CompetitionandtheRealEffectsofUncertainty* RajaPatnaik† LondonBusinessSchool November13,2015 JOBMARKETPAPER Abstract This paper investigates the impact of uncertainty on firm-level capital investment and examineswhetherthiseffectdependsonthedegreeofcompetitionthatfirmsface.Iexploit auniqueempiricalsettingtoconstructatime-varyinguncertaintymeasurethatisexogenous to economic conditions and firm behavior. I show that higher uncertainty results in a decreaseininvestmentforfirmsinmoreconcentratedindustries.Theeffectisstrongerfor firmsthatfacehighercostsassociatedwithreversinginvestments.Thisfindingisinlinewith irreversibleinvestmentmodelsthatpredictanegativerelationshipbetweenuncertaintyand investment. In contrast, firms in highly competitive industries increase investment in response to higher uncertainty, supporting the argument that competition can erode the option value of deferring investment. In that case, other industry and firm characteristics suchasoperationalflexibilitycanresultinincreasedinvestmentinresponsetoheightened uncertainty. I also find economically significant effects of uncertainty on other types of investmentsuchasR&Dspending,advertisingandinvestmentinhumancapital.Collectively, my results illustrate that the degree of competition plays an important role in the link betweenuncertaintyandinvestment. JELclassification:G31,G32 Keywords:Uncertainty;Investment;Competition;RealOptions;Irreversibility *Iamdeeplygratefultomyadvisor,FrancescaCornelli,andtoHowardKung,HenriServaesand StefanLewellenfortheirguidanceandinvaluablesupport.IalsowanttothankTaylorBegley,João Cocco,JamesDow,AlexEdmans,FranciscoGomes,ChristopherHennessy,BrandonJulio,RalphKoijen, Ryan Lewis, Anna Pavlova, Tarun Ramadorai, Hélène Rey, Rui Silva, Vikrant Vig and seminar participantsatLondonBusinessSchoolforhelpfulcommentsanddiscussion.Allerrorsaremyown. †LondonBusinessSchool,Regent’sPark,NW14SA,London.Email:rpatnaik@london.edu 1 1. Introduction Alargebodyofresearchintheirreversibleinvestmentliteraturepredictsanegativelink between uncertainty and investment (e.g., Dixit and Pindyck, 1994). If firms face capital adjustmentcostsandpartialinvestmentirreversibility,uncertaintycreatesanopportunity costofinvestingtodayintheformofapositiveoptionvalueofwaiting.Thisleadsfirmsto deferinvestmentuntilsomeoftheuncertaintyisresolved.Mostoftheempiricalevidenceis consistentwiththispredictionandindicatesthatcapitalinvestmentonaveragedeclinesin response to uncertainty (e.g., Bloom et al., 2007). However, this effect is likely not homogenous across firms in different competitive environments. If firms face competitive access to investment projects, the fear of preemption can render the option to defer investmentlessvaluable.Thus,competitivepressurecanimpedeafirm’sabilitytowaitand subsequently mitigate the negative effect of uncertainty predicted by the irreversible investmentliterature(Aguerrevere,2003;Grenadier,2002). Knowingwhetheruncertaintyhasauniformordifferentialeffectoncapitalinvestment acrossindustriesiscentraltoourunderstandingofhowuncertaintyaffectstherealeconomy. Surprisingly, the existing empirical evidence on the impact of uncertainty on investment acrossdifferentindustriesisverylimited.Twokeychallengeshavehinderedprogress.First, it is difficult to identify changes in uncertainty that occur independently from changes in economic conditions. Such endogeneity issues usually impede the causal estimation of the uncertainty effect on firm-level investment. Second, the lack of exogenous uncertainty measuresthatarerelevantforawidevarietyofindustrieshaveledstudiestofocusonsingle industriessuchasoilandgas(Kellogg,2014)ormanufacturing(Bloometal.,2007;Leahy andWhited,1996).However,suchanapproachmakesitdifficulttotestwhethertheeffectof uncertaintyonfirminvestmentdiffersacrossindustries. In this paper, I use a unique empirical setting to construct a time-varying measure of uncertaintythatisexogenoustoeconomicconditionsandfirmbehavior.Thisallowsmeto providenewempiricalevidenceontheuncertainty-investmentlinkanditsinteractionwith competition.Consistentwiththeoriesintheirreversibleinvestmentliterature,Ishowthat firmsinconcentratedindustriesdecreaseinvestmentinresponsetoheighteneduncertainty. However, I find that uncertainty leads to increased investment in highly competitive industries,suggestingthatcompetitioncanattenuatethenegativeeffectofuncertaintyand allowothermechanismsthatencourageinvestmentinthefaceofuncertaintytodominate. 2 Mycontributiontotheexistingliteratureistwo-fold.First,Iconstructanovelmeasureof uncertainty that is exogenous to firm investment decisions and economic conditions. Specifically,IexploituncertaintyaboutweatherchangesinthePacificassociatedwiththeEl NiñoSouthernOscillation(ENSO)cyclethataffectavarietyofindustriesintheU.S.Industries suchascommodity-producingsectorscanbedirectlyaffectedbychangesinweather,while otherindustriesareimpactedindirectlythroughtherelationbetweencommoditypricesand changesinENSOconditions1.AgrowingliteraturedocumentstheimpactofENSOeventssuch asElNiñoweathershocksoneconomicactivityandcommodityprices(e.g.,Brunner,2002; Cashin et al., 2014). While this literature generally examines the macroeconomic effect of observedENSOshocks,tothebestofmyknowledge,thispaperisthefirststudytousefuture ENSOuncertaintyasaninstrumentforeconomicuncertainty. My empirical strategy is based on the following instrumental variables approach. Consistent with existing studies, I use stock price volatility as a proxy for firm-level uncertainty (e.g., Bloom et al., 2007). Due to the endogenous nature of this measure, I constructaninstrument,whichcapturesindustry-relevantENSOuncertaintythatisplausibly exogenous to economic conditions and firm behavior. The instrument is a combination of industry-level sensitivities to changes in ENSO conditions, calculated using share price returns, and ENSO forecast data 2. This identification strategy relies on the fact that ENSO uncertainty varies across time and ENSO sensitivities vary substantially across industries. IndustriescanhavenegativeandpositiveexposurestoENSOshockssuchasElNiñoepisodes. Thus,anElNiñoeventresultsinachangeineconomicconditions(i.e.,achangeinthefirst moment) that is favorable for industries with positive sensitivities and unfavorable for industries with negative sensitivities. However, an increase in uncertainty about the occurrence of an El Niño (i.e., a change in the second moment) represents an increase in uncertaintyforbothtypesofindustriesandthisincreaseishigherforindustrieswithlarger ENSO sensitivities. Hence, a first moment shock has opposing effects on industries with positiveandnegativesensitivities,whileasecondmomentshockaffectstheseindustriesin the same direction, and the latter effect only varies with the magnitudes of their ENSO 1ChangesinENSOconditionscanbemeasuredusingseasurfacetemperaturedatafromthePacific Ocean.Large-scaleENSOeventssuchasElNiñoepisodesareusuallyassociatedwithincreasesinsea surfacetemperaturethatcausemajorclimatologicalchangesaroundtheworld. 2Asdescribedinsubsequentsections,Iobtaindataontheprobabilitydistributionsof12-month ENSO forecasts from the Climate Prediction Center. These forecasts generally follow normal distributionsandcanthusbecharacterizedbytheirfirstandsecondmoments.Iwillspecificallyfocus onthesecondmomentoftheseforecastprobabilitydistributionstoconstructmyinstrument. 3 sensitivities.Thisintuitionisusedtoconstructtheinstrumentandtodisentanglefirstfrom secondmomenteffects3. My second contribution is that I provide new evidence for a differential effect of uncertainty on firm-level investment across industries. I examine a sample of more than 6,500publiclylistedU.S.firmsfrom1985to2014anddocumentthatcompetitionisamajor determinantofthesignandmagnitudeoftheuncertainty-investmentrelationship.Iusethree different proxies for industry concentration and show that firms in more concentrated industriesdecreaseinvestmentinresponsetohigheruncertainty.Incontrast,firmsinhighly competitiveindustriesincreaseinvestmentwhenuncertaintyrises. Totestwhetherthenegativerelationinconcentratedindustriesisconsistentwithmodels in the irreversible investment literature, I employ two different proxies for the degree of investmentirreversibilitythatfirmsface,whicharebasedoninsightsfromKimandKung (2014) and Zhang (2005). The negative effect is stronger for firms with less redeployable assets and higher average adjustment costs, suggesting that the option to wait and defer investmentismorevaluableforthem.Hence,thenegativeeffectofuncertaintyoninvestment in concentrated industries is consistent with the predictions of real option models in the irreversibleinvestmentliterature(e.g.,DixitandPindyck,1994). Ialsoinvestigatealternativeexplanationsforthenegativeuncertaintyeffectobservedin concentratedindustries.Forinstance,oneexplanationreliesontheargumentthatgreater uncertaintymayincreaseborrowingcostsforfirmsinthepresenceoffinancialconstraints. Gilchristetal.(2014)arguethatuncertaintyaffectsinvestmentindirectlythroughincreases in credit spreads. As a result, firms reduce leverage and decrease investment spending. Examining external financing decisions of the firms in my sample, I find no evidence supportingthisprediction:forfirmsinmoreconcentratedindustries,Idocumentanegative relationshipbetweenuncertaintyandnetequityissueandnoimpactofuncertaintyonnet debt issue 4 . I also test whether firms offset the decrease in capital investment with correspondingincreasesininvestmentinmoreflexibleproductionfactorssuchasworking capitalandlabor(EberlyandVanMieghem,1997;Fischer,2013).Theresultsdonotprovide 3ThiseconometricapproachissimilartotheframeworkinSteinandStone(2013).Theyemploy exposurestooilpriceandexchangeratefluctuationsasinstruments,butdonotaddresstheirpossible endogeneity. My paper employs a distinctly different instrument that is based on changes in ENSO uncertainty,whichareplausiblyexogenoustoinvestmentdynamicsandeconomicconditions. 4The overall debt-to-equity ratio increases in response to uncertainty for firms in concentrated industries. 4 evidenceforsuchashiftininvestments.Uncertaintysimilarlyleadstodecreasesinworking capitalandhiringinmoreconcentratedindustries. Inhighlycompetitiveindustries,Ifindthatfirms,onaverage,increaseinvestmentinthe faceofhigheruncertainty.Thisresultisconsistentwiththeargumentthatcompetitioncan erodetheoptionvalueofdeferringinvestment,allowingothermechanismsthatencourage investmentinresponsetoheighteneduncertaintytodominate.Forinstance,uncertaintycan activelyencourageinvestmentwhenfirmshavehighoperationalflexibility(Mills,1984).If firmscaneasilyexpandtoexploitgoodtimesandcontracttoinsureagainstbadoutcomes, their profits are convex in input or output prices. A mean-preserving increase in cost or demand uncertainty, by Jensen’s inequality, thus increases expected profits 5 . Similar convexitiescanariseundersubstitutabilityofcapitalwithmoreflexibleproductionfactors suchaslabor(Abel,1983;Hartman,1972).LeeandShin(2000)showthattheconvexityin thefirm’sobjectivefunctionincreaseswiththeshareoftheflexibleproductionfactorinthe production technology. Thus, the positive effect of uncertainty on investment should be stronger for firms with higher labor-capital ratios. I test whether the positive effect of uncertaintyinhighlycompetitiveindustriesresultsfromsuchconvexitiesinafirm’sprofit function. Consistent with the above predictions, the positive effect of uncertainty is only statistically significant for firms with higher operational flexibility (e.g., firms in nonunionized industries) and for firms with higher labor-capital ratios where the flexible productionfactorrepresentsalargershareoftheproductiontechnology. The estimated effects of uncertainty are economically significant. In response to a ten percentage point increase in uncertainty, firms in more concentrated industries decrease investmentby10percent.Conversely,firmsinhighlycompetitiveindustriesraiseinvestment ratesbyabout11percentinresponsetoasimilarincreaseinuncertainty. The finding that competition plays a central role in the uncertainty-investment relationshipisrobusttoarangeofadditionaltests6.Specifically,Iprovideevidencethatthe differentialimpactofuncertaintyacrossindustrieswithvaryingdegreesofcompetitionisnot 5Themarginalprofitofcapitalineachperiodtcanbewrittenasmax[0,(p –c )].Ifthemarginal t t unitofcapitaldoesnothavetobeutilizedinbadtimes,aunitofcapitalcanbeseenasanoptionon futureproduction,whichismorevaluablethehigherthevarianceofptorct(Pindyck,1993a). 6 In robustness tests, I address the concern that my instrument might capture systematic differencesinthelevelofcompetitionacrossindustries,whichcouldresultintheobserveddifferential effectofuncertaintyoninvestmentincompetitiveandconcentratedindustries.Ialsodiscussindetail theconcernthatENSOuncertaintycouldaffectthedegreeofcompetitioninanindustry,whichinturn might impact investment, and why it is unlikely that such a causal chain would yield the results reportedinthispaper. 5 auniquefeatureoftheparticularidentificationstrategyanduncertaintymeasureusedinthis paper.Iemploymoreconventionalandaggregateproxiesforuncertaintysuchastheimplied volatilityofS&P100optionstoshowthatcompetitivepressurecanattenuatethenegative effectofuncertainty.Whilepossibleendogeneityconcernsmakeitdifficulttodrawinference about the sign and the magnitude of the uncertainty-investment relationship in such a specification, the results confirm that the level of competition firms face impacts their investmentstrategiesinresponsetoheighteneduncertainty. Whilemyfindingsprovideclearevidencethatfirmsindifferentcompetitiveenvironments respond differently to uncertainty, the economic effect of uncertainty depends on the persistenceofthesefirm-levelresponses.Iinvestigatethedynamicsofinvestmentfollowing anuncertaintyshockinadifference-in-differencesframework.Theresultsshowasignificant wedge in the response of investment spending by firms in competitive and concentrated industriesafterbothalargeincreaseandalargedecreaseinuncertainty.Thiswedgeonly persistsfortwoperiodsafteranincreaseinuncertaintyandforoneperiodafteradecrease inuncertainty.ThesefindingsareconsistentwithsimulationsinBloometal.(2014)showing that the negative effect of uncertainty on investment, prescribed by the real options literature,generallyresultsinashortandsharpdropininvestmentratesfortwoperiods. This is followed by a quick rebound of investment rates and a prolonged recovery of the capitalstocktoitssteadystate. Finally, I document the effect of uncertainty on other types of investment. Uncertainty reducesR&DspendinginconcentratedindustriesandhasnoeffectonR&Dinvestmentfor firmsincompetitiveindustries.ThemagnitudeofthenegativeeffectonR&Dissignificantly smallercomparedtothenegativeeffectofuncertaintyoncapitalinvestmentinconcentrated industries.ThisisconsistentwiththefactthatR&Dspendingismorepersistentacrosstime thancapitalinvestment.Inlinewithdecreasesincapitalinvestment,firmsinconcentrated industriesalsospendlessonadvertising.Iobservenoeffectonadvertisingforfirmsinhighly competitive industries. Hiring on average declines across firms in response to uncertainty indicating that uncertainty has a real impact on labor markets. However, this effect is not significantforfirmsinhighlycompetitiveindustries. 6 1.1 RelatedLiterature 1.1.1 Theory This paper builds on a large theoretical and empirical literature on investment under uncertainty7.Theoreticalresearchhasexaminedmanydifferentsettingsinwhichuncertainty mattersforfirminvestmentdecisions.Byfarthelargeststrandofliteratureemphasizesthe realoptionsframeworktoidentifytheimpactofuncertaintyunderinvestmentirreversibility (Abel and Eberly, 1994; Bernanke, 1983; Brennan and Schwartz, 1985; Dixit and Pindyck, 1994;McDonaldandSiegel,1987).Thecentralideaunderlyingthisargumentisbasedonthe presenceofasymmetricadjustmentcosts,whichmakeitcostliertoreverseinvestmentthan toincreaseit,effectivelyleadingtopartialorfullirreversibilityofinvestment.Undersuch conditions, a mean-preserving rise in uncertainty increases a firm’s option to defer investment until some of the uncertainty is resolved. This creates regions of inaction that expand with uncertainty, leading to a negative link between uncertainty and investment. Negative effects of uncertainty also arise out of managerial risk aversion and ambiguity aversion(IlutandSchneider,2012;PanousiandPapanikolaou,2012).Higheruncertaintycan leadtoincreaseddefaultprobabilitiesandtoinvestorsrequiringhigherriskpremia,which in turn should increase borrowing costs (Bloom, 2014). Several models show that, in the presence of financial constraints, uncertainty can raise the cost of finance for firms and decreaseinvestmentspending(Arellanoetal.,2010;Christianoetal.,2014;Gilchristetal., 2014). Incontrast,othermodelsconsiderconvexitiesinthefirm’sprofitfunctionandpredicta positivelinkbetweenuncertaintyandinvestment(Abel,1983;Hartman,1972;Oi,1961).The basicintuitionbehindtheirresultsisthatuncertaintyraisesthevalueofthemarginalunitof capitalifthefutureprofitabilityofthemarginalunitisaconvexfunctionoftheunderlying stochastic variable such as input or output prices. Certain conditions such as constantreturns-to-scale and complementary production factors achieve such convexity in the marginal revenue product of capital. Similar convexities can also arise due to operational flexibilityandafirm’sabilitytoeasilyvaryoutputaccordingtoeconomicconditions(Mills, 1984; Pindyck, 1993a). In addition, investment lags can lead to a positive link between uncertainty and investment. Many models of irreversible investment assume that investmentsbecomeproductiveassoonasthedecisionismadeandcostsareincurred.In 7ForacomprehensivereviewofthisliteratureseeBloom(2014). 