stocks of past investments in firms’ organizational capabilities, brand equity,andtechnologicalknowledge(R&D);itisnormalizedbynetbookassets. d. Detailedvariabledefinitions areprovidedinAppendixC.. 45

Table3: PanelEvidenceonIntangibleCapitalandFirmFinancing WholeSample R&D>0Firms PanelA:IntangibleCapital,Cash,andNetIndebtedness Cash NetDebt Cash NetDebt OLS FE OLS FE OLS FE OLS FE (1) (2) (3) (4) (5) (6) (7) (8) Intangible Capital t 1 0.086*** 0.061*** -0.111*** -0.047*** 0.104*** 0.067*** -0.125*** -0.067*** (cid:0) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) %PredictedRise 42.5% 43.4% PredictedRise 0.069 0.075 Yearfixedeffects Yes Yes Yes Yes Yes Yes Yes Yes FirmControls Yes Yes Yes Yes Yes Yes Yes Yes AdjustedR2 0.299 0.665 0.235 0.602 0.340 0.689 0.240 0.601 PanelB:CashDynamicsByQuartilesofIntangibleCapital OLS FE GMM OLS FE GMM (1) (2) (3) (4) (5) (6) Q1,SOA 0.463*** 0.721*** 0.557*** 0.527*** 0.748*** 0.729*** Half-life [1.1] [0.5] [0.9] [0.9] [0.5] [0.5] Q2,SOA 0.350*** 0.631*** 0.512*** 0.381*** 0.643*** 0.550*** Half-life [1.6] [0.7] [1.0] [1.4] [0.7] [0.9] Q3,SOA 0.266*** 0.526*** 0.348*** 0.300*** 0.537*** 0.331*** Half-life [2.2] [0.9] [1.6] [1.9] [0.9] [1.7] Q4,SOA 0.210*** 0.424*** 0.294*** 0.225*** 0.431*** 0.319*** Half-life [2.9] [1.3] [2.0] [2.7] [1.2] [1.8] Note: a. ThesampleconsistsofallUSnonfinancialfirmsinCompustatfrom1970to2010. b. PanelAreports estimatesfrompanelregressionsofcashholdingstobookassetsandnetdebttobookassetsonintangiblecapital for OLS and firm fixed effects specifications. Reported coefficients are the change in the dependent variable associatedwithaone-standarddeviationchangeinintangiblecapital.Columns(1)-(4)and(5)-(8)arefortheentire sampleandforthesubsampleoffirmswithpositiveR&D,respectively.c.PanelBreportsestimatesofthespeed ofadjustment(SOA)ofcashfordifferentsub-samplesbasedonquartilesofthedistributionofintangiblecapital. Thisspecificationaddsalaggeddependentvariable(firstlagofcash)tothesamesetofexplanatoryvariablesas inPanelA:Cashit=α 0+(1 α) Cashit 1+β X it 1+(cid:101) it .WereportetimatesofOLSregressionsanalogous (cid:0) (cid:3) (cid:0) (cid:3) (cid:0) toFamaandFrench(2002)(Columns(1)and(4)),OLSregressionswithfirmfixedeffectsanalogoustoFlannery andRangan(2006)(Columns(2)and(5)),GMMestimatesbasedonBlundellandBond(1998)(Columns(3)and (6)). Speedofadjustmentisα. Cashhalf-lifeisthetime(inyears)thatittakesafirmtoadjustbacktothetarget cashafteraone-unitshockto (cid:101), ln(0.5)/ln(1- α). d. Yeardummiesaswellasfirm-levelcontrolsforstandard determinantsoffinancialpoliciesareincludedinallregressions. p-valuesareinparenthesesandareclustered at the firm level. e. Predicted change in cash due to change in a determinant is obtained by taking the point estimatesfromtheOLSregressionestimatedoverthe1970-1989periodandmultiplyingthembythedifferencein averagevalueofeachdeterminantbetweentheestimation(1970-1989)andthepost-estimation2000-2010period. f.DetailedvariabledefinitionsareinAppendixC. 46

Table4: PanelEvidenceonIntangibleCapital,CorporateInvestment,andFirmDynamics WholeSample R&D>0Firms PanelA:SensitivitytoCashConditions Investment SalesGrowth Investment SalesGrowth OLS FE OLS FE OLS FE OLS FE (1) (2) (3) (4) (5) (6) (7) (8) CDF(Cash t 1 ) 0.071*** 0.059*** 0.145*** 0.125*** 0.076*** 0.059*** 0.166*** 0.132*** (cid:0) (0.000) (0.000) (0.001) (0.001) (0.000) (0.000) (0.001) (0.001) Yearfixedeffects Yes Yes Yes Yes Yes Yes Yes Yes FirmControls Yes Yes Yes Yes Yes Yes Yes Yes AdjustedR2 0.459 0.595 0.105 0.295 0.390 0.515 0.102 0.291 PanelB:SensitivitytoCashConditionsByQuartilesofIntangibleCapital Investment SalesGrowth Investment SalesGrowth OLS FE OLS FE OLS FE OLS FE (1) (2) (3) (4) (5) (6) (7) (8) Q1 0.049*** 0.045*** 0.145*** 0.109*** 0.054*** 0.059*** 0.145*** 0.106*** (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) Q2 0.055*** 0.053*** 0.151*** 0.118*** 0.055*** 0.061*** 0.155*** 0.148*** (0.001) (0.001) (0.001) (0.001) (0.000) (0.000) (0.000) (0.000) Q3 0.062*** 0.059*** 0.165*** 0.130*** 0.062*** 0.069*** 0.190*** 0.160*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Q4 0.096*** 0.078*** 0.280*** 0.268*** 0.106*** 0.089*** 0.303*** 