7 fact,manyinvestmentstaketimetocomplete,whichisoftenreferredtoas“time-to-build”or investmentlags8.Bar-IlanandStrange(1996)illustratethatinthepresenceofinvestment lags, the negative effect of price uncertainty on investment predicted by the irreversible investmentliteraturecanbemitigatedorevenreversed.Pindyck(1993b)alsosuggeststhat uncertaintyaboutthetechnicalfeasibilityofaprojectcanincreaseinvestment. Many of the above models rely on specific assumptions about the competitive environmentthatfirmsface.Inparticular,theliteratureonirreversibleinvestmentgenerally assumesthatagentsmakeoptimalinvestmentdecisionsinisolation.Grenadier(2002)takes intoaccountstrategicinteractionsamongfirmsandshowsthatincreasedcompetitioncan erode the option value to defer investment, leading firms to revert to net present value investing strategies. Similarly, Caballero (1991) highlights the lack of robustness of the negativelinkbetweenuncertaintyandinvestmentunderirreversibleinvestmenttochanges inthelevelofcompetition.Otherresearchsimilarlystudiestheeffectsofcompetitiononthe uncertainty-investment relationship (Aguerrevere, 2003; Caballero and Pindyck, 1996; Kogan, 2001; Leahy, 1993; Williams, 1993). Kulatilaka and Perotti (1998) further the argument that strategic competition can encourage investment and illustrate that in the presenceofgrowthoptions,strategicconsiderationstodiscourageentryorinvestmentsof competitorscanprovidestrongincentivesforincreasedinvestment. 1.1.2 EmpiricalEvidence Theambiguouspredictionsofthetheoreticalinvestmentliterature,particularlyintheface ofvaryinglevelsofcompetition,showcasestheneedforempiricalanalysistodeterminethe sign and the magnitude of the relationship between uncertainty and investment. The empiricalliteratureoninvestmentunderuncertaintycanbroadlybeclassifiedintomacro- andmicro-levelstudies.Themacroeconomicapproachconsistsofinvestigatingtheeffectof uncertaintymainlyproxiedforbythevolatilityofmacroeconomicvariables(suchasinflation rates,exchangerates,oilpricesandstockmarketreturns)onaggregatecapitalinvestment (Eisfeldt and Rampini, 2006; Ferderer, 1993; Fernández-Villaverde et al., 2011; Goldberg, 1993; Huizinga, 1993; Servén, 2003). The general consensus in this literature is that uncertainty,onaverage,reduceseconomy-widecapitalinvestment. 8MacRae(1989)observesthatprojectsintheelectricitysectorcantakeupto10yearstocomplete. Similarly,Pindyck(1991)notesthatinvestmentsintheaerospaceandpharmaceuticalindustrieshave comparabletimeframes. 8 Onthemicro-level,papershaveusedindustry-andfirm-specificdatatoprovideevidence on the link between investment and uncertainty. To a large extent, this literature follows macroeconomicstudiesintheconstructionofuncertaintymeasuressuchasthevolatilityof stockreturns(Baumetal.,2008;Bloometal.,2007;Bulan,2005;Gilchristetal.,2014;Leahy and Whited, 1996). Additional measures include the variation in output and input prices (Ghosal and Loungani, 1996; Kellogg, 2014), currency fluctuations (Campa and Goldberg, 1995;Campa,1993)andproxiesforfirmprofitability(GhosalandLoungani,2000;Minton and Schrand, 1999). A common challenge in the empirical literature on investment under uncertaintyistodistinguishtheeffectofuncertaintyfromtheimpactofbusinessconditions. Tomitigateendogeneityconcernsaboutuncertaintyproxiesbasedonreturnvolatility,many of the above studies use “internal” instruments in the spirit of Arellano and Bond (1991), which require strong identifying assumptions about the properties of the underlying investment time series 9 . Stein and Stone (2013) try to address this issue by following a differentidentificationstrategy.Theyuseindustryexposurestooilpriceandexchangerate fluctuationsasinstrumentsforfirm-leveluncertaintymeasuredbyimpliedoptionvolatilities. While their approach arguably presents an improvement to previous methods, it does not accountfortheexistenceoflatentvariablesthatmightaffectinvestmentandexchangerates oroilpricesatthesametime.Anotherapproachmodelschangesinuncertaintybyexploiting naturalexperimentssuchaspoliticalregimechanges,terroristattacks,anddisasters(Baker and Bloom, 2013; Julio and Yook, 2012; Kim and Kung, 2014). The challenge in these frameworks is that these events might impact firm behavior directly rather than through changes in uncertainty. While many of the above studies document suggestive empirical supportforanegativeeffectofuncertainty,resultsareoftenverysensitivetotheinclusionof first-momentcontrols(LeahyandWhited,1996).Thus,inconclusivemicro-levelevidenceon the relationship between capital investment and uncertainty reflects the ambiguity in theoretical predictions particularly in the cross-section 10 . As highlighted earlier, firm and industrycharacteristicssuchasthecompetitiveenvironmentcanhaveasubstantialimpact onthelinkbetweenuncertaintyandinvestment.Surprisinglyfewstudieshaveexaminedthe differentialeffectofuncertaintyacrossindustries.Bulan(2005)andGuisoandParigi(1999) 9SeeAlmeidaetal.(2010)andEricksonandWhited(2012)forcomprehensivediscussionsofthe advantages and disadvantages of using GMM versus OLS type instrumental variables estimators in investmentregressions. 10 Abel and Eberly (1996) and Leahy and Whited (1996) provide in-depth discussions of the possiblepositiveandnegativeeffectsofuncertaintyoncapitalinvestment. 9 examine a sample of manufacturing firms in the U.S. and Italy, respectively, and provide evidence that increased competition can dampen the negative effect of uncertainty. In contrast,GhosalandLoungani(1996)findtheoppositeresultusingindustry-leveldata.The competing nature of these results might arise from studying very specific industries and highlightstheneedforempiricalanalysisofabroadsampleoffirmsinavarietyofdifferent industries. Inthispaper,Iwillconstructanexogenous,time-varyingandforward-lookinguncertainty measuretoinvestigatethedifferentialimpactofuncertaintyoncapitalinvestmentformore than6,500U.S.firmsacrossalargenumberofindustries. Theremainderofthepaperisorganizedasfollows:Section2providesdetailsaboutthe El Niño Southern Oscillation cycle and discusses its impact on economic activity.The data sources are described in Section 3. Section 4 outlines the empirical framework and methodology used. Section 5 presents the main empirical results, while Section 6 covers cross-sectional tests aimed at exploring the economic channels underlying the observed effects.Section7and8investigatethedynamicsoftheuncertaintyeffectandtheimpactof uncertaintyonothertypesofinvestment.Finally,Section8concludesthepaper. 2. ElNiñoSouthernOscillationCycle(ENSO) I examine the effect of uncertainty on firm-level investment by exploiting exogenous variationinuncertaintyinducedbytheElNiñoSouthernOscillation(ENSO)cycle.ENSOisa prominent weather system in the tropical Pacific that heavily influences atmospheric conditions in North and South America as well as Asia and Australia (Adams et al., 1995; Changnon,1999;RopelewskiandHalpert,1987,1986;Rosenzweigetal.,2001;Wangetal., 2000). “Regular” seasons are characterized by a persistent high-pressure system off the west coastofSouthAmerica,specificallyoverthecoastofPeru.Atthesametime,alow-pressure system forms east of Australia. The resulting differential in atmospheric pressure leads to prevailingsurfacewindsthatblowfromtheeasttothewest–oftenreferredtoas“easterlies”. These trade winds carry warm surface water from the central and eastern regions of the equatorialPacifictoAustraliaandAsia.Resultingprecipitationintheseregionsisextensively usedforagriculturalandindustrialpurposes.AlongthewestcoastofSouthAmerica,coldand nutrient-richwaterrisestothesurfaceboostingthefishingindustryintheseregions. 10 Thisweatherpatternisverysensitivetoanomalousdeviationsinatmosphericpressure and sea surface temperature, which can cause serious disruptions to weather conditions aroundthePacific.Occasionally,thehigh-andthelow-pressuresystemscanswitchpositions causingtheeasterliestosubside.ThisphenomenonisoftenreferredtoasElNiño.LaNiña events,ontheotherhand,arecharacterizedbyintensifiedpressuresystemsleadingtoadrop in ocean temperatures and intensification of the trade winds. This periodical interaction between atmospheric conditions and the Pacific Ocean is called the El Niño Southern Oscillation(ENSO)cycle.WhileENSOeventsoccurinthePacificOcean,theycancausewidereachingchangesinglobalweatherpatternsthataffectavarietyofdifferentindustriessuch ascommodity-producingsectors.Asanillustration,pastElNiñoepisodeshaveresultedin extensive rainfall in South America hampering rice cultivation in Ecuador and flooding copperminesinPeruandChile.Suchdirecteffectshavealsobeenobservedinlargepartsof AustraliaandIndonesiawheredroughtconditionsassociatedwithElNiñoeventsresultedin lower agricultural output and forest fires. In addition, ENSO has indirect effects on commodityproductionandtransportation.ENSOrelateddroughtepisodesinIndonesia,the world’slargestnickelproducer,havehistoricallycrippledtheminingindustrywhichrelies heavilyonwaterwaysfortransportationandhydroelectricpower.IntheU.S.,ElNiñoevents havefrequentlybeenassociatedwithunusuallywarmwintersboostingprofitsinindustries engaged in and related to residential construction. Warmer than usual periods have historicallyalsoledtochangesinenergypricesduetodecreasedheatingdemand(Brunner, 2002;Changnon,1999). ThisanecdotalevidenceontheeconomiceffectsofENSOvariationhassparkedagrowing body of research on the link between ENSO and economic activity 11 . Brunner (2002) documentstheeffectofENSOvariationonworldrealnon-oilprimarycommoditypricesas well as GDP growth in G-7 countries. The study shows that ENSO variability can explain almost 20% of movements in commodity prices – in particular, prices of metals and agriculturalproducts.Importantly,whiletheENSOeffectoncommoditypricesissignificant in the short run, it is also pronounced over longer horizons 12 . Similar results have been reported in a recent study by Cashin et al. (2015). They show that ENSO related weather 11Forcomprehensivereviewsoftheliteratureonthemoregeneralrelationshipbetweenclimate andeconomicperformanceseeDelletal.(2014)andTol(2009). 12 The author examines both four-quarter- and 16-quarter-ahead forecast error variances in commoditypricesandfindsthatENSOactivityexplainsalmost10%ofcommoditypricemovements overtheshorterhorizonandupto20%overthelongerhorizon. 11 eventshaveheterogeneouseffectsoneconomicgrowthandinflationacrosscountriesaround theworld.Inaddition,theydocumentthatENSOhasasignificantimpactonenergyandnonfuel commodity prices. Other research finds similar links between ENSO and economic activityaswellascommodityprices(Adamsetal.,1999;Algieri,2014;BerryandOkuliczKozaryn,2008;Chenetal.,2001;Chimelietal.,2008;Chuetal.,2012;HandlerandHandler, 1983;Hansenetal.,1998;Hsiangetal.,2011;Nayloretal.,2001;Solowetal.,1998;Ubilava, 2012). Inthispaper,Iwillbuildontheexistingresultsthatestablishalinkbetweeneconomic activity and ENSO fluctuations. Previous studies focus specifically on the macroeconomic effects of ENSO shocks such as El Niño events, which represent first moment shocks to economicactivity.Incontrast,tothebestofmyknowledge,thisisthefirstpapertoutilize ENSO prediction data to study the impact of variation in future ENSO uncertainty, representingsecondmomentshocks,onrealeconomicoutcomes.Iwillexploitthefactthat firmshavedifferentsensitivitiestochangesinENSOconditions13andconstructaforwardlookinguncertaintymeasureusingtimeseriesdataonENSOpredictions.PleaseseeAppendix BformoredetailsontheElNiñoSouthernOscillationCycle. 3. Data Formyempiricalanalysis,IexaminealargesampleofU.S.firmsintheperiodfrom1985 to2014.Thefollowingsectionprovidesabriefoverviewofthedatasources. 3.1 Firm-andIndustry-LevelData ThefocusofthispaperisonU.S.firmsandIobtainfirm-levelfinancialinformationfrom theCompustatdatabase.Databasedoncashflowstatements,incomestatementsandbalance sheets is available for fiscal years from 1950 to 2014. Daily stock price information is obtainedfromtheCRSPdatabase. I collect industry-level data on union memberships from the Union Membership and Coverage database (www.unionstats.com), described by Hirsch and Macpherson (2003). 13 The variation in ENSO exposure across industries is frequently being highlighted in popular media.ArecentmarketreportbyMacquariestatesthat“companiesinNorthAmericaandAustralia/NZ aremostlikelytobeimpacted,followedbycompaniesinAsia.Importantly,theimpactoncompanies is not uniform within any given region. In Aus/NZ, 38% of highlighted companies are likely to be negatively impacted with 62% expected to be positively impacted. In North America the negative/positivesplitis45/55andinAsiathemixis70/30”(FT,2015). 12 Since this data uses the Census Industry Classification (CIC) to categorize industries, I use matching information provided by the U.S. Census Bureau to map CIC codes to Standard Industry Classification (SIC) and North American Industry Classification System (NAICS) codes.ThisallowsmetomatchtheunionmembershipwithCompustatandCRSPdata. 3.2 ENSOData The intensity of an ENSO event can be measured in various ways. One of the most commonly used measurements relies on sea surface temperature (SST) anomalies in the Pacific.Theseanomaliesrepresentdeviationsofseasurfacetemperatureinagivenregion fromitshistoricalaverage.TheNationalOceanicandAtmosphericAdministration(NOAA) providesthehistoricalmeasurementsofSSTanomaliesforseveralregionsinthePacific.This paperusesSSTdataspecificallyforregion“Niño3.4”forseveralreasons.First,itisthemost centralregioninthePacificforwhichSSTanomaliesarebeingrecorded.Second,theClimate PredictionCenter(CPC)usesdatarecordedinthisregiontoforecastENSOepisodes.Finally, thismeasurementisthefocusofseveralpreviousstudiessuchasBrunner(2002)andChuet al.(2012).Whilethetimeseriesgoesbackasfarasthelate1800s,consistentmeasurements areonlyavailablefrom1950tothepresent.Thedatacanbedownloadedfromthewebsiteof the Climate Prediction Center (CPC, 2015a). Figure 1 illustrates the time series of SST anomaliesinthecentralPacificregion“Niño3.4”.Largepositivevaluesareusuallyassociated withunusuallywarmElNiñoperiods,whilenegativevaluesindicatecolderLaNiñaevents. InadditiontohistoricaldataofobservedENSOevents,Iobtainprobabilisticconsolidation forecastsofSSTforregion“Niño3.4”fromtheCPC14.Theseforecastsrepresentastatistical combinationofvariousmodelsusedbytheCPCtopredictENSOanomaliesthroughoutthe Pacific15.ThedatacontainsmonthlySSTforecastsfrom1985tothepresentforhorizonsof uptotwelvemonths.InadditiontopointestimatesofSST,theentireprobabilitydistribution for these forecasts is provided. The first and second moments of the twelve-month SST forecastdistributions,whichgenerallyarenormaldistributions,aredepictedinFigure2. CompustatandCRSPdataismergedwithENSOpredictiondata,whichlimitsthesample periodto1985to2014.Thefinalsampleincludes12,017firmsand92,572observations. 14IthanktheCPCandNOAAforsharingthisforecastdatauponrequest. 15ThemodelsusedbytheCPCtoforecastENSOeventsincludecanonicalcorrelationanalysis(CCA), Markovmodels,ConstructedAnalogmodelsandfullycoupled,one-tierocean-atmospheredynamical models.TheCPCprovidesdetailedinformationabouttheconstructionofthesemodels(CPC,2015a). 13 4. EmpiricalFramework I construct a panel data set of U.S. firms from the period between 1985 and 2014 to investigatetherelationshipbetweenuncertaintyandinvestment.Myfocusonfirm-levelas opposed to more aggregated data is motivated by several factors. First, it allows me to investigate the differential effect of uncertainty across firms with different firm- and industry-levelcharacteristics.Second,Icanaccountfortheeffectofmacroeconomicshocks andthebusinesscyclethroughtimefixedeffects. The following sections describe the empirical strategy in more detail and discuss the centralquestionofhowImeasureuncertainty,andmyestimationmethodology. 4.1 EstimatingtheEffectofUncertainty:AnIVApproach Theexistingempiricalliteratureoninvestmentunderuncertaintytakesmanydifferent approaches to measuring uncertainty. While some studies build structural models, most researchrelieseitheroneventsthatplausiblycreateunexpectedspikesinuncertainty16or on time-series measures of aggregate and idiosyncratic uncertainty. The former research design works well if uncertainty shocks are indeed unexpected, but is problematic if the occurrenceofeventsiscorrelatedwithotherunobservedshocks.Studiesthatfallintothe secondcategoryfrequentlyuseuncertaintymeasuresbasedonvolatilityindicessuchasthe VIXfromtheChicagoBoardofOptionsExchangeoronfirm-levelstockreturnvolatility(e.g., Bloometal.,2007;LeahyandWhited,1996),whicharegenerallyrathernoisyproxiesfor uncertainty.Changesinstockreturnsoftenreflectchangesinthefirstmomentofinvestment opportunities rather than changes in uncertainty. In fact, a growing number of studies documentthecounter-cyclicalnatureofreturnvolatility(Bloom,2009;Campbelletal.,2001; Storeslettenetal.,2004). In this paper, I take a different approach. To estimate the effect of uncertainty on firm investment,Iconsiderasimplelinearinvestmentmodelofthefollowingform !",$ %",$&' = )* + ,- + ./ 01,2 + 34 ∙ 6*,- + 7*,- (1) where )* represents firm fixed effects and ,- denotes year fixed effects. The dependent variableiscapitalexpenditure,8*,- ,scaledbythepreviousperiod’scapitalstock,9*,-:4 .01,2 is 16 Some studies exploit uncertainty shocks caused by terrorist attacks, military action, natural disastersorpoliticalshocks(BakerandBloom,2013;JulioandYook,2012;KimandKung,2014). 14 avectorofrelevantcontrolvariables(includingTobin’sq)and6*,- denotestheuncertainty measure. Causal identification of the uncertainty effect requires 6*,- to be exogenously determined.Thelackofsuchanexogenousmeasureleadsmuchoftheempiricalliterature on investment under uncertainty to use realized stock return volatilities to proxy for uncertainty.Duetotheendogenousnatureofthismeasure,moststudiesrelyon“internal” instrumentssuchaslaggedlevelsanddifferencesoftheexplanatoryanddependentvariables (Bloom et al., 2007; Bulan, 2005; Leahy and Whited, 1996). However, the validity of this methodologyreliesonstrongassumptionsaboutthepropertiesoftheunderlyingdata,which arenotrequiredfortheestimationinmypaper. MyidentificationstrategybuildsonthemethodologyusedinSteinandStone(2013)and intheliteratureoninvestmentundereconomicpolicyuncertainty(e.g.,BrogaardandDetzel, 2015).Inlinewiththeexistingempiricalliterature,Iemployfirm-levelstockreturnvolatility as a proxy for firm-level uncertainty. However, I address concerns about the possible endogenous nature of such a measure by constructing a novel instrument based on ENSO uncertaintytoidentifyexogenousvariationinstockreturnvolatility.Iestimatetheeffectof uncertaintyonfirminvestmentintwosteps. 4.1.1 Industry-LevelSensitivitiestoENSO First,IidentifyindustriesthatareaffectedbychangesinENSOconditions.TheClimate PredictionCenter(CPC)providesmonthlymeasurementsofENSOintensityintermsofsea surface temperature (SST) anomalies (see Figure 1) 17. Positive deviations from long-term regional means are usually associated with El Niño episodes, while negative deviations indicate La Niña events. The climatology literature has established a strong association betweenpositiveornegativeSSTanomaliesandchangesinweatherconditionsinthePacific region (Adams et al., 1995; Ropelewski and Halpert, 1986; Wang et al., 2000)18. Thus, the literatureontheeconomiceffectsofENSOfrequentlyusesthetimeseriesofSSTanomalies to document the link between El Niño events, which represent the largest disruptions to regular weather patterns, and economic activity (see discussion in Section 2). Figure 1 illustratesthemonthlytimeseriesvariationinSSTanomalies(indegreeCelsius)from1980 to 2014. The positive spikes in temperature indicate El Niño episodes. For instance, the 17TheNationalOceanicandAtmosphericAdministration(NOAA)definesanomalyasadeviationof theseasurfacetemperatureinagivenregionfromitshistoricalaverage.Thismeasurewasinitially developedinRasmussonandCarpenter(1982)andRopelewskiandHalpert(1987). 18SeeAppendixBformoredetailsontheglobaleffectsofENSOonweather. 15 positivespikeinthetimeseriesaround1997and1998reflectsthestrongElNiñoeventat thetimethatinducedaprolongedperiodofdroughtinpartsoftheU.S.andledtoenormous lossestoagriculturalindustries(WolterandTimlin,1998). I use return data for firms in the CRSP-Compustat merged sample and monthly SST measurements to identify industries whose operations are affected by changes in ENSO conditions. To this end, I estimate the following regression for each year t (from 1985 to 2014)usingmonthlyreturndatafromthepreviousfiveyears(t-6tot-1) CD?E CD?E ;*,< = =>* + =* ∙ ;<?@A>> + =B,∙ F< + 7*,< (2) where;*,< isthemonthlystockreturnforfirmi,;<?@A>> isthemonthlyreturnontheS&P500 CD?E CD?E index andF< are the monthly observations of SST anomalies. The coefficient=B,in each year t is constrained to be the same for all firms in industry k (based on 3-digit SIC CD?E codes19).Ineachyeart,=B,representstheaveragesensitivityofstockreturnsinindustry k to anomalous ENSO activity over the previous five years. In other words, industry sensitivitiesaretime-varyingandforanygivenyearttheindustry-specificsensitivitiesare calculatedusingmonthlyreturnsandSSTdatafromthepreviousfiveyears.Giventhelong sampleperiod(from1985to2014),thismethodologyaccountsforthefactthatindustries andtheirexposurestoENSOmightchangeovertime.Icalculateindustry-levelinsteadoffirm level sensitivities due to several reasons. First, the panel data set is highly unbalanced. In addition,thismethodologymitigatesconcernsabouttheendogeneityofENSOsensitivities withrespecttofirminvestment.Somefirmsmightchoosecertaininvestmentstoreducetheir exposuretoENSO.However,itishighlyunlikelythatallfirmsinthesameindustryengagein thesamediversificationeffortsthroughinvestment20. For my subsequent analysis, I identify industries for which the estimated sensitivity CD?E coefficient,=B,,isstatisticallysignificantfromzero(atthe5%level).Firmsinindustries thatarenotsignificantlyaffectedbyENSOwillbedroppedfrommysample,sinceIdonot expectthesefirmstoreacttoENSOuncertainty21. 19Theresultspresentedthroughoutthispaperarerobusttobasingtheindustryclassificationon 2-digitSICcodes. 20InAppendixD,Ishowthatthemainresultsofthepaperarerobusttocalculatingtimeinvariant ENSOsensitivitiesusingreturndatainyears1980to1984,whichispriortomymainestimationperiod of1985to2014. 21Someindustrieshavelargesensitivitycoefficientswithsimilarlylargestandarderrorsrendering thecoefficientestimatenotstatisticallysignificantfromzero.Includingsuchobservationsthroughout 16 4.1.2 InstrumentalVariablesEstimation Tomitigateanyendogeneityconcernswithrespecttostockreturnvolatility,Iestimate model(1)usinganinstrumentalvariablesapproach.Iconstructaninstrumentthatidentifies exogenous variation in firm-level stock returns by combining the estimated ENSO CD?E sensitivities,=B,,withENSOpredictiondata. The CPC uses various statistical models to provide forecasts of ENSO anomalies for horizonsofuptotwelvemonths22.TheseforecastsusetrendsinmonthlySSTmeasurements topredictfutureENSOactivitymakingtheunderlyingstochasticprocessplausiblyexogenous to economic conditions. The prediction data obtained from the CPC includes information about the probability distribution of these forecasts including its mean and standard deviation. Since the forecasts are generally normally distributed, the first and second momentsaresufficientstatisticstocharacterizetheforecastprobabilitydistributions.Figure 2 depicts the first and second moments of the probability distributions for the 12-month forward-lookingENSOforecasts23.Asillustrated,thereissubstantialvariationinbothtime series.TheidentificationstrategyinthispaperreliesontheargumentthatENSOuncertainty is higher, the higher the standard deviation of the 12-month ENSO forecast. Figure 3 illustrates this idea. Using predicted rather than observed variation in ENSO activity, I constructthefollowingforward-lookingmeasurethatcapturesindustry-leveluncertaintyfor firms affected by changes in ENSO conditions (through uncertainty about weather and commoditypricesorotherchannels) CD?E =B,∙ 6-CD?E CD?E where=B,is the estimated ENSO sensitivity for industry k in year t and6-CD?E is the standarddeviation(secondmoment)oftheprobabilitydistributionforthe12-monthENSO forecast,whichwasissuedinyeart24.Asafirstmomentcontrol,Iconstruct subsequentanalysismayleadtoanoverestimationoftheeffectofENSOonthesefirmsandthusto misleadingresults.SeeSection5.3andAppendixDforanadditionaldiscussionofthisissue. 22 Starting in the early 1980s, the CPC started publishing monthly and weekly bulletins that summarizeinformationoncurrentand futureexpectedENSOconditions(seeAppendixBformore details). 23Thegraphillustratesthefirstandsecondmomentoftheforecastdistributions.Iremovealinear trendinthetimeseriesforthesecondmomenttomitigateconcernsaboutpossibletimetrendsinthe expectedaccuracyofforecasts.Theresultsofthepaperarerobusttoomittingthisstep.Thefirstand secondmomentsofmonthlyforecastsareaveragedovereachyeartoarriveatannualmeasurements. 24 The construction of the uncertainty measure is in line with the following simple intuition:6 =BCD?E ∙ H-CD?E = =BCD?E ∙ 6 H-CD?E = =BCD?E ∙ 6-CD?E . 17 CD?E =B,∙ H-CD?E whereH-CD?E denotesthefirstmomentoftheforecastprobabilitydistribution. In line with two-stage least squares methodology, the first stage of the instrumental variablesapproachtoestimatemodel(1)takesthefollowingform CD?E CD?E 6*,- = I* + J- + K/ 01,2 + )4 ∙ =B,∙ 6-CD?E + =B,∙ H-CD?E + L*,- (3) where I* and J- are firm and time fixed effects, and 01,2 is a vector of relevant control variables. Finally,substitutingthepredictedvaluesof6*,- fromregression(3)intomodel(1)yields thefollowingspecification !",$ %",$&' CD?E = )* + ,- + ./ 01,2 + 34 ∙ 6*,- + 3M ∙ =B,∙ H-CD?E + 7*,- (4) where)* represents firm fixed effects and,- denotes year fixed effects.01,2 is a vector of relevant control variables and6*,- denotes the predicted values for stock return volatility CD?E fromestimatingthefirststageregression(3).=B,∙ H-CD?E capturestheeffectthatthe12- monthENSOlevelforecast,H-CD?E ,hasonfirminvestmentinindustrykinyeart. 8*,- inequation(4)iscapitalexpenditurebyfirmiinperiodt,whichisscaledbythecapital stockattheendofthepreviousperiod9*,-:4 .Sincefinancialstatementsreportcapitalstock at book rather than replacement value, 9*,- is calculated recursively using a perpetualinventorymethod.ThiscalculationisdiscussedinmoredetailinAppendixA. Inmanytheoreticalmodels,uncertaintyaffectsfirminvestmentpartlyorentirelythrough marginalTobin’sq25.Therefore,itisstandardpracticetoincludemeasuresofTobin’sqas controls in investment models. Empirically, it is challenging to measure a firm’s marginal 25Mostnotably,AbelandEberly(1994)introduceirreversibilityintothemodelofAbel(1983)and showthatinvestmentdependsonlyonthecapitalstockandmarginalq.Therefore,theyarguethat uncertaintyimpactsinvestmentonlytotheextentthatitaffectsmarginalq.DixitandPindyck(1994) advance a similar argument and conclude that uncertainty changes the threshold value of q above whichfirmsarewillingtoinvest. 18 Tobin’sq,definedastheratioofthemarketvalueandthereplacementcostofanadditional unitofcapital.Thus,IincludethefollowingmeasureofaverageTobin’sq26 N*,- = OPQBR-SPT*-PU*VP-*WX",$ YZR[-",$ :S\QQRX-]^^R-^".$ %",$ Y!X`RX-WQa",$ Y!X-PXb*[UR^",$ Y!X`R^-<RX-^PXcPc`PXdR^",$ (5) which represents the ratio between the market value of a firm’s capital stock and its replacementcost.Othercontrolvariablesincludedarecashflowscaledbypreviousyear’s capitalstock,debt-to-equityratio,totalassetsandsales. Identificationoftheuncertaintyeffectoninvestmentinmodel(4)isachievedthroughthe combined effect of two sources of variation: First, SST forecasts are time-varying and generatedbyanunderlyingstochasticprocessthatisexogenoustothebusinesscycleand economic conditions. Second, the sensitivity to changes in ENSO conditions varies across industries. The first- and second-moment shocks are separated through the specific constructionofthesensitivities.Specifically,whileachangeinthefirstmomentofENSOhas opposing effects on industries with positive and negative ENSO sensitivities, an increase (decrease) in ENSO uncertainty, the second moment, represents an increase (decrease) in uncertainty for both types of industries. In addition, the first moment of the SST forecast probabilitydistribution,H-CD?E ,doesnotco-varywithitssecondmoment,6-CD?E ,inthetime series. 5. CompetitionandInvestmentunderUncertainty 5.1 IVEstimation–ENSOUncertainty Inthissection,Iexaminetheimpactofuncertaintyonfirminvestmentandtestwhether thiseffectdiffersdependingonthelevelofcompetitionthatfirmsfaceintheirindustries.For thatpurpose,Iconstructthreecommonlyusedproxiesforcompetitionthatarebasedonthe Herfindahl-Hirschmanindex(HHI)andprice-costmargins.First,IcalculatetheHHIbasedon 26 Average and marginal Tobin’s q to be equal requires, among other conditions, perfect competitionandconstantreturnstoscale(AbelandEberly,1994;Hayashi,1982).Itisunlikelythat theseconditionsaregenerallymetacrossindustriesandfirms.Therefore,evenundertheassumption thatuncertaintyonlyaffectsinvestmentthroughmarginalq,itisreasonabletobelievethattheimpact ofuncertaintyoninvestmentisnotinitsentiretycapturedbyaverageTobin’sq. 19 publicly listed firms using accounting data from Compustat (Compustat HHI)27 . Following Campello(2006)andByounandXu(2012),IconsideranindustryascompetitiveiftheHHI (expressed as fraction of the total market) is below 0.1. Due to concerns about the representativenessofacompetitionmeasurebasedentirelyonpublicfirms(Alietal.,2009), IalsoutilizetheindexconstructedbyHobergandPhillips(2010).Theirmeasure(HPHHI) captures industry concentration by using data for both private and public firms and is availablefrom1984to2005.IcategorizeindustriesashighlycompetitiveiftheirHPHHIfalls belowthesample33rdpercentile.Third,Iproxyforcompetitionusingtheprice-costmargin (PCM)orLernerindex,constructedasthefirm’soperatingprofitmargin(salesminuscosts dividedbysales)28.FollowingPeress(2010),Isubtractthevalue-weightedaverageindustry (basedon3-digitSICcodes)price-costmargintocontrolforsystematicdifferencesinthis measure across industries which are unrelated to the level of competition. Higher values indicatestrongermarketpowerandweakercompetition.Companiesareassumedtobein competitiveindustriesiftheirPCMsfallbelowthesample33rdpercentile. Prior to estimating the effect of uncertainty on investment using the instrumental variables approach described in Section 4.1.2, I estimate model (1) by simple OLS. I use annual,firm-levelreturnvolatilityasanuncertaintyproxywithoutaddressingitspossible endogeneity.TheresultsofsuchanaïveOLSestimationarepresentedincolumn1inPanelA ofTable2.Inlinewithpreviousstudiesusingreturnvolatilityasanuncertaintyproxy,the coefficient suggests a significantly negative relationship between uncertainty and capital investment.Asexpected,firmswithhigherTobin’sqratioshavehigherinvestmentspending. Inthenextstep,Isplitthesampleoffirmsaccordingtothedegreeofcompetitionintheir respective industries using the Compustat HHI measure. Columns 2 and 3 reveal that the negativeeffectofuncertaintyoninvestmentisobservedforbothsubsamples.However,due to the possible endogeneity of stock return volatility, it is difficult to infer any causal relationshipsusingasimpleOLSestimation. 27 This measure has been widely employed in the literature and I construct it following the methodologyusedinpreviousstudiessuchasAlietal.(2009),Baranchuketal.(2014)andHouand Robinson (2006). The Compustat based HHI measure is calculated for each year using the market shares of all firms in an industry and is averaged over the past three years. Industries are defined accordingto3-digitSICcodes. 28 The price-cost margin has been extensively used as a proxy for market power in empirical studies,particularlyintheindustrialorganizationliterature(Aghionetal.,2005;ByounandXu,2012; GasparandMassa,2006;LindenbergandRoss,1981;Nickell,1996;Peress,2010). 