0.283*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Note: a. ThesampleconsistsofallUSnonfinancialfirmsinCompustatfrom1970to2010. b. PanelAreports estimatesfrompanelregressionsoftotalcorporateinvestment(capex+R&D)andannualsalesgrowthonthe empirical cumulative distribution function (CDF) of lagged cash to book assets ratio for OLS and firm fixed effectsspecifications. Reportedcoefficientsarethechangeinthedependentvariableassociatedwithachange fromthelowesttothehighestvaluesoflaggedcash.Columns(1)-(4)and(5)-(8)arefortheentiresampleandfor thesubsampleoffirmswithpositiveR&D,respectively. c. PanelBreportsestimatesofthesameregressionsas inPanelAfordifferentsub-samplesbasedonquartilesofthedistributionofintangiblecapital.d.Yeardummies aswellasfirm-levelcontrolsforstandarddeterminantsofcorporateinvestmentareincludedinallregressions. p-valuesareinparenthesesandareclusteredatthefirmlevel.e.DetailedvariabledefinitionsareinAppendixC. 47

Table5: WhydoesIntangibleCapitalMatter? PanelEvidenceonFinancialandRealFrictions PanelA:FinancialFrictions PanelB:RealFrictions WholeSample R&D>0Firms WholeSample R&D>0Firms OLS FE OLS FE OLS FE OLS FE (1) (2) (3) (4) (5) (6) (7) (8) ByFirmSize ByIndustryFrequencyofInvestmentInaction [1] Q1 0.102 0.112 0.120 0.120 [1] Q1 0.024 0.031 0.030 0.036 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) [2] Q4 0.030 0.028 0.037 0.039 [2] Q4 0.137 0.112 0.162 0.133 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) ByDividendPayerStatus ByInvestmentSpikesintheIndustry [3] No 0.100 0.075 0.124 0.090 [3] No 0.049 0.048 0.061 0.049 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) [4] Yes 0.028 0.034 0.033 0.035 [4] Yes 0.095 0.079 0.120 0.081 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) ByWW-Index ByTime-SeriesSkewnessofIndustryInvestment [5] Q4 0.104 0.090 0.126 0.107 [5] Q1 0.044 0.043 0.060 0.047 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) [6] Q1 0.055 0.050 0.065 0.058 [6] Q4 0.100 0.079 0.145 0.104 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) ByAssetLiquidationValue ByTime-SeriesKurtosisofIndustryInvestment [7] Q1 0.145 0.109 0.152 0.115 [7] Q1 0.048 0.046 0.065 0.051 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) [8] Q4 0.051 0.049 0.059 0.054 [8] Q4 0.092 0.078 0.133 0.093 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) ByDegreeofAssetRedeployability ByTime-SeriesVariabilityofOperatingCosts [9] Q1 0.199 0.126 0.207 0.132 [9] Q4 0.038 0.040 0.054 0.050 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) [10] Q4 0.062 0.048 0.084 0.059 [10] Q1 0.102 0.080 0.133 0.096 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Note: a. The sample consists of all US nonfinancial firms in Compustat from 1970 to 2010. The table reports parameterestimatesfrompanelregressionsofcashholdingstobookassetsonintangiblecapitalforseveralsubsamplesplitsbasedonex-anteproxiesfortheseverityoffinancial(PanelA)andinvestment(PanelB)frictions facedbyfirms. b. Reportedcoefficientsarethechangeinthedependentvariableassociatedwithaone-standard deviation change in intangible capital, which is defined as the sum of stocks of past investments in firms’ organizationalcapabilities, brandequity, andtechnologicalknowledge(R&D)normalizedbynetbookassets. c. Columns(1)-(2)and(5)-(6)reportresultsforthewholesample,whileColumns(3)-(4)and(7)-(8)areforthesubsampleoffirmsthatreportpositiveR&D. Foreachofthetwosamples,wereportestimatesofOLSregressions andregresssionswithfirmfixedeffects. ControlsareasinTable2. d. InPanelA,thesampleissplitbetween bottomandtopquartilesof(year-prior)valuesof:firmsize(Rows[1]to[2]),WhitedandWu(2006)WW-Index (Rows[5]to[6]),Bergeretal.(1996)assetliquidationvalue(Rows[7]to[8]),andBalasubramanianandSivadasan (2009)indexofindustryassetredeployability(Rows[9]to[10]),andbydividendpayerstatus(Rows[3]to[4]). p-values clustered at the firm level are in parentheses. e. In Panel B, the sample is split between bottom and topquartilesof:(4-SIC)industryfrequencyofinvestmentinaction- Capex/bookassets<.01(Rows[1]to[2]), j j and whether in the industry there are investment spikes - Capex/book assets>.2 (Rows [3] to [4]), all based j j onCooperandHaltiwanger(2006); time-seriesskewness(Rows[5]to[6])andkurtosis(Rows[7]to[8])ofannualaggregateindustryinvestment(Capex/bookassets),basedonCaballero(1999);andthetime-seriesstandard deviationofaggregateindustryoperatingcosts(Rows[9]to[10]).f.VariabledefinitionsareinAppendixC. 48