20 Toaddressthisissue,IfollowtheinstrumentalvariablesapproachdescribedinSection 4.1.2.First,Icalculateindustry-levelsensitivitiestoENSOactivitybyestimatingequation(2). IremoveanyfirmsfrommysamplethatbelongtoindustriesforwhichtheENSOsensitivity coefficients are not statistically significant (at the 5% level). Figure 4 illustrates the distribution of these sensitivities for the full (left) and the restricted sample (right). IndustrieswiththehighestsensitivitiestochangesinENSOconditionsareindustriesthatare either engaged in the production of commodities such as agricultural products and metal, industries that are affected by abnormal changes in weather conditions (e.g., construction andrelatedsectors)orindustriesthatrelyoncommoditypricesintermsoftheirinputsor outputs29.TestsusingthefullsampleoffirmsarediscussedintheSection5.3andinAppendix D.Theresultingsampleincludesmorethan6,500firmsand51,000observations. Inordertoexaminetheeffectofuncertaintyoncapitalinvestment,Iestimatemodel(4) byinstrumentingfirm-levelstockreturnvolatilitiesusingindustry-levelENSOsensitivities and ENSO prediction data. All of the following specifications include firm and year fixed effects to account for systematic differences in investment responses to heightened uncertaintyacrossfirmsandforshockstoinvestmentthataffectallofthefirmsatthesame time.Standarderrorsareclusteredatthefirmlevelinallspecificationsthroughoutthepaper. Theresults,basedontwo-stageleastsquaresestimation(2SLS),arepresentedincolumn1 inPanelBofTable2.Thecoefficientfortheinstrumentedstockreturnvolatilityisnegative but not statistically significant.While I observe no average effect of uncertainty on capital investment,theoreticalpredictionspointtoadifferentialeffectofuncertaintyoninvestment across firms facing different degrees of competitive pressure. In the next step, I test these predictionsbysplittingmysamplebasedonindustryconcentrationusingthesamethreshold levelfortheCompustatHHImeasureasbefore. Resultsofthisanalysisarepresentedincolumns2and3inPanelBofTable2.Ifindthat firms in highly competitive industries increase capital investment in response to higher uncertainty. This is in line with arguments that competition can erode the option value of deferring investment in the face of uncertainty allowing other channels through which uncertaintycanencourageinvestmenttodominate.Incontrast,firmsinmoreconcentrated industries tend to decrease investment in response to heightened uncertainty, which is 29Theseresultsareinlinewithconclusionsdrawninpreviousstudies.Brunner(2002)suggests thatresidentialconstructionandrelatedindustriesarestronglyaffectedbyENSOactivity,sincetheir operations depend heavily on weather conditions. In addition, Cashin et al. (2015) show a strong relationshipbetweenENSOvariationandcommoditypricesincludingenergyprices. 21 consistentwiththeargumentthatforsuchfirmsthepositiveoptionvalueofwaitingleads firmstodeferinvestmentspending. ThesignificantdifferencebetweentheOLSandIVestimateshighlightsendogeneityissues that bias the OLS estimates in the subsample of competitive industries more than in concentrated industries (see discussion in Appendix C). One finding of note is that the magnitudeofthecoefficientontheuncertaintymeasureislargerthanthemagnitudeofthe correspondingOLSestimateinmoreconcentratedindustries.Apotentialexplanationforthis finding is based on possible measurement errors in the uncertainty measures. Since stock return volatility is a very noisy proxy for uncertainty, the OLS estimates are likely biased towardzero.Suchabiasiscommonlyobservedinthecontextofmeasurementerrorsinthe endogenous explanatory variable and is usually eliminated by the instrumental variable approach(Almeidaetal.,2010;AngristandKrueger,2001;Card,1993;Gujarati,2003;Theil, 1971).Iinvestigatethemeasurementerrorexplanationbycomparingthevarianceofstock returnvolatilitywiththevarianceofthevaluespredictedbythefirststageregression.The variance of the raw stock return volatility variable is significantly larger than that of the predictedcomponent.Thissignificantdifferenceisconsistentwiththelargermagnitudeof thecoefficientinthe2SLSestimationandsupportstheerrors-in-variableexplanation. In addition to being statistically significant, the effect of uncertainty on investment illustrated in Panel B of Table 2 is also economically significant. A ten percentage point increaseinuncertaintyasmeasuredbystockreturnvolatilityleadstoa2.4percentagepoint decrease in the investment rate for firms in concentrated industries and a 2.8 percentage point increase for firms in highly competitive industries. Given the average sample investment rate of 25 percent these numbers approximately correspond to a 10 percent decreaseandan11percentincreaseinfirminvestmentrates,respectively. PanelCandDofTable2showresultsofsimilarestimationsusingtheHobergandPhillips (2010) HHI variable and price-cost margins to measure the degree of product market competition.Thesample-splitisbasedonthe33rdpercentileofeachmeasuretocapturethe differencebetweenhighlycompetitiveandmoreconcentratedindustries.Theresultsofthese estimationsarequalitativelysimilar.ComparedtotheestimatesinPanelB,themagnitudes ofthecoefficientsarelargerwhenusingtheHPHHImeasureandlowerwhensplittingthe sampleaccordingtoprice-costmargins.However,thedirectionoftheuncertaintyeffectis the same for each of the two subsamples regardless of the competition proxy used to categorize firms. Since the HP HHI time series ends in 2005, I will mostly focus on the 22 CompustatHHImeasureinsubsequenttests.However,themainconclusionspresentedinthe nextsectionsholdregardlessofthecompetitionmeasureemployed. TheaboveresultspresentedinTable2arebasedonseparateestimatesoftheimpactof uncertaintyoninvestmentforfirmsinhigh-andlow-competitionindustries.Ifocusonthe split-sample methodology for several reasons. First, empirical evidence shows that firms have different investment dynamics depending on the level of competition they face (Bustamante, 2015). Firm investment might change differently in response to changes of relevantcontrolvariablesintheinvestmentmodel,whichwarrantsaseparateestimationof theseparametersinthedifferentsubsamples.Second,theeffectofcompetitiononthelink betweenuncertaintyandinvestmentislikelynotlinear.Whileanincrementalchangeinthe degreeofcompetitionmighthavearathersmalleffectontheuncertainty-investmentlinkfor firmsinveryconcentratedindustries,achangeincompetitionofsimilarmagnitudemight havelargeeffectsaroundcertainthresholdlevelswhenanindustrytransitionsfrombeing moderatelyconcentratedtobeingcompetitive30. However, to demonstrate the robustness of the results, we can relax these two assumptions. Assuming that the effect of competition on the uncertainty-investment relationshipislinear,wecanestimateitbyincludingthecontinuouscompetitionmeasureas aregressorinmodel(4)togetherwithaninteractiontermbetweenthecompetitionproxy and return volatility. The results of estimating this specification are presented in Table 3. Compdenotesthe competitionproxy andVolxCompisthe interactionbetweenfirm-level return volatility and the competition measure 31 . The column headers in Table 3 indicate which competition measure is used. Columns 1 and 2 employ the Compustat HHI and the Hoberg and Phillips HHI measure, respectively, and illustrate that firms in concentrated industriesdecreaseinvestmentmorerelativetofirmsincompetitiveindustriesinresponse toheighteneduncertainty.Thecoefficientontheinteractionbetweenreturnvolatilityand price-costmarginsincolumn3isalsonegativealbeitnotstatisticallysignificant.However, the low F-statistic of the first stage regression for the interaction term indicates that this specificationincolumn3maysufferfromaweakinstrumentproblem.Themagnitudeofthe 30 In their Horizontal Merger Guidelines, the U.S. Department of Justice and the Federal Trade Commissionclassifyindustriesaseitherunconcentrated(competitive),moderatelyconcentratedor highlyconcentrateddependingontheirHHIvalues(DoJ,2010). 31Duetotheendogenousnatureoffirmreturnvolatility,theinteractiontermisinstrumentedas well.Theinteractiontermisconstructedbymultiplyingtheendogenousvariablereturnvolatilitywith the competition measure. In addition to using =BCD?E ∙ 6-CD?E as an instrument for stock return volatility,Iconstruct =BCD?E ∙ 6-CD?E ×fghiasaninstrumentfortheinteractionterm. 23 effectsincolumns1and2ofTable3areagaineconomicallysignificantandcanbeinterpreted asfollows.Asanillustration,firmsinindustrieswithHHIvaluesequaltothe5thpercentile plausiblyfacehighlevelsofcompetitivepressure,whilefirmsinindustrieswithHHIvalues equaltothe95thpercentilefacelowerlevelsofcompetition.Consideragainatenpercentage pointincreaseinuncertainty.Thecoefficientsincolumns1and2indicatethatinresponseto such a change in uncertainty, firms in the 5th percentile increase while firms in the 95th percentile decrease their investment rates. In column 1 the difference in the change in investment rates is 3 percentage points and in column 2 this difference is 12 percentage points.TheresultsinbothTable2andTable3providestrongevidencethatcompetitionisa majordeterminantofthesignandthemagnitudeoftheuncertainty-investmentrelationship. 5.2 ConventionalUncertaintyMeasures Oneconcernmightbethattheeffectofcompetitionontheuncertainty-investmentlink documentedinprevioussectionsisspecifictomyuniqueuncertaintymeasureandempirical identificationstrategy.Forthisreason,inthissection,Ishowthatthedifferentialeffectof uncertaintyacrossindustrieswithdifferentdegreesofcompetitioncansimilarlybeobserved whenusingmoreconventionaluncertaintymeasures. Aggregate and idiosyncratic uncertainty measures based on return volatility are often unsuitabletodeterminethesignandmagnitudeoftheuncertainty-investmentrelationship due to their possibly endogenous nature. However, they can still be useful to investigate whether firms adopt different investment strategies in response to increased uncertainty based on certain industry characteristics. I will exploit heterogeneity in the structure of product markets across industries to determine whether firms in competitive industries reactdifferentlytoheightenedaggregateuncertaintycomparedtotheircounterpartsinmore concentratedindustries.Inlinewithpreviousstudies,IemploytheS&P100VolatilityIndex (VXO) 32 as a proxy for aggregate uncertainty. Simple correlations between firm-level investment rates for the Compustat-CRSP merged sample and the VXO are presented in Figure5.Thelinesrepresentlinearregressionlinesforthefullsampleoffirmsandforthe sample of firms in competitive and concentrated firms (based on the Compustat HHI thresholdvalueof0.1).TherightgraphinFigure5clearlydepictsthedifferentslopesforthe threeregressionlines.Allfirms,onaverage,decreaseinvestmentinresponsetohigherVXO 32TheChicagoBoardofOptionsExchangeprovidesdatafortheS&P100volatilityindex(VXO)from 1986onwards,whilecoveragefortheVIXonlystartsin1989.UsingtheVIXindexyieldssimilarresults. 24 values.Thisrelationismorepronouncedforfirmsinconcentratedindustries,butreversed for firms in competitive industries, which seemingly increase investment when aggregate volatilityishigher. I test this difference in slopes more formally by estimating a specification of model (1) wherelaggedvaluesoftheVXOareusedastheuncertaintymeasure.Firmfixedeffectsand controlvariablessuchasTobin’sq,totalassets,cashflowscaledbycapitalstockandthedebtto-equityratioareincluded.TheresultsarepresentedinTable4.NotethatIamunableto include year fixed effects in this first specification, since the VXO index only varies across time. Column 1 indicates that an increase in aggregate volatility has a negative effect on investmentacrossallfirms.Thespecificationsincolumns2and3ofTable4includeadummy variableCompindicatingwhetherafirmbelongstoacompetitiveindustryaccordingtothe CompustatHHI(thresholdvalueof0.1asinprevioussections).VXOxCompisaninteraction termbetweenCompandtheVXOindex.Theresultsincolumns2and3illustratethatwhile anincreaseinaggregateuncertaintygenerallydecreasesinvestment,thedecreaseislessfor competitive firms. However, the effect of competition on the uncertainty-investment relationshipseemstobenon-linear.Column4showsthatthecoefficientontheinteraction betweentheHHIindex,whichisacontinuousmeasurethatvariesacrosstimeandindustries, and the VXO index is not statistically significant. Thus, a marginal increase or decrease in competition does not seem to affect the investment-uncertainty relationship unless this changeoccursaroundtheHHIthresholdlevelof0.1. Collectively,thefindingsinTable4providestrongevidencethatincreasesinaggregate uncertainty,onaverage,leadfirmstoreducecapitalinvestment.However,thiseffectdepends on the degree of product market competition that firms face. The negative effect of uncertaintyonfirminvestmentismuchlesspronouncedforfirmsincompetitivemarkets. 5.3 Robustness In this section, I briefly discuss additional empirical evidence to demonstrate the robustnessofmymainresults.PleaseseeAppendixDforamoredetaileddiscussionandthe resultsforthefollowingaswellasadditionalrobustnesstests. FullSampleofFirms In Section 5.1, I estimate the effect of uncertainty on firm investment using the instrumental variables approach outlined in Section 4.1.2. The first step consists of identifyingindustriesthatareaffectedbychangesinENSOconditionsandIusestockreturn 25 datatoestimateindustry-levelsensitivitiestoENSO.Subsequently,Iremoveanyfirmsfrom myanalysissampleinindustriesforwhichtheestimatedENSOsensitivityisnotstatistically significant (at the 5 percent level). This step was taken to increase the precision of the subsequent instrumental variables estimates. If industries are not affected by changes in ENSOconditions,theirestimatedsensitivitiesshouldbezero.Subsequently,therewillbeno variationintheuncertaintymeasureforthesefirmsthatcanbeusedtoidentifytheeffectof uncertaintyonfirminvestment.However,someindustrieshavelargesensitivitycoefficients with similarly large standard errors rendering the coefficient estimate not statistically significantfromzero.Includingsuchobservationsthroughoutsubsequentanalysismaylead tomeasurementerrorintheENSObaseduncertaintymeasure.Nonetheless,Icanestimate model(4)usingthefullsampleoffirmswithoutexcludingindustrieswithENSOsensitivities that are not statistically significant. The results of the estimation (see Appendix D.2) are similartotheonespresentedinPanelBofTable2.However,themagnitudeoftheuncertainty effect is smaller, suggesting that measurement error in the ENSO based uncertainty proxy mightbiasthoseestimatestowardszero.Firms,onaverage,reduceinvestmentandthiseffect is stronger for firms in concentrated industries. In contrast, firms facing high levels of competitionincreaseinvestment. CompetitionasaRiskFactorandthePossibleEndogeneityofCompetition One concern might be that industry-level ENSO sensitivities calculated by estimating regression(2)capturethedegreeofcompetitioninanindustry.Inthatcase,thedifferential effect of uncertainty in competitive versus concentrated industries might be a result of systematicdifferencesintheENSOsensitivitiesbetweenthetwosubsamples.Iaddressthis concernbyincludingcompetitionasariskfactorintheestimationofspecification(2).Using theENSOsensitivitiesestimatedthiswayyieldsasimilardifferentialeffectofuncertaintyon investment as presented in Panel B of Table 2 (see Appendix D.5). Firms in competitive industriesincreaseinvestmentandfirmsinconcentratedindustriesdecreaseinvestmentin responsetoheighteneduncertainty. Another concern relates to the causal links among uncertainty, investment, and competition. One might argue that uncertainty could affect the level of competition in an industrywhichinturnaffectsinvestmentandleadstothedifferentialeffectofuncertainty observedinPanelBofTable2.Fromatheoryperspective,itisunclearwhethersuchacausal linkbetweenuncertaintyandcompetitionexists,sincemodelsintheliteratureoninvestment underuncertaintygenerallydonotmodelitexplicitly.However,itisunlikelythatacausal 26 chainamonguncertainty,competitionandinvestmentcanexplaintheresultspresentedin this paper for several reasons. The unreported coefficients in the first stage, model (3), of CD?E estimatingmodel(4)usinginstrumentsarepositivefor =B,∙ 6-CD?E forallsubsamplesof firms, irrespective of the level of competition in an industry. This indicates that ENSO uncertainty has a positive effect on firm-level return volatility. Under the assumption that uncertaintyaffectsproductmarketcompetition,theresultswouldsuggestthatuncertainty increasesthelevelofcompetition,sinceithasbeendocumentedthathigherproductmarket competitionleadstoincreasedreturnvolatility(GasparandMassa,2006;IrvineandPontiff, 2009;Peress,2010).Foranincreaseincompetitiontoexplainthedifferentialresponseof investment,firmsinhighlycompetitiveindustrieswouldhavetoincreaseinvestment,while firmsinconcentratedindustrieswouldhavetodecreaseinvestmentinresponsetohigher competition. However, empirically, the opposite effect has been documented. Frésard and Valta(2013)examinetheeffectoftariffreductionsandtheassociatedincreaseincompetition from foreign rivals on firm-level investment. Their main result is that the effect of higher competitiondependsstronglyontheex-antecompetitivestructureoftheproductmarket. While an increase in competition leads to significant decreases in investment in highly competitive industries, the same increase in competition has no significant effect on firmlevelinvestmentinconcentratedindustries.Giventheseresults,itisunlikelythattheresults in this paper can be explained by an indirect effect of uncertainty on investment through changesincompetition. AdditionalTests TheresultspresentedinPanelBofTable2arealsorobusttocalculatingENSOsensitivities based on a four-factor model (see Appendix D.3), calculating ENSO sensitivities using an industryclassificationbasedontwo-digitSICcodes(seeAppendixD.6),andcalculatingtime invariantENSOsensitivitiesusingreturnsinyears1980to1984,whichispriortothemain estimation period (see Appendix D.4). In addition, the results are robust to using the unpredictablecomponentofreturnvolatilityastheuncertaintyproxy,6*,- ,inmodel(4)(see AppendixD.7). 