Table6: BaselineCalibration Description Calibration ParametersofTechnologyandPreferences Returns-to-scale ξ = 0.83 Value-addedshareofcapital α= 0.30 Depreciation δ = 0.10 Elasticityofsubstitutionbetweencapitalinputs ρ = ∞ , 0.3 (cid:0) Purchasepriceofcapital p+ = 1.00 Partialirreversibility p = 0.95 (cid:0) Fixedcostofadjustment F = 0.01k k (cid:3) Fixedcostofoperation F = 0.05k 0 (cid:3) Persistenceoftechnologyshock ρ = 0.80 z Constantrelativeriskaversion σ = 1.00 Volatilityoftechnologyshock σ = 0.30 z InverseofFrischelasticityoflaborsupply φ = 1.00 Financial Parameters Risk-freerate r = 0.06 AgencyCostofCashHoldings κ = 5bps Fixedcostofissuance ϕ = 0.015k 0 (cid:3) Linearcostofissuance ϕ = 0.15 1 Interestrateincometaxrate τ = 0.30 i Corporateincometaxrate τ = 0.35 c Note:k (cid:3)isthesteadystatelevelofcapitalaccumulationinafrictionlessmodel. Table7: AssetTangibility,CashHoardingsandNet-Leverage FIX+IRR IRR Technologicalparameterθ 0.80 0.50 0.30 0.80 0.50 0.30 Cash-to-tangibleassets(model) 0.02 0.18 0.42 0.00 0.05 0.20 Cash-to-tangibleassets(data)a 0.09 0.21 - 0.09 0.21 - Netdebt-to-tangibleassets(model) 0.47 0.17 -0.24 0.60 0.43 0.13 Netdebt-to-tangibleassets(data)a 0.18 0.05 - 0.18 0.05 - Utilizationofdebtcapacity(model) 0.57 0.46 0.29 0.76 0.61 0.50 Note:Thetableshowsthemomentsofendogenousvariablesintheeconomywithirreversibilityandfixed adjustmentcost, usingstationarydistribution. Thestationarydistribution µ (cid:3) (K,B,Z) isdiscretizedona space(50,50,5).a.Datamomentsareaveragevaluesof1970sand2000s. 49

Table8: AssetTangibilityandCashFlowSensitivityofCashandInvestment FIX+IRR IRR Technologicalparameterθ 0.80 0.50 0.30 0.80 0.50 0.30 Corr(Investment,Cashflow) 0.28 0.30 0.33 0.21 0.25 0.30 Corr(Cash,Cashflow) 0.49 0.76 0.75 0.05 0.61 0.71 Corr(Investment,Tobin’sQ) -0.20 -0.21 -0.20 -0.31 -0.32 -0.31 Corr(Cash,Tobin’sQ) 0.60 0.88 0.91 0.05 0.83 0.86 Skewness(Investment) 2.19 2.27 2.27 0.68 0.38 0.41 Kurtosis(Investment) 8.10 8.47 8.22 5.23 4.69 5.11 Tobin’sQ(Mean) 1.63 1.62 1.61 1.64 1.66 1.66 Note:Thetableshowsthemomentsofendogenousvariablesintheeconomywithirreversibilityandfixed adjustmentcost, usingstationarydistribution. Thestationarydistribution µ (cid:3) (K,B,Z) isdiscretizedona space(50,50,5). Table9: AssetTangibilityandRealEconomyinStationaryEquilibrium FIX+IRR IRR Technologicalparameterθ 0.80 0.50 0.30 0.80 0.50 0.30 CapitalAccumulation( 10) 9.85 9.74 9.75 10.0 9.85 9.77 (cid:2) Consumption( 10) 5.57 5.54 5.55 5.64 5.59 5.57 (cid:2) Output( 10) 6.54 6.51 6.52 6.62 6.57 6.55 (cid:2) Hours( 10) 4.71 4.70 4.70 4.74 4.72 4.72 (cid:2) TotalValueofFirm 1.96 1.96 1.97 2.01 2.01 2.01 Note:Thetableshowsthemomentsofendogenousvariablesintheeconomywithirreversibilityandfixed adjustmentcost, usingstationarydistribution. Thestationarydistribution µ (cid:3) (K,B,Z) isdiscretizedona space(50,50,5). Table10: ComparativeStatics: theCaseofCES Technologicalparameter,θ 0.80 0.50 0.30 Cash-to-tangibleassets 0.00 0.12 0.25 Netdebt-to-tangibleassets 0.46 -0.04 -0.23 Corr(Cash,Tobin’sQ) n/a 0.94 0.97 Corr(Netleverageratio,Tangibleassetratio) 0.33 0.41 0.59 Note:Thetableshowsthemomentsofendogenousvariablesintheeconomywithirreversibilityandfixed adjustmentcost,usingactualsimulationofsizeT=200andN=10,000. TheMarkovChaininthebaselineis replacedwithaGauss-Hermitequadraturewithsamenumberofgrids. 50

Appendices A Stationary Equilibrium The model economy consists of a continuum of firms that combine capital and labor to produce final outputs, and a continuum of households that provide labor hours to firms to earn market wages,consumefinaloutputsandinvestinfirm’ssharesanddebtstoaccumulatewealth. Forthe descriptionofthefirmproblem,seethemaintext. A.1 StationaryDistribution The presence of persistent idiosyncratic shocks, DRS production technology and financial market friction imply a non-degenerate joint distribution of technology, capital accumulation and financial balances. We denote the joint distribution by µ(K,B,Z). At any point in time, the joint distributionsatisfiesthefollowingofmotion37: µ (K B Z) = 1(g (K,B,Z;µ) K) 1(g (K,B,Z;µ) B) Q(Z,Z)µ(dK,dB,dZ) (22) 0 k b (cid:2) (cid:2) 2 (cid:1) 2 (cid:1) Z where g (K,B,Z;µ) and g (K,B,Z;µ) are the optimal capital and financial policies that solve the k b program(11). Notethatthedistributionnextperiodµ isdeterminedbytoday’soptimalpolicies, 0 which then depends on today’s distribution µ via market clearing wage. The stationary distribution is the fixed point solution to the above functional equation: µ = µ = µ . The stationary 0 (cid:3) distributioncanbeusedtocomputeexactmomentsofanyorderaswellasaggregates.38 A.2 Households Since our focus is on firms’ investment and financial policies, we assume the existence of a representative agent for the household sector. The household is assumed to maximize the expected present value of utility flows discounted at β < 1: max E ∑∞ βtu(C ,H ). The utility flow is 0 t=0 t t strictly increasing and concave in consumption, and is strictly decreasing and concave in labor hours. For ease of interpretation of general equilibrium effects, we adopt a utility form that is nonseparableinconsumptionandhoursfollowingGreenwood, Hercowitz, andHuffman(1988). Morespecifically,wespecify 1 σ u(C,N) = 1 C ζ N1+φ (cid:0) 1 . (23) 1 σ (cid:0) 1+φ (cid:0) " # (cid:0) (cid:18) (cid:19) Suchspecificationabstractsfromwealtheffectsonlaborsupply,thusmakingitstraightforwardto interprettheeffectsoftechnologicalchangesoncapitalaccumulation. 37SeeStokey,Prescott,andLucas(1989)fortheoreticaldiscussion,andGourioandMiao(2010)foracomputational example. 38Inprinciple,itisconceivablethatthefirmmaywanttoexitowingtothepresenceoffixedcostsofoperation. For thisreason,weallowthefirmstohaveanexitoptionduringoursimulationsuchthatthevalueofequityistruncated belowbyacertainthresholdvalueV¯. Thiscanbeachievedbyreplacing V(K 0,B 0,Z 0 ) withmax V¯,V(K 0,B 0,Z 0 ) in f g thecontinuationvalueterm. However,exitdoesnotoccurinoursimulationwithbaselinecalibrationofthemodelin whichwesetV¯ =0.If,however,theoutsideoptionofthefirm,V¯ issubstantiallygreaterthanzerothefirmsmaywant toexitincertaincircumstances. Forsimplicity,weadoptanimplicitassumptionthatthesunkcostsofentryaregiven suchthatthatthevalueofoutsideoptionremainsatalowvalue,andthevalueofnewfirmentryisnogreaterthanthe requiredgrossreturnonthecostsofentry,i.e.,E[V(0,0,Z )] [1+r(1 τ )]F 0.whereZ isaninitialdrawfrom 0 (cid:0) (cid:0) i S (cid:20) 0 theerogodicdistributionofZandF isthesunkentrycosts,whichcanbeinterpretedasanaturalentrybarrier. S 51

The household earns competitive market wage w per work hour, and saves by investing in sharesanddebtissuedbythefirms. Thebudgetconstraintofthehouseholdisgivenby C+ (P S +B )µ(dK,dB,dZ) = wN+T +T (24) S 0 0 G F Z + (D+P˜ )S+[1+r(1 τ )]B µ(dK,dB,dZ). S i f (cid:0) g Z whereP isthe(ex-dividend)valueofequitytoday,P˜ isthe(ex-dividend)valuetodayofexisting S S sharesoutstandingyesterday. SandS arethenumberofsharesoutstandingyesterdayandtoday, 0 respectively. We assume that the proceeds of interest and corporate income taxes are transferred to the household in a lump sum, denoted by T . We also assume that all fixed costs of operation G and investment are transferred to the household in the same way, denoted by T . The two value F terms P and P˜ arelinkedtoeachotherbyanaccountingidentity, S S P S = [P˜ +E ϕ(E)]S (25) S 0 S (cid:0) Thisidentitysimplysaysthatthetotalvalueofequitytodayisthesumofthetotalvalueofshares outstanding yesterday and the value of new shares issued today. Substituting (25) in (24) and imposingstockmarketclearingconditionS = S = 1yields 0 C+ B µ(dK,dB,dZ) = wN+T +T (26) 0 G F Z + D [E ϕ(E)]+[1+r(1 τ )]B µ(dK,dB,dZ). i f (cid:0) (cid:0) (cid:0) g Z Afewremarksareinorder. First,bysubstituting(9)in(26),onecanseethatthetermE ϕ(E) (cid:0) vanishes. This makes it clear that the costs of issuing equity take the form of discount sales of new shares such that the dilution costs to the old shareholders are exactly offset by the gains of new shareholders in general equilibrium, thus leaving the resource constraint of the economy intact. Second, we use