6. TheEffectofUncertainty:UnderlyingMechanisms Theevidencepresentedintheprevioussectionsuggeststhattheeffectofuncertaintyon firm investment depends heavily on the competitive environment of firms. Firms in less 27 competitive industries decrease investment, while firms in highly competitive industries increase investment in response to higher uncertainty. In this section, I explore the underlyingmechanismsthatdrivethenegativeandpositiveeffects. 6.1 NegativeEffect:IrreversibleInvestmentunderUncertainty WhiletheresultsinPanelBofTable2illustratethatfirmsinlesscompetitiveindustries reduce investment under uncertainty, it remains unclear whether the negative effect of uncertainty (as observed in column 3) is in fact a result of the real option mechanism predictedbytheirreversibleinvestmentliterature.Toanswerthisquestion,Iwillonlyfocus onfirmsinconcentratedindustriesforthefollowingtestsinthissection. In models such as Abel and Eberly (1996) or Dixit and Pindyck (1994), the increased option value of waiting and deferring investment in uncertain times arises out of specific assumptionsaboutthedegreeofinvestmentreversibility.Inparticular,iffirmsfacepartial orfullirreversibilityofinvestment,anincreaseinuncertaintyexpandstheregionofinaction wherefirmsprefertowaitanddeferinvestment,creatinganegativelinkbetweenuncertainty andinvestment.Thiseffectislikelytobestrongerforfirmsfacinglessreversibleinvestment decisionsandhighercapitaladjustmentcosts.Iexaminethesepredictionsusingtwodifferent approachesthatidentifyheterogeneityintheabilityoffirmstoreverseinvestmentsandin theadjustmentcoststheyface. 6.1.1 Assetredeployability First, I test whether costly reversibility of investment leads to a negative investmentuncertainty relationship by exploiting variation in the type of capital that firms invest in. Consideringcapitalinvestmentasanacquisitionofoneormoreassets,firmscanreversesuch investmentdecisionsbyresellingthesecapitalgoodsonthesecondarymarketforcorporate assets. Empirical evidence is consistent with the idea that firms heavily engage in such activity(Gavazza,2011;Warusawitharana,2008).However,afirm’sabilitytoresellassetson thesecondarymarketheavilydependsonthetypeofcapitalgoodstheyinvestin.Frictions such as search costs for buyers and sellers as well as financial constraints can make such transactions more or less costly depending on the asset type (Kim and Kung, 2014). Irreversibilitygenerallyariseseitherduetoalackofsecondarymarketsforcertainassetsor ifassetsareindustry-orfirm-specificandcannoteasilybeusedbyotherfirmsinthesameor otherindustries.Assetsthatarewidelyusedwithinandacrossindustriesareplausiblyless costlytoresellonthesecondarymarket.Consequently,investmentinsuchassetsislikelyto 28 bemorereversiblethaninvestmentinassetswithalownumberofalternativeusers.Based onthislogic,KimandKung(2014)constructameasureofassetredeployabilityusingBureau ofEconomicAnalysis(BEA)capitalflowdata.Theassetredeployabilityscoreforanindustry depends on the usability of the assets employed by that industry and is higher the more industriesusethesameassets.Therefore,theauthorsarguethatfirmsinindustrieswithlow redeployabilityscoresfacemoreasymmetricadjustmentcoststhanfirmsinindustrieswith higherscores,sinceitiscostlierforthemtoreversetheirinvestmentsbysellingtheirassets on the secondary market. I split my sample of firms in concentrated industries into two groups based on whether their respective industry has a redeployability score above or belowthesamplemedianandestimatespecification(4)foreachsubsample.PanelAofTable 5illustratestheregressionresults.Inlinewiththepredictionsoftheliteratureonirreversible investment, firms with less redeployable assets decrease investment significantly more in response to higher uncertainty than their counterparts with more redeployable assets as indicatedincolumns1and2.Theseresultsprovidestrongevidencethatcostlyreversibility ofinvestmentisamajordriverofthenegativeuncertainty-investmentrelationshipobserved inlesscompetitiveindustries. 6.1.2 Valuevs.GrowthFirms Second, I use insights from Zhang (2005) to further test whether heterogeneity in the capital adjustment costs faced by firms can explain the negative effect of uncertainty on investment observed in more concentrated industries. Zhang (2005) argues that costly reversibilityofcapitalandcountercyclicalpriceofriskmakeitmoredifficultforvaluefirms to reduce capital, making them riskier than growth firms, particularly in bad times. The authorusesthislogictoexplainthedifferenceinexpectedreturnsbetweenvalueandgrowth firms.Theintuitionbehindtheargumentissimple.Valuefirmsderivemoreoftheirfirmvalue fromassetsinplaceandfromtheirrelativelylargercapitalstock,whilethemaindriversof valueforgrowthfirmsaregrowthoptions.Thus,inbadtimesvaluefirmswanttoreducetheir capital stock more sharply than growth firms, since they are faced with relatively more unproductive capital. In contrast, in good times growth firms increase their capital stock more than value firms, for which most of their unproductive capital becomes productive again.Sinceitiscostliertodecreasethanitistoincreasecapitalstock,ameanpreserving increaseinuncertaintyaboutfuturestatesoftheworld,onaverage,leadstohigherexpected adjustmentcostsforvaluefirms.Accordingtotherealoptionargument,valuefirmsshould thusdecreaseinvestmentinresponsetoheighteneduncertaintymoresharplythangrowth 29 firms to avoid relatively higher costs of reversing investment in bad times. To test this prediction empirically, I split my sample of firms into two groups based on whether their book-to-marketratioishigherorlowerthanthesamplemedian.Estimatingmodel(4)for eachofthesubsamplesyieldsresultspresentedinPanelAofTable5.Estimationsincolumns 3 and 4 indicate that the negative effect of uncertainty on investment in less competitive industriesisstrongerandstatisticallysignificantonlyforvaluefirms.Thedifferencebetween coefficientsforeachsubsamplebecomeslargerandstatisticallysignificantwhensplittingthe sample according to the 25th percentile of book-to-market ratios. These results are in line withpredictionsoftheliteratureonirreversibleinvestment. 6.1.3 TestingAlternativeExplanationsfortheNegativeUncertaintyEffect Onealternativeexplanationforthenegativeeffectofuncertaintyisbasedontheargument thatfirmsmayshiftinvestmentfromcapitalstocktowardsmoreflexibleandpotentiallyless productiveproductionfactors(EberlyandVanMieghem,1997).Fischer(2013)examinesa micro-level data set of firms in the Dominican Republic and finds that higher inflation uncertaintydecreasescapitalinvestmentandincreasesinvestmentinworkingcapital,which isoftenassumedtobeamoreflexibleandlessproductivefactorofproduction33.Itestthis ideabyestimatingtheeffectofuncertaintyonworkingcapital(scaledbythecapitalstock). Resultsofthisestimationarepresentedincolumn5inPanelAofTable5andillustratethat uncertaintyaffectsworkingcapitalinthesamedirectionascapitalinvestment.Firmsinmore concentrated industries decrease working capital in response to higher uncertainty. The sameinferencecanbedrawnundertheassumptionthatlaborisamoreflexibleproduction factorthancapital(Hartman,1972).Column6demonstratesthatuncertaintyalsonegatively affects hiring in concentrated industries. These findings do not support the idea that uncertaintydistortsinvestmentandshiftsitfromfixedassetstowardsmoreflexiblefactors suchasworkingcapitalorlabor. Anotherchannelthroughwhichuncertaintycanhaveadampeningeffectoninvestmentis based on the presence of financial market frictions. Gilchrist et al. (2014) argue that uncertainty has an indirect effect on investment. Increases in uncertainty raise default probabilitiesandthusincreasecreditspreadstocompensatebondholdersforhigherdefault risk. In response, firms decrease their leverage and reduce investment rates. I test this 33Forinstance,FazzariandPeterson(1993)discusstheroleofworkingcapitalasaninputfactor andinsmoothingfixedinvestmentsunderfinancingconstraintsduetolowercostsassociatedwith adjustingworkingcapital. 30 predictionbyexaminingtheexternalfinancingdecisionsoffirmsinconcentratedindustries. Columns7and8inPanelAofTable5illustratetheeffectofuncertaintyonnetdebtandnet equity issue, respectively 34 . The results contrast the predictions of the risk premium argument.Forfirmsinconcentratedindustries,uncertaintyhasnoeffectonnetdebtissue andanegativeimpactonequityissuewhichlikelyleadstohigherleverage.Infact,additional results presented in Appendix D.8 illustrate that the debt-to-equity ratio of firms in concentratedindustriesincreasesinresponsetoheighteneduncertainty.Thisfindinglends furthersupporttotheideathattheobservednegativeeffectofuncertaintyoninvestmentin concentrated industries is consistent with a real option as opposed to a risk premium argument. TheresultspresentedinPanelAofTable5indicatethatinvestmentirreversibilityand higher adjustment costs can strengthen the negative link between uncertainty and investmentinconcentratedindustries.Thesefindingsareconsistentwiththeargumentthat theoptionvaluetodeferinvestmentincreaseswiththedegreeofinvestmentirreversibility andthelevelofcapitaladjustmentcosts.Whilecompetitivepressurecandecreasethisoption value(asshowninTable2),thedecreaseshouldbemorepronouncedforfirmsthatcanmore easilyreverseinvestmentsandforfirmsthatfaceloweradjustmentcosts.PanelBofTable5 illustratesresultsofthetestsperformedinthissectionforthesampleofhighlycompetitive firmsthat,onaverage,increaseinvestmentinresponsetohigheruncertainty.Columns1and 2indicatethattheaveragepositiveuncertaintyeffectobservedinTable2isonlysignificant for firms with more redeployable assets. Similarly, only growth firms, whose expected adjustment costs are lower in uncertain times compared to those of value firms, increase investment in the face of heightened uncertainty. These findings suggest that even under higherlevelsofcompetition,therealoptionmechanism,onaverage,counteractsanypositive effects of uncertainty on investment for firms with less reversible investment and higher averageadjustmentcosts. 6.2 PositiveEffect:ConvexitiesintheFirm’sProfitFunction ResultsinTable2showthatfirmsinhighlycompetitiveindustries,onaverage,increase theirinvestmentspendinginresponsetouncertainty.Thesefindingsareconsistentwiththe idea that competition can erode the real option of waiting and deferring investment 34 Net debt and net equity issue are calculated according to Frank and Goyal (2003) using Compustatdata(seeAppendixA). 31 (Grenadier, 2002), allowing other channels through which uncertainty can encourage investmenttodominate. Theoretically,uncertaintycanhaveapositiveeffectoninvestmentifitincreasesthevalue of the marginal unit of capital. This idea is based on a simple Jensen’s argument. If future profits generated by the marginal unit of capital are a convex function of the underlying stochasticvariable(e.g.,inputoroutputprices),then,byJensen’sinequality,anincreasein uncertaintyraisesexpectedfutureprofits.Suchconvexitiesintheprofitfunctioncanarisein avarietyofdifferentsettings. 6.2.1 Labor-CapitalRatios InmodelssuchasOi(1961),Hartman(1972)andAbel(1983),themarginalproductof capitalisaconvexfunctionofoutputpricesandwageratesiflaborisaflexibleproduction factorthatcaneasilybeadjustedinresponsetopricechanges.Aspriceschange,firmsalso adjusttheirlabor-capitalratios,causingthemarginalrevenueproductofcapitaltochangeby morethanjustthefluctuationinprices.LeeandShin(2000)expandonthatresultandshow thattheconvexityintheprofitfunctionishigher,thegreatertheproportionoftheflexible production factor such as labor. I test this prediction by splitting firms in competitive industries into two subsamples based on whether their labor-capital ratios are above or belowthesamplemedian.Resultsofestimatingmodel(4)foreachsubsamplearepresented in columns 1 and 2 in Panel A of Table 6. Consistent with the discussion above, the significantlypositiveeffectofuncertaintyisonlyobservedforfirmswithhighlabor-capital ratios. The coefficients on return volatility for the two subsamples are not statistically significant from each other. However, splitting firms into two subsamples using the 25th percentileofthesamplelabor-capitalratiosyieldscoefficientsthatarestatisticallydifferent fromeachother,asshownincolumns3and4inPanelAofTable6. 6.2.2 OperationalFlexibility Evenwithfixedproportionsoftheproductionfactors,convexitiesintheprofitfunction canresultfromoperationalflexibilityandafirm’sabilitytovaryoutputaccordingtomarket conditions(MarschakandNelson,1962;Oi,1961).Iffirmscaneasilyexpandtoexploitgood outcomes and easily contract in response to bad outcomes, they might desire a meanpreservingincreaseinuncertainty.Inotherwords,iffirmsdonotneedtoutilizethemarginal unitofcapitalinbadtimes,aunitofcapitalcanbeseenasanoptiononfutureproduction. Thevalueofthisoptionwillincreaseinthevarianceofinputoroutputprices,leadingtoa 32 positive relationship between uncertainty and capital investment. Mills (1984) formalizes thisintuitionandshowsthattheexpectedprofitofacompetitivefirmfacingstochasticprices increasesinitsoperationalflexibility j k i =k j i + lm n M , wherek i istheprofitgivenpricepandpricevariance6 M .Theparameter=representsthe degree of operational flexibility a firm has and varies inversely with the cost of changing output in response to price fluctuations. Following Grullon et el. (2012), I measure operational flexibility using industry-level labor union membership. Intuitively, firms face moreconstraintswhenadjustinglaborinputinresponsetochangesineconomicconditions (particularly if it involves reductions in labor in economic downturns) if the workforce is heavilyunionized(AbrahamandMedoff,1984;Chenetal.,2011;GrammandSchnell,2001). Thus,Itreatunionmembershipasapercentageoftheworkforceinanindustryasaninverse proxyforoperationalflexibility.Isplitthesampleofcompetitivefirmsintotwosubsamples usingthesamplemedianofthismeasureandestimatemodel(4)foreachofthesesubsamples separately.Theresultsarepresentedincolumns5and6inPanelAofTable6.Thecoefficient oninstrumentedreturnvolatilityispositiveforfirmswithlowunionmembershipratesand negative for firms with high union membership rates. Both estimates are statistically significant.Thesefindingssupporttheideathatoperationalflexibilitycombinedwithhigh levelsofcompetitioncanencourageinvestmentinresponsetoheighteneduncertainty. PanelBofTable6presentsresultsofthetestsinthissectioninthesampleoffirmsinmore concentratedindustries.Columns2and4illustratethatfirmswithhighlabor-capitalratios inconcentratedindustriesdecreaseinvestmentinresponsetouncertainty.Theuncertainty effectisnotstatisticallysignificantforfirmswithlowlabor-capitalratios.However,theFstatisticforthefirststageregressionindicatesthatthesespecificationsmightsufferfroma weak instrument problem. Columns 5 and 6 indicate that union membership does not significantlychangetheimpactofuncertaintyoninvestmentinconcentratedindustries.This result suggests that convexities in the firm’s profit function associated with operational flexibility cannot mitigate the negative effect of uncertainty arising from the real option mechanisminconcentratedindustries. 