the same notation B to denote the debt issued by a firm and held by the household. When B is positive, this means that the household has a financial claim on a firm. When B is negative, this implies that the household owes money to a firm. Finally, we assume that the differential tax treatment of interest incomes and expenses applies only to the firms, but nottothehousehold. Thisimpliesthattheafter-taxinterestrateisequalizedtothetimediscount rateofthehouseholdinthestationaryequilibrium,1+r(1 τ i ) = β(cid:0) 1. (cid:0) A.3 Government Thegovernmentfollowsabalancedbudgetingrule,collectingtaxesandthentransferringthemto thehousehold. Thebudgetconstraintisgivenby T +T = [τ [Π(K,Z) F ] τ δK (τ τ )max 0,B ]µ(dK,dB,dZ) (27) G F c 0 c c i (cid:0) (cid:0) (cid:0) (cid:0) f g Z When a firm holds a strictly negative financial balance B, the taxes on interest incomes and the deductiononinterestexpensesofthehouseholdareoffset. A.4 StationaryEquilibrium Thestationaryequilibriumconsistsofaconstantmarketwagew,thestationarydistributionµ ,the (cid:3) individualpolicyrulesofthefirms,K = g (K,B,Z;µ ), B = g (K,B,Z;µ ), E = g (K,B,Z;µ ), 0 K (cid:3) 0 B (cid:3) E (cid:3) 52

D = g (K,B,Z;µ ), HD = HD(K,Z,w;µ ), I = p(g (K,B,Z;µ ),K) [g (K,B,Z;µ ) (1 D (cid:3) (cid:3) K (cid:3) K (cid:3) (cid:1) (cid:0) (cid:0) δ)K] g (K,B,Z;µ ), the policy rules of the representative household, C = C(w;µ ), and HS = I (cid:3) (cid:3) (cid:17) HS(w;µ )suchthatlaborandgoodsmarketsclear39: (cid:3) HS(w;µ ) = HD(K,Z,w;µ )µ (dK,dB,dZ) (28) (cid:3) (cid:3) (cid:3) Z C(w;µ ) = [Y(K,Z,w;µ ) g (K,B,Z;µ )+κmin 0,g (K,B,Z;µ ) ]µ (dK,dB,dZ). (29) (cid:3) (cid:3) I (cid:3) B (cid:3) (cid:3) (cid:0) f g Z B Investment with a Non-Convex Cost with Financial Friction Toanalyzeunderwhatconditionanactive(dis)investmentiswarranted,wereformulatetheprogram (11) as a discrete choice problem, V(K,B,Z) = max V(K,B,Z ν = 1),V(K,B,Z ν = 0) K K f j j g wherethetwoauxiliaryvaluefunctionscanbedefinedas p (1 δ)θ V(K,B,Z j ν K = 1) = m λ, i ψ n K 0 m ,B 0 a ,E x ,νE(cid:26) (1+λ)D (cid:0) E+ψ (cid:20) 1+ (cid:0) r(1 (cid:0) (cid:0) τ i ) K 0 (cid:0) B 0 (cid:21) (30) 1 + V(K ,B ,Z )Q(Z,dZ ) 0 0 0 0 1+r(1 τ ) (cid:0) i Z (cid:27) s.t. D = (1 τ )Π(K,Z;w)+τ δK Γ(K ,K) [1+r (B)]B+B +ν [E ϕ(E)] c c 0 B 0 E (cid:0) (cid:0) (cid:0) (cid:0) p (1 δ)θ V(K,B,Z j ν K = 0) = m λ, i ψ n B m 0 ,E a ,ν x E(cid:26) (1+λ)D (cid:0) E+ψ (cid:20) 1+ (cid:0) r(1 (cid:0) (cid:0) τ i ) (1 (cid:0) δ)K (cid:0) B 0 (cid:21) (31) 1 + V((1 δ)K,B ,Z )Q(Z,dZ ) 0 0 0 1+r(1 τ ) (cid:0) (cid:0) i Z (cid:27) s.t. D = (1 τ )Π(K,Z;w)+τ δK [1+r (B)]B+B +ν [E ϕ(E)] c c B 0 E (cid:0) (cid:0) (cid:0) The investment action is warranted if and only if V(K,B,Z ν = 1) V(K,B,Z ν = 0) > 0. K K j (cid:0) j Aftersubstitutingincomplementaryslacknessconditions,wecanexpresstheconditionas Γ(K ,K) [B B(0)]+ ν [E ϕ(E)] ν (0) [E(0) ϕ(E(0))] (32) 0 (cid:20) (cid:0) 0 (cid:0) 0 f E (cid:0) (cid:0) E (cid:0) g 1 + [V(K ,B ,Z ) V((1 δ)K,B(0) ,Z )]Q(Z,dZ ) 0 0 0 0 0 0 1+r(1 τ ) (cid:0) (cid:0) (cid:0) i Z wherewemakedistinctionsbetween B ,E ,ν and B(0),E(0),ν (0) todistinguishthesolutions f 0 0 E g f 0 0 E g for(30)and(31). ByconstructionK(0) = (1 δ)K. Notethatholding B , E ,ν , B(0), E(0) andν (0) 0 (cid:0) 0 0 E 0 E constant, both sides of the inequality are strictly increasing in K . We denote the level of capital 0 stock tomorrow that satisfies (32) with an equality by Kˆ. Such a capital stock may not exist. The derivativeoftherighthandsideisgivenbyqM(K ,B ,Z)sincealltermsexceptV(K ,B ,Z )does 0 0 0 0 0 notdependonK . Thederivativeoftheleftsideisgivenby Γ (K ,K). Wesummarizetheoptimal 0 K 0 0 investmentstrategyinthefollowingproposition. 39Thelastterminthegoodsmarketclearingconditionisduetotheagencycostofcashholdings, whichweview asanefficiencylosstotheeconomy. Thegoodsmarketclearingconditionscanbederivedbyaddingupthreeflowof fundsconstraintsofthefirms,therepresentativehouseholdandthegovernment. 53

Figure8: IllustrationofInvestmentProblem: ANeoclassicalCase 0.6 0.5 0.4 0.3 Γ(K’,K) 0.2 0.1 0 −0.1 −0.2 K* K(−) K(+) K* 0.2 0.4 0.6 0.8 (1−δ)*K 1 1.2 1.4 Note:Thehorizontallinemeasuresthelevelofcapitalstocktomorrowandtheverticallinemeasures the corresponding values of adjustment cost and the expected gains from adjustment. Blue, solid line depicts the shape of the adjustment cost function, black, dhash-dotted line and red, dash line illustratepossibleshapesofexpectedgainsfromadjustmentasafunctionofcapitalstockchosenfor tomorrow.Thetwogreenlinesegmentsdepicttheslopesoftheexpectedgainfunctions. Proposition1 If (i) Kˆ exists, (ii) qM(Kˆ,B ,Z) Γ (Kˆ,K), and (iii) K (K,B,Z) Kˆ, then the cap- 0 K (cid:3) (cid:21) 0 (cid:21) ital stock tomorrow is determined by (13), i.e., K = K (K,B,Z) > (1 δ)K; if (i) Kˆ exists, (ii) 0 (cid:3) qM(Kˆ,B ,Z) Γ (Kˆ,K) and (iii) K (K,B,Z) Kˆ, then the capital sto (cid:0) ck tomorrow is determined 0 K (cid:3) (cid:20) 0 (cid:20) by(13),i.e.,K = K (K,B,Z) < (1 δ)K;Otherwise,inactionisoptimal,i.e.,K = (1 δ)K. 0 (cid:3) 0 (cid:0) (cid:0) NotethatincasewhereV canbeshowntobestrictlyconcavewithrespecttoK in[Kˆ (cid:101),K +(cid:101)] 0 (cid:3) (cid:0) for a finite real number (cid:101) > 0, the third conditions, K (K,B,Z) Kˆ for the expansion problem, (cid:3) (cid:21) and K (K,B,Z) Kˆ for the contraction problem, are not required to establish the optimality of (cid:3) (cid:20) action. To build intuition, it is useful to consider a neoclassical case without financial distortions. If ϕ(E) = 0forall Eandτ = τ ,thevaluematchingcondition(32)issimplifiedinto i c 1 Γ(Kˆ,K) = [V(Kˆ,Z ) V((1 δ)K,Z )]Q(Z,dZ ) (33) 0 0 0 1+r(1 τ ) (cid:0) (cid:0) (cid:0) i Z The right hand side of the above, which measures the expected gain from adjustment, is strictly increasing in K as it depends on K only through V(K ,Z ). Furthermore, the right hand side is 0 0 0 0 equaltozeroatK = (1 δ)K. ThismeansthatatK = (1 δ)K,therighthandsideof(33)stays 0 0 strictlybelowtheleftha (cid:0) ndsideas Γ((1 δ)K,K) = F > 0 (cid:0) . k (cid:0) Figure 8 illustrates the investment problem for a frictionless neoclassical firm. Solid blue line depictstheadjustmentcostfunction Γ(K ,K). ThehorizontalaxismeasuresK andtheverticalaxis 0 0 showsthecorrespondingvaluesoftheleftandrighthandsidesof(33). Duetotheasymmetryof purchase and resale prices, the function is kinked at K = (1 δ)K. The fixed component of the 0 (cid:0) adjustment cost places the function strictly above zero at K = (1 δ)K. Black, dash-dotted line 0 (cid:0) 54

andred,dashlinedepicttwopossiblecontoursoftherighthandsideof(33). Asnotedbefore,the righthandsideisstrictlyincreasinginK andpassesthroughthepoint,((1 δ)K,0). 0 (cid:0) Consider the case of black, dash-dotted line. The line intersects the adjustment cost at K = 0 K(+). Notethattheslopeoftheblack,dash-dottedlineismeasuredbyqM. Asthelinecutsthrough the adjustment cost function from below, one can see that the second condition of Proposition 1, qM(Kˆ,Z) Γ (Kˆ,K) is satisfied. On the left side of the vertical line K = K(+), the expected K 0 (cid:21) 0 gain from adjustment is strictly less than the cost of adjustment. On the right side of the vertical line, a strictly positive adjustment is warranted as the gain dominates the cost. In the latter case, the optimal level of capital stock tomorrow K is determined such that the FOC qM(K ,Z) = (cid:3) (cid:3) Γ (K ,K) is met. The green line segment located in the northeast corner of the figure depicts K (cid:3) 0 the slope of the gain from adjustment, the slope of which coincides with qM(K ,Z). Note that (cid:3) qM(K ,Z) = Γ (K ,K) is a necessary condition for adjustment, but not a sufficient condition (cid:3) K (cid:3) 0 as it may be satisfied with the expected gain from adjustment stays below and never crosses the adjustmentcostfunction. Usingthesamelogic,onecanseethatK locatedatthesouthwestcorner (cid:3) of the figure corresponds to the case of contraction when the expected gain function is given by thered,dashline. Itisimportantthattheexpectedgainfunction(therighthandsideof(33))may neverintersecttheblue,solidline. Insuchacase,inaction,i.e.,0 = K (1 δ)K isoptimal. 