33 7. LargeChangesinUncertaintyandtheDynamicsofInvestment Section5illustratesthattheeffectofuncertaintyonfirminvestmentheavilydependson the level of competition that firms face in their industries. The results show a significant wedge in the response of investment spending to increased uncertainty between firms in competitiveandconcentratedindustries.Idocumentthatthiswedgeispartlycausedbya decline in investment in concentrated industries, which is consistent with real options argumentsintheirreversibleinvestmentliterature. Sofar,weknowlittleabouthowpersistentthiseffectofuncertaintyisacrosstime.Bloom etal.(2014)showthatthenegativeeffectofuncertaintyoninvestmentprescribedbythereal options literature generally results in a short and sharp drop in investment. In their simulations,thisdropisobservedintheperiodinwhichtheuncertaintyshockoccursandin thefollowingperiod.Onceuncertaintydecreases,investmentratesrecovertolevelsslightly above the ones observed prior to the uncertainty shock. The authors conclude that uncertainty shocks lead to a short and sharp drop in investment, which results ina lower capitalstocksincecapitalcontinuestodepreciate.Thisisfollowedbyaprolongedrecovery ofthecapitalstocktoitssteadystate.Basedontheseresults,thewedgeintheresponseof investment to heightened uncertainty between firms in competitive and concentrated industriesshouldbesimilarlyshort-livedifitismainlyaresultoftherealoptionsmechanism. Inthissection,Itestthesepredictionsbyfocusingonlargechangesinuncertainty,which I define as the five largest yearly increases and the five largest yearly decreases in my CD?E measureofENSOuncertainty,6B,,overtheentireperiodof1985to2015.Iinvestigatethe dynamicsofinvestmentfollowinganuncertaintyshockbyusingadifference-in-differences approachandbyestimatingthefollowingmodel 8*,= )* + o- + 3M f*,- + 3p q-:M f*,- + 3p q-:4 f*,- +3p q-> f*,- + 3p q-Y4 f*,- + 3p q-YM f*,- + rs*,- + 7*,- 9*,-:4 wheref*,- isadummyindicatingwhetherfirmiisinahighlycompetitiveindustry(basedon the threshold value of 0.1 for the Compustat HHI).q-YM ,q-Y4 ,q-> ,q-:4 , andq-:M are dummy variablesindicatingwhetheralargechangeinuncertaintytookplaceinyeart-2,t-1,t,t+1, andt+2,respectively.s*,- denotesthesamevectorofcontrolvariablesasinsection5.Aswith allpreviousspecifications,Iincludefirmandtimefixedeffectsandclusterstandarderrorsat thefirmlevel. 34 First,Iexaminethefiveyearsinwhich6-CD?E increasedthemostandestimatetheabove specification.Resultsarepresentedincolumns1and2ofTable7.Column2revealsthatthere arenosignificantchangesininvestmentpriortoanuncertaintyshockandhighlightsthatthe differential effect of uncertainty between competitive and concentrated industries is only significant in the year of the uncertainty shock and in the subsequent year. In these two periods,firmsinmorecompetitiveindustriesinvestmoreinresponse touncertaintythan their counterparts in more concentrated industries. This wedge in investment rates is not statisticallysignificantinyear2aftertheuncertaintyshock.Incolumns3and4ofTable7,I investigate the response of firm investment to a decrease in uncertainty. The average magnitudeofthefivelargestdecreasesinuncertaintyissimilartotheaveragemagnitudeof thefivelargestincreasesinuncertaintyexaminedincolumns1and235.Column4indicates that in response to decreases in uncertainty, firms in concentrated industries invest more than firms in competitive industries. This significant effect is only observed in the year in whichuncertaintydecreases.Thedifferentialeffectofadeclineinuncertaintyoninvestment betweenfirmsincompetitiveandconcentratedindustriesissmallerinmagnitude.Following anincreaseinuncertainty,firmsincompetitiveindustrieshave2.8and4.3percentagepoints higherinvestmentratesthantheircounterpartsinconcentratedindustriesinthesameyear of the uncertainty shock and in the following year, respectively. In contrast, firms in concentrated industries have 1.9 percentage points higher investment rates than firms in competitive industries in the year of an uncertainty decrease. This suggests that the differential effect of uncertainty on investment rates after an uncertainty increase is not immediatelyoffsetonceuncertaintydecreasestoasimilarextent. ThefindingspresentedinTable7arehighlyconsistentwiththeresultsinBloometal. (2014)andprovidefurtherevidencefortheargumentthatthedifferenceininvestmentrates followinganuncertaintyshockbetweenfirmsincompetitiveandconcentratedindustriesis infactaresultoftherealoptionsmechanismintheirreversibleinvestmentliterature. 35Theaverageincreasein6 CD?E duringtheyearswiththehighestspikesinENSOuncertaintyis - 0.069andtheaveragedecreasein6-CD?E duringtheyearswiththehighestdeclinesinuncertaintyis0.057.However,thedifferencebetweentheabsolutevaluesofthesetwonumbersisnotstatistically significant(p-valueof0.63). 35 8. UncertaintyandInvestmentinOtherFactors Whileuncertaintyhasasignificantimpactoncapitalinvestment,itislessclearwhether the effect is similar for other types of investment such as R&D, advertising spending or investment in labor through hiring. I aim to address these empirical questions by first estimating model (4) using R&D spending scaled by the knowledge stock as a dependent variable36.Theresultsofthisanalysisarepresentedincolumns1to3ofTable8.Uncertainty hasnosignificanteffectoninvestmentforfirmsinhighlycompetitiveindustries.Incontrast, firms in more concentrated industries reduce R&D spending in response to heightened uncertainty.Columns4to6showsimilarresultswhenusinganalternativeproxyforR&D investment opportunities instead of Tobin’s Q. R&D Q is constructed using the same nominator as in equation (5) but using the knowledge stock as the denominator. These findingssuggestthatsimilarmechanismsareatworkforbothcapitalandR&Dinvestment underuncertainty.However,themagnitudeoftheuncertaintyeffectonR&Dissmaller(about halfthesizeofthecoefficientsforcapitalinvestment),whichisconsistentwiththeempirical regularitythatR&Dspendingishighlypersistentinthetimeseries(Bloom,2007). Table9showssimilarresultsforadvertising(scaledbyadvertisingcapitalstock37).Firms, on average reduce their advertising spending in response to higher uncertainty. However, thiseffectisnotstatisticallysignificantforfirmsinhighlycompetitiveindustries. Finally,Ireportresultsforhiringasmeasuredbythenumberofemployeesscaledbytotal assetsinTable10.Onaverage,Ifindanegativerelationshipbetweenuncertaintyandhiring forallfirms.However,thiseffectseemstobedrivenbyfirmsinmoreconcentratedindustries asillustratedincolumns2and3whereIsplitthesamplebasedonindustrycompetition.The negativeeffectofuncertaintyonhiringisinlinewithpreviousstudiesthatrelateuncertainty associatedwithrecessionstoanaggregatefallinhiring(SteinandStone,2013). 9. Conclusion Themaincontributionofthispaperistwo-fold:First,Iexploitauniqueempiricalsetting to construct an uncertainty measure that is exogenous to economic conditions and firm behavior. This allows me to estimate the causal effect of uncertainty on firm investment. Second, I examine a broad data set of more than 6,500 U.S. firms in a variety of different 36Thecalculationoftheknowledgestockfollowstheperpetual-inventorymethodusedinHalland Mairesse(1995).SeeAppendixAformoredetails. 37SeeAppendixAfordetailsaboutthecalculationoftheadvertisingcapitalstock. 36 industries and document that the uncertainty effect differs depending on industry characteristicssuchasproductmarketcompetition. Theresultsinthispaperchallengethecommonviewamongacademicsandpolicymakers that increased uncertainty leads to a decline in capital investment effectively dampening economicgrowth.Ishowthatthecompetitiveenvironmentplaysalargeroleindetermining the sign and magnitude of the uncertainty and investment relationship. Firms in highly competitive industries tend to increase capital investment, while firms in less competitive industriesreduceinvestmentinmoreuncertaintimes. Ifindsubstantialevidencesupportingtheargumentthatthenegativeeffectofuncertainty inmoreconcentratedindustriesisaresultoftherealoptionsmechanismintheirreversible investment literature. Firms with less redeployable capital and firms incurring higher expectedadjustmentcostsinthefaceofuncertaintydecreaseinvestmentinuncertaintimes. This reduction in investment is not offset by corresponding changes in more flexible production factors such as working capital or labor. Similarly, I find no support for the argumentthatuncertaintyaffectsfirminvestmentthroughborrowingcosts.Incontrastto predictionsinGilchristetal.(2014),uncertaintyhasnoeffectonnetdebtissueandanegative effectonnetequityissueforfirmsinconcentratedindustries. Thepositiveeffectofuncertaintyincompetitiveindustriessuggeststhatcompetitioncan counteractthenegativeimpactofuncertaintyoninvestmentpredictedbytherealoptions literature.Instead,uncertaintycanactivelyencourageinvestmentincompetitiveindustries iffirmshaveoperationalflexibility.Inlinewiththisprediction,Ifindthatthepositiveeffect is only significant for firms in competitive industries that have operational flexibility (e.g., firms with low union membership rates). Similarly, firms where labor, which is often assumedtobeamoreflexibleproductionfactorthancapital,representsalargershareofthe productiontechnologyincreaseinvestmentinresponsetohigheruncertainty.Ifurthershow thatuncertaintyhasaneconomicallysignificanteffectonothertypesofinvestmentsuchas R&Dspending,advertising,andhiring. The findings in this paper have important policy implications. My results provide new empirical evidence and corroborate existing theoretical predictions on the importance of competitioninthelinkbetweenuncertaintyandfirminvestment.Policymakersaimingto attenuate the negative effects of uncertainty on the real economy would need to put substantialattentiononindustry-levelcharacteristicsinthedesignofoptimalpolicies. 37 TablesandFigures Figure1 Seasurfacetemperatureanomaliesin“Niño3.4”region Thisgraphdepictsthetimeseriesofseasurfacetemperature(SST)anomaliesinthecentralPacificregion “Niño3.4”from1980to2014.TheNationalOceanicandAtmosphericAdministration(NOAA)definesanomalies asseasurfacetemperaturedeviationsinaregionfromitshistoricalmean.MonthlySSTdatacanbedownloaded fromthewebsiteoftheClimatePredictionCenter(CPC)attheNOAA. 4 2 1 0 -1 -2 -3 -4 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 3 SSTanomalies Figure2 Seasurfacetemperatureforecastsin“Niño3.4”region This graph shows the time series of the 12-month sea surface temperature (SST) predictions and their standarddeviationsforthePacificregion“Niño3.4”.ThesepredictionsarebasedonestimatesbytheClimate Prediction Center (CPC) at the National Oceanic and Atmospheric Administration (NOAA), which kindly providedthedataforthisstudy. 1.1 1 0.9 0.8 0.7 0.6 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 -0.6 -0.8 -1 SSTforecastdistributionmeanH-CD?E SSTforecastdistributionSD6-CD?E 38 Figure3 Illustrationoflowandhighuncertaintystate Thisgraphdepictsexampleprobabilitydistributionsfor12-monthENSOforecasts(bluecurve).Uncertainty islowerinyearswhenthestandarddeviationofthedistributionissmaller(leftgraph).Incontrast,uncertainty ishigherwhenthestandarddeviationislarger(rightgraph). tuvwx 2 yuvwx 2 Figure4 IndustrysensitivitiestoENSO This graph illustrates the distribution of ENSO sensitivities across industries calculated using regression model(2).Forillustrationpurposes,thesesensitivitieswereestimatedbasedontheentiresampleoffirmsforthe periodfrom1985to2014andarenottime-varying.Intheestimationofmodel(4),time-varyingsensitivitiesare usedbasedonthecalculationsdescribedinSection4.1.1.Theleftfigureshowssensitivitiesfortheentiresample offirms,whiletherightfigureillustratesENSOindustrysensitivitiesthatarestatisticallysignificantata5%level. CD?E =zB, CD?E =zB, 39 Figure5 Correlationbetweenvolatilityindex(VXO)andinvestmentrates ThisgraphillustratesthecorrelationbetweenthevolatilityindexVXOprovidedbytheChicagoBoardof OptionsExchange andfirm-levelinvestmentrates.The VXO is basedonimpliedvolatilitiesofthe S&P 100 options.Thelinesrepresentsimplelinearregressionlinesforthefullsampleoffirmsandthesampleoffirms incompetitiveandconcentratedindustries. 40 Table1 Summarystatistics ThistableshowssummarystatisticsandthedistributionoffirmcharacteristicsinthefullsampleofCompustat firmsandtheanalysissample,whichonlyincludesfirmsinindustrieswithstatisticallysignificantsensitivitiesto changesinENSOconditions. Compustat Mean Median Std.dev. Sales($M) 1689 82 9598 TotalAssets($M) 2197 95 Investment($M) 143 Capitalstock($M) It/Kt-1 Analysissample Mean Median Std.dev. 1574 67 9464 12883 2128 85 12321 4 922 145 4 982 1048 28 6772 1033 25 7058 24% 13% 27% 25% 14% 28% R&D($M) 70 2 436 75 3 447 R&Dstock($M) 322 13 2185 338 16 2217 Rt/Gt-1 27% 22% 20% 27% 22% 21% Cashholdings($M) 183 7 1298 191 7 1276 CEt/TAt-1 19% 8% 20% 9% 25% 92,341 Firms 12,017 N 24% 51,203 6,695 41 Table2 Theeffectofuncertaintyoncapitalinvestmentunderdifferinglevelsofproductmarketcompetition This table presents results of various regression specifications to investigate the impact of uncertainty on capitalinvestment(investmentscaledbythepreviousyear’scapitalstock).PanelAillustratesresultsofasimple OLSestimationofmodel(1).PanelsB,CandDpresentresultsoftheinstrumentalvariableestimationdescribed inSection4.1.2usingthreedifferentmeasuresofcompetitiontosplitthesampleintofirmsinhighlycompetitive industries(HighCompetition)andfirmsinlesscompetitiveindustries(LowCompetition).CompustatHHI,HPHHI and PCM denote a Compustat based Herfindahl-Hirschman Index (HHI), the HHI measure used in Hoberg and Philips(2010)andaprice-costmarginmeasure,respectively.Controlsincludetotalassets,debt-to-equityratio, cashflowsscaledbycapitalstock,salesandthe12-monthSSTlevelforecast.Standarderrorsareclusteredatthe firmlevel.***,**and*indicate1%,5%,and10%significance,respectively. PanelA:OLS-CompustatHHI AllFirms HighCompetition LowCompetition (1) (2) (3) Returnvolatility Tobin'sQ -0.0760*** -0.0606*** -0.0877*** (0.01) (0.01) (0.01) 0.0162*** 0.0123*** (0.00) 0.0176*** (0.00) (0.00) N 51,203 24,663 26,540 Controls Yes Yes Yes YearFE Yes Yes Yes Yes Yes Yes FirmFE PanelB:IV-CompustatHHI AllFirms HighCompetition LowCompetition (1) (2) (3) Returnvolatility -0.0471 F-Stat.-1stStage (0.05) 0.0172*** (0.00) N -0.2384*** (0.07) 0.0197*** 0.2792*** (0.03) Tobin'sQ 0.0192*** (0.00) (0.00) 51,203 24,663 26,540 229.12 87.99 117.93 Controls Yes Yes Yes YearFE Yes Yes Yes FirmFE Yes Yes Yes 42 Table2continued PanelC:IV–HPHHI HighCompetition LowCompetition (1) (2) (3) Returnvolatility -0.0466 F-Stat.-1stStage (0.08) 0.0213*** (0.00) N -0.3786*** (0.04) 0.0197*** 0.3726*** (0.03) Tobin'sQ AllFirms 0.0157*** (0.00) (0.00) 34,168 11,390 22,778 325.82 473.45 104.82 Controls Yes Yes Yes YearFE Yes Yes Yes Yes Yes Yes FirmFE PanelD:IV-PCM AllFirms HighCompetition LowCompetition (1) (2) (3) Returnvolatility -0.0506 F-Stat.-1stStage (0.04) 0.0177*** (0.00) N -0.1114** (0.06) 0.0197*** 0.1157** (0.03) Tobin'sQ 0.0224*** (0.00) (0.00) 49,750 16,583 33,167 262.95 97.37 153.65 Controls Yes Yes Yes YearFE Yes Yes Yes FirmFE Yes Yes Yes 43 Table3 The effect of competition on the investment-uncertainty relationship using interaction terms and IV estimates ThistablepresentsresultsofIVestimatesformodel(4)wherethedependentvariableisinvestmentscaled bythepreviousyear’scapitalstock.CompindicatesthecompetitionproxyandVolxCompisaninteractionterm betweenthecompetitionmeasureandreturnvolatility.Thecolumnheadersindicatewhichcompetitionmeasure isused.CompustatHHI,HPHHIandPCMdenoteaCompustatbasedHerfindahl-HirschmanIndex(HHI),theHHI measureusedinHobergandPhilips(2010)andaprice-costmarginmeasure,respectively.Controlsincludetotal assets,debt-to-equityratio,cashflowsscaledbycapitalstock,salesandthe12-monthSSTlevelforecast.Standard errorsareclusteredatthefirmlevel.***,**and*indicate1%,5%,and10%significance,respectively. Returnvolatility CompustatHHI HPHHI PCM (1) (2) (3) 0.3071*** 0.0003 (1.75) -0.9048*** (0.00) 12.9676*** (0.17) VolxComp 0.0213*** (0.00) 0.6929*** (0.04) 0.0216*** (0.00) Comp 0.1627*** (0.15) 0.0206*** 1.2789*** (0.06) Tobin'sQ (0.53) -18.6904*** (0.32) -0.0006 (2.73) (0.00) N 51,203 34,168 49,750 F(Vol) 121.25 179.8 114.8 43.98 170.21 0.48 F(VolxComp) Controls Yes Yes Yes YearFE Yes Yes Yes FirmFE Yes Yes Yes 44 Table4 Theeffectofaggregateuncertaintyonfirm-levelinvestmentincompetitiveandconcentratedindustries This table presents results of estimating model (1) using OLS where the dependent variable isinvestment scaled by the previous year’s capital stock. Comp is a dummy variable indicating whether a firm is in a highly competitiveindustry(basedonCompustatHHIthresholdvalueof0.1).VolxCompisaninteractiontermbetween Comp and the volatility index VXO. HHI is the Compustat HHI measure and VXO x HHI is an interaction term between HHI and the volatility index. Controls include total assets, debt-to-equity ratio, cash flows scaled by capitalstock,salesandthe12-monthSSTlevelforecast.Standarderrorsareclusteredatthefirmlevel.