0 (cid:0) (cid:0) C Details of Variable Definition Thevariablesusedintheanalysisaredefinedasfollows: Cashtobookasset–ourmaindependentvariable–isdefinedcashandmarketablesecurities (cid:15) (dataitem#1)dividedbybookassets(#6) Othercashmeasures(robustness): CashtoNetBookAssetsiscashandmarketablesecurities (cid:15) (#1) divided by book assets (#6) minus cash and marketable securities (#1); Cash to Market Value of Assets is cash and marketable securities (#1) divided by long-term debt (#9) plus debtincurrentliabilities(#34)plusmarketvalueofequity. Net-Leverage is the ratio of long-term debt (#9) plus debt in current liabilities (#34) minus (cid:15) cashandmarketablesecurities(dataitem#1)tobookassets(#6). Industrysigma(cashflowrisk)isthestandarddeviationofindustrycashflowtobookassets. (cid:15) Standard deviation of cash flow to book assets is computed for every firm-year using data overtheprevioustenyears. Wethenaveragethesecashflowstandarddeviationsover2SIC industriesandeachyear. Market-to-book ratio is the ratio of the book value of assets (#6) minus the book value of (cid:15) equity(#60)plusthemarketvalueofequity(#199*#25)tothebookvalueofassets(#6). Firmsizeisthenaturallogarithmofbookassets(#6)in1990dollars(usingCPI). (cid:15) Cash flow is earnings after interest, dividends, and taxes before depreciation divided by (cid:15) bookassets((#13–#15–#16–#21)/#6). Capitalexpendituresistheratioofcapitalexpenditures(#128)tobookassets(#6). (cid:15) Dividendisadummyvariableequaltooneinyearsinwhichafirmpaysacommondividend (cid:15) (#21). Otherwise,thedummyequalszero. Acquisitionsistheratioofacquisitions(#129)tobookassets(#6). (cid:15) Networkingcapitalistheratioofnetworkingcapital(#179)minuscash(#1)tobookassets (cid:15) (#6). 55

Leverageistheratiooflong-termdebt(#9)plusdebtincurrentliabilities(#34)tobookassets (cid:15) (#6). Net debt (equity) issuance is annual total debt (equity issuance minus debt retirement (eq- (cid:15) uityrepurchases),dividedbybookassets. R&D(flow)istheratioofR&Dexpenditures(#46)tobookassets(#6). (cid:15) Asset Tangibility is the ratio of net PPE (#8) to book assets (#6) minus cash and marketable (cid:15) securities(#1). High-tech industries are defined following Loughran and Ritter (2004) as SIC codes 3571, (cid:15) 3572, 3575, 3577, 3578, 3661, 3663, 3669, 3674, 3812, 3823, 3825, 3826, 3827, 3829, 3841, 3845, 4812,4813,4899,7370,7371,7372,7373,7374,7375,7378,and7379. WW-IndexisbasedonWhitedandWu(2006)andisasfollows: WW-Index=-0.091*CashFlow (cid:15) -0.062*Dividend+0.021*Leverage-0.044*Size+0.102*IndustryGrowth-0.035*Growth,where IndustryGrowthisthe4-SICindustrysalesgrowth, Growthisown–firmrealsalesgrowth, andtheothervariablesareasdefinedabove. Asset liquidation value is based on Berger et al. (1996) and is the sum of 0.715*Receiv- (cid:15) ables(#2),0.547*Inventory(#3),and0.535*Capital(#8). IndustryassetredeployabilityindexisbasedonBalasubramanianandSivadasan(2009)and (cid:15) is the fraction of total capital expenditures in an industry accounted for by purchases of used(asopposedtonew)capital,computedat4-digitSIClevelandconstructedusinghandcollectedUSCensusBureaudata. Sincethesedataareavailableonlyonceevery5yearsand not for more recent years, we compute a time-invariant index by averaging the available quinquennialindicesatthe4-SIClevel. Thismeasureisonlyavailableforarestrictedsample ofmanufacturingfirms. Investmentinaction, smallinvestments, andinvestmentspikesaredefinedatthefirmlevel (cid:15) basedonCooperandHaltiwanger(2006)asthosefirm–yearobservationscorrespondingto Capex/book assets <.01, Capex/book assets .01, and Capex/book assets >.2, respecj j j j(cid:21) j j tively. Industry is 4-SIC. In each industry-year, we compute frequency as number of observations involving investment inaction (small investment) to total number of observations in the industry. This procedure results in a time-invariant cross-sectional ranking of 4-SIC industries. Time-series skewness and kurtosis of annual aggregate industry investment are based on (cid:15) Caballero (1999) and calculated as the skewness and kurtosis of average annual Capex to book assets ratios in each (4-SIC) industry. In every year, we calculate annual averages in eachindustryasindustry-yearmeansofindividualfirm-yearCapextobookassetratios.This procedureresultsinatime-invariantcross-sectionalrankingof4-SICindustries. Time-series standard deviation of aggregate industry operating costs is calculated after ag- (cid:15) gregatingfirm-leveloperatingcostsbytakingannualmeansatthe4-SICindustrylevel. For eachindustry,themeasureisthestandarddeviationoftheseannualindustrymeansofoperatingcosts. Operatingcostsarecostsofgoodsold(#41). Thismeasuregivesatime-invariant cross-sectionalrankingof4-SICindustries. 56

D Details of Regression Results in Panel A of Table 2 TableA-1: DetailsofRegressionAnalysisofIntangibleCapitalandCorporateFinancing WholeSample R&D>0Firmsonly Cash NetDebt Cash NetDebt OLS FE OLS FE OLS FE OLS FE (1) (2) (3) (4) (5) (6) (7) (8) IntangibleCapital 0.086 0.061 -0.111 -0.047 0.104 0.067 -0.125 -0.067 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Industrysigma 0.037 -0.007 -0.062 0.022 0.041 -0.001 -0.064 0.037 (0.000) (0.190) (0.000) (0.000) (0.000) (0.739) (0.000) (0.000) Market-to-book 0.023 0.011 -0.032 -0.010 0.028 0.016 -0.037 -0.009 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Firmsize -0.024 -0.039 0.025 0.011 -0.022 -0.032 0.005 0.067 (0.000) (0.001) (0.000) (0.000) (0.000) (0.001) (0.000) (0.000) Cashflow 0.031 0.017 -0.096 -0.091 0.037 0.023 -0.109 -0.108 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Capex -0.008 -0.012 0.017 0.005 -0.011 -0.011 0.018 0.002 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.038) Dividend -0.011 0.007 -0.037 -0.036 -0.023 0.005 -0.016 -0.033 (0.000) (0.000) (0.000) (0.000) (0.000) (0.007) (0.000) (0.000) Acquisitions -0.010 -0.008 0.034 0.018 -0.012 -0.009 0.036 0.023 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Yearfixedeffects Yes Yes Yes Yes Yes Yes Yes Yes AdjustedR2 0.299 0.665 0.235 0.602 0.340 0.689 0.240 0.601 The table reports detailed parameter estimates for the control variables included in the cash and netdebtregressionsreportedinPanelAofTable2. E Numerical Method Thenumericalprocedurecanbethoughtasaninner-outerloopiteration. 1. Initiate the procedure assuming a uniform distribution for the joint distribution of technology,capitalandnetdebt,denotebyµ0. 2. Wethenlookforamarketclearingwagew0 = w(µ0)thatsolves(29)usinganumericalroot finder. 3. Intheinnerloopproblem,anindividualfirmsolvesitsvaluemaximizationproblemtaking as given the stationary joint distribution, and hence the market clearing wage. This procedureupdatesthepolicyfunctions, g (K,B,Z;µ0)and g (K,B,Z;µ0). k b 4. We then move to the outer loop where we iterate (22) until convergence to obtain µ1. We then look for a new market clearing wage w1 = w(µ1) that solves (29) using a numerical rootfinder. 57

5. Checkifw1 = w0. Ifyes,stop. Ifno,gobacktostep3anditeratethestepsusingw1 asanew initialguess. Fortheinnerloopproblem,weuseanadaptivegridpointmethodforvaluefunctioniteration. We proceed in two stages. In the first stages, we specify 30 40 5 grid points in state dimension (cid:2) (cid:2) for (K,B,Z). We use Markov chain process to discretize the technology shock. In the choice dimension for (K ,B ), we use 300 300 grid points. We then iterate on Bellman equation to obtain 0 0 (cid:2) thefirststagevaluefunction. Wedenotetheoptimalchoiceofcapitalandnetdebtby(K ,B ). (cid:3) (cid:3) Inthesecondstage,weallocateallgridpointsfor(K ,B )in([K δ ,K +δ ] [B δ ,B + 0 0 (cid:3) K (cid:3) K (cid:3) B (cid:3) (cid:0) (cid:2) (cid:0) δ ]), i.e., in a rectangle around the optimal policies obtained in the first stage Bellman iteration. B This way we allocate all computing resources around the region where the optimal choices are mostly likely to exist. In each time we update the value function in the second stage, ([K (cid:3)0 (cid:0) δ0K ,K (cid:3) +δ0K ] (cid:2) [B (cid:3)0 (cid:0) δ0B ,B (cid:3)0 +δ0B ])isalsoupdatedwith δ0K ,δ0B < δ K ,δ B suchthatwefocusonfiner andfinergridpoints. Thisallowustoconstructafairlysmoothpolicyfunctions. Itisimportantto makesurethatforeachgridpointK instatedimension,(1 δ)K isalwaysincludedinthechoice (cid:0) dimensionsuchthattheprogramallowsforinactionexactly. For the outer problem, we use a fixed (50 50 5) grid points for µ(K,B,Z). This means that (cid:2) (cid:2) weneedtouseinterpolationtoevaluatepolicyfunctionsatpointsoffthegridpointsusedinthe valuefunctioniteration. Forspeed,weuselineartensorproductinterpolation. 58