***,**and *indicate1%,5%,and10%significance,respectively. (1) VXO Tobin'sQ (2) -0.0024*** (0.00) (0.00) 0.0235*** (0.00) Comp. -0.0021*** 0.0235*** (3) VXOxComp. 0.0213*** (0.00) (0.00) -0.0377*** -0.0188*** (0.01) (0.01) (0.00) 0.0006*** (0.00) HHI 0.0213*** 0.0009*** (4) (0.00) 0.0790*** (0.02) VXOxHHI 0.0004 (0.00) N 119,989 119,989 119,989 119,989 Controls Yes Yes Yes Yes YearFE No No Yes Yes FirmFE Yes Yes Yes Yes 45 Table5 Testingtherealoptionmechanismintheirreversibleinvestmentliteratureasthedriverofthenegativeeffectofuncertaintyoninvestmentinconcentrated industries Thistablepresentsresultsofestimatingmodel(4)usinginstrumentalvariables.TheCompustatbasedmeasureoftheHerfindahl-HirschmanIndex(HHI)isusedto split the sample into firms in less competitive industries (Panel A) and firms in highly competitive industries (Panel B). The dependent variable in columns 1 to 4 is investment scaled by the previous year’s capital stock. In columns 1 and 2, the sample is split into firms with more or less redeployable assets based on the asset redeployabilityindex(RI)constructedbyKimandKung(2014).Incolumns3and4,thesampleissplitintogrowthandvaluefirmsbasedonwhetherthebook-to-market valueexceedsorisbelowthesamplemedian.Thedependentvariableforcolumns5,6,7and8isworkingcapital,employeesscaledbytotalassets,netdebtissue,andnet equityissue,respectively.Controlsincludetotalassets,debt-to-equityratio,cashflowsscaledbycapitalstock,salesandthe12-monthSSTlevelforecast.Standarderrors areclusteredatthefirmlevel.***,**and*indicate1%,5%,and10%significance,respectively. PanelA:LessCompetitive(MoreConcentrated)Industries Redeployability Valuevs.Growth FlexibleFactors ExternalFinancing LowRI HighRI (1) (2) Returnvol. -0.8117*** -0.1805** (0.18) (0.08) Tobin'sQ 0.0117*** (3) (4) -0.1634 -0.2934*** (0.13) (0.08) 0.0172*** (0.00) Value 0.0208*** (0.00) Growth (0.01) (5) (6) NDI NEI (7) (8) -0.1962** -0.0247* (0.09) (0.01) (0.03) -0.0001 (0.00) 0.0296*** Labor 0.0360*** (0.00) WC 0.0009* (0.00) (0.00) 0.0042 -0.1298*** (0.05) 0.0149*** (0.00) N F-Stat.-1st Stage 9,517 9,397 10,802 14,192 15,105 25,587 24,739 24,307 44.13 76.35 56.66 76.22 60.25 143.84 139.43 126.46 Controls Yes Yes Yes Yes Yes Yes Yes Yes YearFE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes FirmFE Yes Yes Yes Yes Yes 46 Table5continued PanelB:HighlyCompetitiveIndustries Redeployability Valuevs.Growth FlexibleFactors ExternalFinancing LowRI HighRI (1) (2) Returnvol. 0.0798 (0.16) 0.0139*** 0.0230*** (0.00) (0.00) Growth Value (3) (4) 0.3720** (0.11) Tobin'sQ 0.4990*** (0.13) 0.0164*** (0.00) WC Labor (5) (6) -0.0096 (0.15) 0.0342*** (0.00) 0.2773** (0.00) 0.0288*** 0.0000 (0.00) NDI NEI (7) (8) 0.0034 (0.13) (0.00) -0.0120 (0.05) 0.3337*** (0.10) 0.0003 0.0209*** (0.00) (0.00) N F-Stat.-1st Stage 9,766 9,864 12,649 9,582 12,660 22,644 22,351 20,532 84.50 20.50 90.66 17.37 122.33 82.28 79.63 51.39 Controls Yes Yes Yes Yes Yes Yes Yes Yes YearFE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes FirmFE Yes Yes 47 Table6 Driversofthepositiveeffectofuncertaintyoninvestmentinhighlycompetitiveindustries Thistablepresentsresultsofestimatingmodel(4)usinginstrumentalvariables.TheCompustatbasedmeasureoftheHerfindahl-HirschmanIndex(HHI)isusedto splitthesampleintofirmsinhighlycompetitiveindustries(PanelA)andfirmsinlesscompetitiveindustries(PanelB).Thedependentvariableinallcolumnsisinvestment scaledbythepreviousyear’scapitalstock.Incolumns1to4,firmsarecategorizedaccordingtotheirlabor-capitalratios.Incolumns5and6,thesampleissplitintofirms inindustrieswithlowandhighlevelsofunionmembershipratesbasedonthesamplemedian.Controlsincludetotalassets,debt-to-equityratio,cashflowsscaledby capitalstock,salesandthe12-monthSSTlevelforecast.Standarderrorsareclusteredatthefirmlevel.***,**and*indicate1%,5%,and10%significance,respectively. PanelA:HighlyCompetitiveIndustries Labor-CapitalRatio LaborUnionMembership LowL/K(<median) HighL/K(>median) LowL/K(<25thpctile) (1) (2) (3) Returnvol. 0.1287 0.1535** (0.12) Tobin'sQ 0.2969*** (0.17) 0.0143*** (4) -0.1242 (0.07) 0.0127*** HighL/K(>25thpctile) (0.07) 0.0166*** 0.0152*** (0.00) (0.00) LowLUM HighLUM (5) (6) 0.5730* -0.3144** (0.30) (0.13) 0.0166*** 0.0136*** (0.00) (0.00) (0.00) (0.00) 10,661 11,235 4,456 17,734 8,691 7,191 44.91 137.37 22.78 157.84 20.71 51.99 N F-Stat.-1stStage Controls Yes Yes Yes Yes Yes Yes YearFE Yes Yes Yes Yes Yes Yes FirmFE Yes Yes Yes Yes Yes Yes 48 Table6continued PanelB:LessCompetitive(MoreConcentrated)Industries Labor-CapitalRatio LaborUnionMembership LowL/K(<median) HighL/K(>median) LowL/K(<25thpctile) (1) (2) (3) Returnvol. 0.0960 (0.11) Tobin'sQ 0.0219*** (4) -0.3494*** -0.8705 -0.2709*** (0.05) (0.92) (0.04) 0.0151*** HighL/K(>25thpctile) 0.0000 HighLUM (5) (6) 0.0190*** (0.02) LowLUM -0.3945*** -0.2305*** (0.09) (0.06) 0.0172*** (0.00) 0.0179*** (0.00) (0.00) (0.00) (0.00) 10,025 14,818 4,314 20,828 8,493 13,701 12.31 235.87 1.19 148.95 62.66 106.73 N F-Stat.-1stStage Controls Yes Yes Yes Yes Yes Yes YearFE Yes Yes Yes Yes Yes Yes FirmFE Yes Yes Yes Yes Yes Yes 49 Table7 Largechangesinuncertaintyandthedynamicsofinvestment Thistablepresentsresultsofestimatingthepersistenceofthedifferentialeffectofuncertaintyoninvestment between competitive and concentrated industries using the difference-in-differences approach presented in Section7.Controlsincludetotalassets,debt-to-equityratio,cashflowsscaledbycapitalstock,salesandthe12monthSSTlevelforecast.Standarderrorsareclusteredatthefirmlevel.***,**and*indicate1%,5%,and10% significance,respectively. LargeIncreaseinUncertainty LargeDecreaseinUncertainty (1) Tobin'sQ C !×# $% (2) 0.0149*** !×# (0.00) -0.0152*** -0.0237*** -0.0107** -0.0103* (0.00) (0.01) (0.00) (0.01) -0.0078 (0.01) (0.01) -0.0080 (0.01) 0.0034 0.0152*** 0.0284*** (0.01) !×# (& (0.01) 0.0433*** (0.01) !×# (% 0.0149*** (0.00) (0.01) !×# 0.0149*** ' (0.00) (4) (0.00) $& (3) 0.0149*** 0.0030 0.0074 -0.0181*** -0.0185*** (0.00) (0.01) -0.0049 (0.01) (0.00) 0.0041 (0.01) N 51,203 51,203 51,203 51,203 Controls Yes Yes Yes Yes YearFE Yes Yes Yes Yes FirmFE Yes Yes Yes Yes 50 Table8 TheeffectofuncertaintyonR&Dinvestmentundervaryinglevelsofcompetition This table presents results of estimating model (4) using an instrumental variables approach where the dependent variable is R&D spending scaled by the previous year’s knowledge stock. The Compustat based measure of the Herfindahl-Hirschman Index (HHI) is used to split the sample into firms in highly competitive industries(HighComp.)andfirmsinlesscompetitiveindustries(LowComp.).R&DQisanalternativeproxyfor R&Drelatedinvestmentopportunitiesandcalculatedbyusingthesamenumeratorasinequation(5)butusing theknowledgestockasthedenominator.Controlsincludetotalassets,debt-to-equityratio,cashflowsscaledby capitalstock,salesandthe12-monthSSTlevelforecast.Standarderrorsareclusteredatthefirmlevel.***,**and *indicate1%,5%,and10%significance,respectively. AllFirms HighComp. LowComp. AllFirms HighComp. LowComp. (1) (2) (3) (4) (5) (6) Returnvol. -0.0500 (0.03) Tobin'sQ 0.0075*** 0.0945 (0.06) 0.0062*** (0.00) (0.00) -0.0941** -0.0209 0.1823*** -0.1277*** (0.04) (0.03) (0.05) (0.04) 0.0081*** (0.00) R&DQ 0.0001* (0.00) 0.0001 0.0000 (0.00) (0.00) N 24,834 14,110 10,724 31,588 17,391 14,197 F-Stat.-1stStage 181.45 103.63 65.90 262.56 176.58 100.70 Controls Yes Yes Yes Yes Yes Yes YearFE Yes Yes Yes Yes Yes Yes FirmFE Yes Yes Yes Yes Yes Yes 51 Table9 Theeffectofuncertaintyonadvertisingspendingundervaryinglevelsofcompetition This table presents results of estimating model (4) using an instrumental variables approach where the dependent variable is advertising spending scaled by total assets. The Compustat based measure of the Herfindahl-HirschmanIndex(HHI)isusedtosplitthesampleintofirmsinhighlycompetitiveindustries(High Comp.)andfirmsinlesscompetitiveindustries(LowComp.).AdvertisingQisanalternativeproxyforadvertising relatedinvestmentopportunitiesandcalculatedbyusingthesamenumeratorasinequation(5)butusingthe advertising stock as the denominator. Controls include total assets, debt-to-equity ratio, cash flows scaled by capitalstock,salesandthe12-monthSSTlevelforecast.Standarderrorsareclusteredatthefirmlevel.***,**and *indicate1%,5%,and10%significance,respectively. AllFirms HighComp. LowComp. AllFirms HighComp. LowComp. (1) (2) (3) (4) (5) (6) Returnvol. -0.3603*** -0.0836 -0.5022*** -0.3454*** -0.0596 -0.4779*** (0.05) (0.09) (0.08) (0.07) (0.06) Tobin'sQ 0.0118*** 0.0119*** (0.00) AdvertisingQ (0.04) 0.0048** (0.00) 0.0000 (0.00) 0.0000** (0.00) -0.0000 (0.00) (0.00) N F-Stat.-1st Stage 13,400 6,096 7,304 17,351 7,654 9,697 153.31 52.94 80.14 260.14 93.79 158.40 Controls Yes Yes Yes Yes Yes Yes YearFE Yes Yes Yes Yes Yes Yes FirmFE Yes Yes Yes Yes Yes Yes 52 Table10 Theeffectofuncertaintyonhiringundervaryinglevelsofcompetition This table presents results of estimating model (4) using an instrumental variables approach where the dependent variable is the number of employees scaled by total assets. The Compustat based measure of the Herfindahl-HirschmanIndex(HHI)isusedtosplitthesampleintofirmsinhighlycompetitiveindustries(High Comp.) and firms in less competitive industries (Low Comp.). Labor Q is an alternative proxy for labor related investmentopportunitiesandcalculatedbyusingthesamenumeratorasinequation(5)butusingthenumberof employeesasthedenominator.Controlsincludetotalassets,debt-to-equityratio,cashflowsscaledbycapital stock, sales and the 12-month SST level forecast. Standard errors are clustered at the firm level. ***, ** and * indicate1%,5%,and10%significance,respectively. AllFirms (1) (2) (3) Returnvol. -0.0164* (0.01) 0.0034 0.0003* AllFirms Low Comp. (4) (5) (6) -0.0116** 0.0032* (0.01) 0.0009* (0.00) High Comp. (0.01) 0.0000 -0.0247* (0.00) (0.00) LaborQ Low Comp. Tobin'sQ High Comp. -0.0213** (0.00) (0.01) -0.0000 -0.0000 -0.0000 (0.00) (0.00) (0.00) (0.00) N 48,231 22,644 25,587 60,935 28,094 32,841 F-Stat.-1stStage 264.42 82.28 143.84 325.65 249.05 159.62 Controls Yes Yes Yes Yes Yes Yes YearFE Yes Yes Yes Yes Yes Yes FirmFE Yes Yes Yes Yes Yes Yes 53 References Abel,A.B.,1983.OptimalInvestmentUnderUncertainty.AmericanEconomicReview73,228– 233. Abel, A.B., Eberly, J.C., 1996. Optimal Investment With Costly Reversibility. Review of EconomicStudies63,581–593. Abel, A.B., Eberly, J.C., 1994. A Unified Model Of Investment Under Uncertainty. American EconomicReview84,1369–1384. Abraham,K.G.,Medoff,J.L.,1984.LengthOfServiceAndLayoffsInUnionAndNonunionWork Groups.IndustrialandLaborRelationsReview38,87–97. Adams, R.M., Bryant, K.J., McCarl, B.A., Legler, D.M., O’Brien, J., Solow, A., Weiher, R., 1995. ValueOfImprovedLong-RangeWeatherInformation.ContemporaryEconomicPolicy13, 10–19. Adams, R.M., Chen, C.-C., McCarl, B.A., Weiher, R.F., 1999. The Economic Consequences Of ENSOEventsForAgriculture.ClimateResearch13,165–172. Aghion,P.,Bloom,N.,Blundell,R.,Griffith,R.,Howitt,P.,2005.CompetitionAndInnovation: AnInverted-URelationship.QuarterlyJournalofEconomics120,701–728. Aguerrevere, F.L., 2003. Equilibrium Investment Strategies And Output Price Behavior: A Real-OptionsApproach.ReviewofFinancialStudies16,1239–1272. Algieri,B.,2014.ARollerCoasterRide:AnEmpiricalInvestigationOfTheMainDriversOfThe InternationalWheatPrice.AgriculturalEconomics45,459–475. Ali, A., Klasa, S., Yeung, E., 2009. The Limitations Of Industry Concentration Measures Constructed With Compustat Data: Implications For Finance Research. Review of FinancialStudies22,3839–3871. Almeida,H.,Campello,M.,Galvao,A.F.,2010.MeasurementErrorsInInvestmentEquations. ReviewofFinancialStudies23,3279–3328. Angrist,J.D.,Krueger,A.B.,2001.InstrumentalVariablesAndTheSearchForIdentification: FromSupplyAndDemandToNaturalExperiments.JournalofEconomicPerspectives15, 69–85. Arellano, C., Bai, Y., Kehoe, P., 2010. Financial Markets And Fluctuations In Uncertainty, FederalReserveBankofMinneapolis,StaffReport466. Arellano,M.,Bond,S.,1991.SomeTestsOfSpecificationForPanelData:MonteCarloEvidence AndAnApplicationToEmploymentEquations.ReviewofEconomicStudies58,277–297. Bagwell, K., 2007. The Economic Analysis Of Advertising, in: Armstrong, M., Porter, R.H. 54 (Eds.),HandbookofIndustrialOrganization.Elsevier,pp.1701–1844. Baker, S., Bloom, N., 2013. Does Uncertainty Drive Business Cycles? Using Disasters As NaturalExperiments,NBERWorkingPaper19475. Bar-Ilan,A.,Strange,W.C.,1996.InvestmentLags.AmericanEconomicReview610–622. Baranchuk, N., Kieschnick, R., Moussawi, R., 2014. Motivating Innovation In Newly Public Firms.JournalofFinancialEconomics111,578–588. Baum,C.F.,Caglayan,M.,Talavera,O.,2008.UncertaintyDeterminantsOfFirmInvestment. EconomicsLetters98,282–287. Bernanke,B.S.,1983.Irreversibility,Uncertainty,AndCyclicalInvestment.QuarterlyJournal ofEconomics98,85–106. Berry, B.J.L., Okulicz-Kozaryn, A., 2008. Are There ENSO Signals In The Macroeconomy? EcologicalEconomics64,625–633. Bloom,N.,2014.FluctuationsInUncertainty.JournalofEconomicPerspectives28,153–175. Bloom,N.,2007.UncertaintyAndTheDynamicsOfR&D.AmericanEconomicReview97,250– 255. Bloom,N.,Bond,S.,VanReenen,J.,2007.UncertaintyAndInvestmentDynamics.Reviewof EconomicStudies74,391–415. Bloom,N.,Floetotto,M.,Jaimovich,N.,Saporta-Eksten,I.,Terry,S.J.,2014.ReallyUncertain BusinessCycles,WorkingPaper,StanfordUniversity. Bloom,N.N.,2009.TheImpactOfUncertaintyShocks.Econometrica77,623–685. Brennan, M.J., Schwartz, E.S., 1985. Evaluating Natural Resource Investments. Journal of Business58,135–157. Brogaard,J.,Detzel,A.,2015.TheAsset-PricingImplicationsOfGovernmentEconomicPolicy Uncertainty.ManagementScience61,3–18. Brunner,A.D.,2002.ElNiñoAndWorldPrimaryCommodityPrices:WarmWaterOrHotAir? ReviewofEconomicsandStatistics84,176–183. Bulan,L.T.,2005.RealOptions,IrreversibleInvestmentAndFirmUncertainty:NewEvidence FromU.S.Firms.ReviewofFinancialEconomics14,255–279. Bustamante, M.C., 2015. Strategic Investment And Industry Risk Dynamics. Review of FinancialStudies28,297–341. Byoun, S., Xu, Z., 2012. Shift In Capital Flows And Product Market Competition During A FinancialCrisis,WorkingPaper,PolytechnicInstituteofNewYorkUniversity. Caballero, R.J., 1991. On The Sign Of The Investment-uncertainty Relationship. American 55 EconomicReview279–288. Caballero, R.J., Pindyck, R.S., 1996. Uncertainty, Investment, And Industry Evolution. InternationalEconomicReview37,641. Campa,J.,Goldberg,L.S.,1995.InvestmentInManufacturing,ExchangeRatesAndExternal Exposure.JournalofInternationalEconomics38,297–320. Campa, J.M., 1993. Entry By Foreign Firms In The United States Under Exchange Rate Uncertainty.ReviewofEconomicsandStatistics75,614–622. Campbell, J.Y., Lettau, M., Malkiel, B.G., Xu, Y., 2001. Have Individual Stocks Become More Volatile?AnEmpiricalExplorationOfIdiosyncraticRisk.JournalofFinance56,1–43. Campello, M., 2006. Debt Financing: Does It Boost Or Hurt Firm Performance In Product Markets?JournalofFinancialEconomics82,135–172. Card,D.,1993.UsingGeographicVariationInCollegeProximityToEstimateTheReturnTo Schooling,WorkingPaper4483,NationalBureauofEconomicResearch. Carhart,M.M.,1997.OnPersistenceInMutualFundPerformance.JournalofFinance52,57– 82. Cashin,P.,Mohaddes,K.,Raissi,M.,2015.FairWeatherOrFoul?TheMacroeconomicEffects OfElNiño,WorkingPaper,UniversityofCambridge. Changnon,S.A.,1999.ImpactsOf1997—98ElNiñoGeneratedWeatherInTheUnitedStates. BulletinoftheAmericanMeteorologicalSociety80,1819–1827. Chen, C.-C., McCarl, B.A., Adams, R.M., 2001. Economic Implications Of Potential ENSO FrequencyAndStrengthShifts.ClimaticChange49,147–159. Chen, H.J., Kacperczyk, M., Ortiz-Molina, H., 2011. Labor Unions, Operating Flexibility, And TheCostOfEquity.JournalofFinancialandQuantitativeAnalysis46,25–58. Chimeli, A.B., De Souza Filho, F.D.A., Holanda, M.C., Petterini, F.C., 2008. Forecasting The ImpactsOfClimateVariability:LessonsFromTheRainfedCornMarketInCeará,Brazil. EnvironmentandDevelopmentEconomics13,201–227. Christiano,L.J.,Motto,R.,Rostagno,M.,2014.RiskShocks.AmericanEconomicReview104, 27–65. Chu,L.-F.,McAleer,M.,Chen,C.-C.,2012.HowVolatileIsENSOForGlobalGreenhouseGas EmissionsAndTheGlobalEconomy?JournalofReviewsonGlobalEconomics1,1–12. CPC, 2015a. Website Of The Climate Prediction Center [WWW Document]. URL www.cpc.noaa.gov(accessed7.16.15). CPC,2015b.ElNiño/SouthernOscillation(ENSO)DiagnosticDiscussion(October8,2015), Report,ClimatePredictionCenter. 56 CPC,2014.ElNiño/SouthernOscillation(ENSO)DiagnosticDiscussion(June5,2014),Report, ClimatePredictionCenter. Dell, M., Jones, B.F., Olken, B.A., 2014. What Do We Learn From The Weather? The New Climate-EconomyLiterature.JournalofEconomicLiterature52,740–798. Dixit, A.K., Pindyck, R.S., 1994. Investment Under Uncertainty. Princeton University Press, Princeton. DoJ,2010.HorizontalMergerGuidelines,Report,U.S.DepartmentofJusticeandtheFederal TradeCommission. Eberly, J.C., Van Mieghem, J.A., 1997. Multi-factor Dynamic Investment Under Uncertainty. JournalofEconomicTheory75,345–387. Eisfeldt, A.L., Rampini, A.A., 2006. Capital Reallocation And Liquidity. Journal of Monetary Economics53,369–399. Erickson, T., Whited, T.M., 2012. Treating Measurement Error In Tobin’s Q . Review of FinancialStudies25,1286–1329. Fama, E.F., French, K.R., 1992. The Cross-Section Of Expected Stock Returns. Journal of Finance47,427–465. Fazzari,S.M.,Petersen,B.C.,1993.WorkingCapitalAndFixedInvestment:NewEvidenceOn FinancingConstraints.RANDJournalofEconomics24,328–342. Ferderer, J.P., 1993. The Impact Of Uncertainty On Aggregate Investment Spending: An EmpiricalAnalysis.JournalofMoney,CreditandBanking25,30–48. Fernández-Villaverde, J., Guerrón-Quintana, P., Rubio-Ramírez, J.F., Uribe, M., 2011. Risk Matters: The Real Effects Of Volatility Shocks. American Economic Review 101, 2530– 2561. Fischer, G., 2013. Investment Choice And Inflation Uncertainty, Working Paper, London SchoolofEconomics. Frank,M.Z.,Goyal,V.K.,2003.TestingThePeckingOrderTheoryOfCapitalStructure.Journal ofFinancialEconomics67,217–248. Frésard,L.,Valta,P.,2013.CompetitivePressureAndCorporateInvestment:EvidenceFrom TradeLiberalization,WorkingPaper. FT, 2015. How A Strong El Niño Affects Stocks: Macquarie [WWW Document]. URL http://www.ft.com/fastft/415031/stocks-affected-el-nino(accessed11.1.15). Gaspar,J.,Massa,M.,2006.IdiosyncraticVolatilityAndProductMarketCompetition.Journal ofBusiness79,3125–3152. Gavazza,A.,2011.TheRoleOfTradingFrictionsInRealAssetMarkets.AmericanEconomic 57 Review101,1106–1143. Ghosal,V.,Loungani,P.,2000.TheDifferentialImpactOfUncertaintyOnInvestmentInSmall AndLargeBusinesses.ReviewofEconomicsandStatistics82,338–343. Ghosal, V., Loungani, P., 1996. Product Market Competition And The Impact Of Price UncertaintyOnInvestment:SomeEvidenceFromUSManufacturingIndustries.Journal ofIndustrialEconomics44,217–228. Gilchrist, S., Sim, J., Zakrajšek, E., 2014. Uncertainty, Financial Frictions, And Investment Dynamics,WorkingPaperNo.20038,NationalBureauofEconomicResearch. Goldberg, L.S., 1993. Exchange Rates And Investment In United States Industry. Review of EconomicsandStatistics75,575–588. Gramm,C.L.,Schnell,J.F.,2001.TheUseOfFlexibleStaffingArrangementsInCoreProduction Jobs.Industrial&LaborRelationsReview54,245–258. Grenadier,S.R.,2002.OptionExerciseGames:AnApplicationToTheEquilibriumInvestment StrategiesOfFirms.ReviewofFinancialStudies15,691–721. Grullon, G., Lyandres, E., Zhdanov, A., 2012. Real Options, Volatility, And Stock Returns. JournalofFinance67,1499–1537. Guiso, L., Parigi, G., 1999. Investment And Demand Uncertainty. Quarterly Journal of Economics114,185–227. Gujarati,D.,2003.BasicEconometrics,4thed.McGrawHillInc.,NewYork,NY. Hall,B.H.,Mairesse,J.,1995.ExploringTheRelationshipBetweenR&DAndProductivityIn FrenchManufacturingFirms.JournalofEconometrics65,263–293. Handler, P., Handler, E., 1983. Climatic Anomalies In The Tropical Pacific Ocean And Corn YieldsInTheUnitedStates.Science220,1155–1156. Hansen, J.W., Hodges, A.W., Jones, J.W., 1998. ENSO Influences On Agriculture In The SoutheasternUnitedStates.JournalofClimate11,404–411. Hartman, R., 1972. The Effects Of Price And Cost Uncertainty On Investment. Journal of EconomicTheory5,258–266. Hayashi, F., 1982. Tobin’s Marginal Q And Average Q: A Neoclassical Interpretation. Econometrica:JournaloftheEconometricSociety213–224. Hirsch,B.T.,MacPherson,D.A.,2003.UnionMembershipAndCoverageDatabaseFromThe CurrentPopulationSurvey:Note.Industrial&LaborRelationsReview56,349–354. Hoberg,G.,Philips,G.,2010.RealAndFinancialIndustryBoomsAndBusts.JournalofFinance 65,45–86. Hou,K.,Robinson,D.T.,2006.IndustryConcentrationAndAverageStockReturns.Journalof 58 Finance61,1927–1956. Hsiang, S.M., Meng, K.C., Cane, M.A., 2011. Civil Conflicts Are Associated With The Global Climate.Nature476,438–441. Huizinga,J.,1993.InflationUncertainty,RelativePriceUncertainty,AndInvestmentInU.S. Manufacturing.JournalofMoney,Credit&Banking25,521–549. Ilut,C.,Schneider,M.,2012.AmbiguousBusinessCycles.WorkingPaperNo.17900,National BureauofEconomicResearch. Irvine,P.J.,Pontiff,J.,2009.IdiosyncraticReturnVolatility,CashFlows,AndProductMarket Competition.ReviewofFinancialStudies22,1149–1177. Julio, B., Yook, Y., 2012. Political Uncertainty And Corporate Investment Cycles. Journal of Finance67,45–83. Kellogg,R.,2014.TheEffectOfUncertaintyOnInvestment:EvidenceFromTexasOilDrilling. AmericanEconomicReview104,1698–1734. Kim, H., Kung, H., 2014. The Asset Redeployability Channel: How Uncertainty Affects CorporateInvestment,WorkingPaper. Kogan, L., 2001. An Equilibrium Model Of Irreversible Investment. Journal of Financial Economics62,201–245. Kulatilaka,N.,Perotti,E.C.,1998.StrategicGrowthOptions.ManagementScience44,1021– 1031. Leahy,J.V,1993.InvestmentInCompetitiveEquilibrium:TheOptimalityOfMyopicBehavior. QuarterlyJournalofEconomics108,1105–1133. Leahy,J.V,Whited,T.M.,1996.TheEffectOfUncertaintyOnInvestment:SomeStylizedFacts. JournalofMoney,CreditandBanking28,64–83. Lee,J.,Shin,K.,2000.TheRoleOfAVariableInputInTheRelationshipBetweenInvestment AndUncertainty.AmericanEconomicReview90,667–680. Lindenberg, E.B., Ross, S.A., 1981. Tobin’s Q Ratio And Industrial Organization. Journal of Business54,1–32. MacRae, K.M., 1989. Critical Issues In Electric Power Planning In The 1990s: Executive Summary.CanadianEnergyResearchInstitute,Calgary. Marschak, T., Nelson, R., 1962. Flexibility, Uncertainty, And Economic Theory. Metroeconomica14,42–58. McDonald, R.L., Siegel, D., 1987. The Value Of Waiting To Invest. Quarterly Journal of Economics101,707–727. Mills,D.E.,1984.DemandFluctuationsAndEndogenousFirmFlexibility.JournalofIndustrial 59 Economics33,55–71. Minton, B.A., Schrand, C., 1999. The Impact Of Cash Flow Volatility On Discretionary InvestmentAndTheCostsOfDebtAndEquityFinancing.JournalofFinancialEconomics 54,423–460. Naylor, R., Falcon, W., Rochberg, D., Wada, N., 2001. Using El Niño/Southern Oscillation ClimateDataToPredictRiceProductionInIndonesia.ClimaticChange50,255–265. Nickell,S.J.,1996.CompetitionAndCorporatePerformance.JournalofPoliticalEconomy104, 724–746. Oi,W.Y.,1961.TheDesirabilityOfPriceInstabilityUnderPerfectCompetition.Econometrica: JournaloftheEconometricSociety58–64. Panousi,V.,Papanikolaou,D.,2012.Investment,IdiosyncraticRisk,AndOwnership.Journal ofFinance67,1113–1148. Peress,J.,2010.ProductMarketCompetition,InsiderTrading,AndStockMarketEfficiency. JournalofFinance65,1–43. Pindyck, R.S., 1993a. A Note On Competitive Investment Under Uncertainty. American EconomicReview83,273–277. Pindyck,R.S.,1993b.InvestmentsOfUncertainCost.JournalofFinancialEconomics34,53– 76. Pindyck, R.S., 1991. Irreversibility, Uncertainty, And Investment. Journal of Economic Literature29,1110–1148. Rasmusson,E.M.,Carpenter,T.H.,1982.VariationsInTropicalSeaSurfaceTemperatureAnd Surface Wind Fields Associated With The Southern Oscillation/El Niño. Monthly WeatherReview110,354–384. Ropelewski, C.F., Halpert, M.S., 1987. Global And Regional Scale Precipitation Patterns AssociatedWithTheElNiño/SouthernOscillation.MonthlyWeatherReview115,1606– 1626. Ropelewski, C.F., Halpert, M.S., 1986. North American Precipitation And Temperature Patterns Associated With The El Niño/Southern Oscillation (ENSO). Monthly Weather Review114,2352–2362. Rosenzweig, C., Iglesias, A., Yang, X.B., Epstein, P., Chivian, E., 2001. Climate Change And ExtremeWeatherEvents;ImplicationsForFoodProduction,PlantDiseases,AndPests. GlobalChangeandHumanHealth2,90–104. Salinger, M., Summers, L.H., 1983. Tax Reform And Corporate Investment: A Microeconometric Simulation Study, in: Behavioral Simulation Methods in Tax Policy Analysis.UniversityofChicagoPress,pp.247–288. 60 Servén,L.,2003.Real-Exchange-RateUncertaintyAndPrivateInvestmentInLDCS.Reviewof EconomicsandStatistics85,212–218. Solow,A.R.,Adams,R.F.,Bryant,K.J.,Legler,D.M.,O’Brien,J.J.,McCarl,B.A.,Nayda,W.,Weiher, R.,1998.TheValueOfImprovedENSOPredictionToU.S.Agriculture.ClimaticChange 39,47–60. Stein, L.C.D., Stone, E., 2013. The Effect Of Uncertainty On Investment, Hiring, And R&D: CausalEvidenceFromEquityOptions,WorkingPaper. Storesletten,K.,Telmer,C.I.,Yaron,A.,2004.CyclicalDynamicsInIdiosyncraticLaborMarket Risk.JournalofPoliticalEconomy112,695–717. Theil,H.,1971.PrinciplesOfEconometrics.JohnWiley&Sons,Inc.,NewYork,NY. Tol,R.S.J.,2009.TheEconomicEffectsOfClimateChange.JournalofEconomicPerspectives 29–51. Ubilava,D.,2012.ElNiño,LaNiña,AndWorldCoffeePriceDynamics.AgriculturalEconomics 43,17–26. Wang,B.,Wu,R.,Fu,X.,2000.Pacific-EastAsianTeleconnection:HowDoesENSOAffectEast AsianClimate?JournalofClimate13,1517–1536. Warusawitharana, M., 2008. Corporate Asset Purchases And Sales: Theory And Evidence. JournalofFinancialEconomics87,471–497. Williams,J.T.,1993.EquilibriumAndOptionsOnRealAssets.ReviewofFinancialStudies6, 825–850. Wolter,K.,Timlin,M.S.,1998.MeasuringTheStrengthOfENSOEvents:HowDoes1997/98 Rank?Weather53,315–324. Zhang,L.,2005.TheValuePremium.JournalofFinance60,67–103. 61 Appendix A DataandVariableConstruction Thissectiondescribesthedatathatisusedforthemainanalysisandtheconstructionof relevantvariablesinmoredetail. Iusefirm-levelfinancialdataobtainedfromtheCompustatdatabase,whichIsupplement withreturndatafromCRSP.Thisdatasetismergedwithseasurfacetemperature(SST)and SST prediction data from the Climate Prediction Center (CPC) at the National Oceanic and Atmospheric Administration. This results in a data sample of more than 12,000 firms and 90,000firm-yearobservation.Thissampleisfurtherrestrictedbyexcludingindustriesfor -./0 whichtheindustry-levelsensitivitytoENSO,)*,, ,estimatedinmodel(2)isnotstatistically differentfromzero(atthefivepercentlevel).Thefinalanalysissamplecomprises6,695firms and51,203firm-yearobservations. The construction of variables included in the estimation of model (4) follows standard approaches in the existing literature.12,, in equation (4) is capital expenditure by firm i in periodt,whichisscaledbythecapitalstockattheendofthepreviousperiod32,,$& .Since financial statements report capital stock at book rather than replacement value, 32,, is calculatedrecursivelyusingaperpetual-inventorymethod.Ibroadlyfollowthemethodology utilized in Salinger and Summers (1983) and Stein and Stone (2013) and set the first observation for each firm spell equal to the book value of property, plant and equipment (PPE): 32,' = 5562,' Inthenextstep,Irecursivelycalculatethereplacementvalueofcapitalstockforeachfirmin eachperiodtasfollows 32,, = 551,7 1 − : 7 32,,$& + 12,, 7 551,$& 62 where551,7 is the Producer Price Index for Finished Capital Equipment Goods,12,, is the capitalexpenditureforfirmiinperiodtand: 7 isthedepreciationrate(assumedtobe10% asinSteinandStone(2013)38). I follow the same methodology to calculate the “stock” variable used to scale R&D spending.Duetotheunavailabilityofabookvalueforthisvariable(equivalenttoproperty, plantandequipmentforcapitalinvestment),foreachfirmspellthevariableisinitializedas <=>2,' = <=2,' + A?@/ : ?@/ where<=2,' isR&Dspendinginthefirstyearoffirm’sispellofdata.AsinHallandMairesse (1995),IuseadepreciationrateforR&Dstock,: ?@/ ,of15%andagrowthrate,A?@/ ,of5% (netofdepreciation).TheR&Dstockissubsequentlydeflatedusingtheaggregatepriceindex forprivatebusinessR&Dspending,whichisobtainedfromtheBEA.Thestockvariablefor advertisingspendingissimilarlycalculatedusingagrowthrateof5%andadepreciationrate of50%,sinceestimateddepreciationratesforadvertisingrangefrom15%to80%(Bagwell, 2007).Investmentinlaborisconstructedasthenumberofemployeesscaledbythefirm’s totalassets. Finally,thecapitalinvestmentrate,theR&Dspendingrate,andtheadvertisinginvestment ratearewinsorizedtoliebetween-0.1and1,0and1,and0and1.5,respectively. ItisstandardpracticetoincludemeasuresofTobin’sqascontrolsininvestmentmodels. Empirically,itischallengingtomeasureafirm’smarginalTobin’sq,definedastheratioof themarketvalueandthereplacementcostofanadditionalunitofcapital.Thus,Iincludethe followingmeasureofaverageTobin’sq: B2,, = CDEFGH!DJKHDLKMDHKNO2,, + =GPH2,, − !QEEGOHRSSGHS2., 32,, + 1OUGOHNEV2,, + 1OHDOAKPLGS2,, + 1OUGSHWGOHSDOXDXUDOYGS2,, which represents the ratio between the market value of a firm’s capital stock and its replacementcost.Othercontrolvariablesincludedintheestimationsthroughoutthepaper 38Theresultspresentedinthepaperarerobusttoassumingadepreciationrateof8%asinBloom etal.(2007). 63 are cash flow scaled by previous year’s capital stock, debt-to-equity ratio, total assets and sales. Changeinnetworkingcapital,netdebtandnetequityissuearecalculatedaccordingto FrankandGoyal(2003).Intheconstructionofworkingcapital,Idistinguishbetweenfirms withreportingformatcode1,2,3and7.Itemnumbersdenotethenumbersassociatedwith eachCompustatvariable. Reportingformatcode1:ZNetWorkingcapital=Item236+Item274+Item301 Reportingformatcode2,3:ZNetWorkingcapital=-Item236+Item274-Item301 Reportingformatcode7:ZNetWorkingcapital=-Item302-Item303-Item304-Item305 -Item307+Item274-Item312-Item301 NetdebtandnetequityissuesarecalculatedusingthefollowingCompustatdataitems.Net debtissuesequalsItem111minusItem114andnetequityissuesequalsItem108minus Item115. 64 B ElNiñoSouthernOscillation(ENSO)–AdditionalDetails ChangesinENSOconditionscanhavealarge-scaleimpactonweatheraroundtheworld. FigureB.1illustratestheclimatologicaleffectsofwarm(ElNiño)andcoldepisodes(LaNiña). Thesechangesinweatherpatternshavepronouncedeffectsoncommodity-producingand constructionindustriesaswellasglobalandnationalcommodityprices(Cashinetal.,2015). Asanillustration,FigureB.1showsthatwarmepisodescanresultinunusuallywarmwinters in large parts of the U.S., while cold episodes have the opposite effect. These changes can significantly impact energy demand in the U.S. and thus the price of commodities such as naturalgasandheatingoil.Amoredetaileddiscussionoftheimpactoftheseverestandmost recentElNiñoeventondifferentindustriesintheU.S.andtheoverallU.S.economycanbe foundinChangnon(1999). TheintensityofENSOeventscanbemeasuredusingseasurfacetemperature(SST)and sea-level air pressure in the Pacific Ocean. The most commonly used measure for ENSO activity in the academic literature is based on sea surface temperature measurements for Pacificregion“Niño3.4”(see,e.g.,Brunner,2002)andcanbedownloadedfromthewebsite oftheClimatePredictionCenter,whichispartoftheU.S.NationalOceanicandAtmospheric Administration (CPC, 2015a). This data is also used by the CPC to predict future ENSO conditions. B.1 CPCWeeklyandMonthlyBulletins Starting in the early 1980s, the Climate Prediction Center (CPC) has been summarizing current and future expected ENSO activity in weekly and monthly bulletins, which are availableontheirwebsite.Theyprovideaplethoraofinformationaboutcurrentconditions suchashistoricaldeviationsofseasurfacetemperaturefromhistorical,regionalmeans.As anexample,FigureB.2,whichwaspartoftheweeklyENSOupdatepublishedonNovember 9,2015,illustratesSSTdeviationsintheTropicalPacificintheperiodfromOctober11,2015 toNovember7,2015.Thepositivedeviationsillustratedinthatfigureareconsistentwiththe strongElNiñoeventexperiencedintheyear2015. Along with data about current ENSO conditions, the weekly and monthly bulletins also contain extensive information about expected future ENSO activity in the form of probabilistic forecasts of ENSO events (see Figure B.3) and forecasts of future sea surface temperature (see Figure B.4). Figure B.4 illustrates consolidation forecasts of sea surface 65 temperature that are based on several statistical models. The graph also indicates the standarddeviationsfortheseforecasts. The graphs illustrating ENSO forecasts are also accompanied by comprehensive discussions of current and future expected ENSO conditions. In particular, the degree of uncertaintyaboutfutureENSOconditionsisusuallyclearlystated.Thisisillustratedinrecent reports issued by the Climate Prediction Center. As an example, consider the “El Niño/Southern Oscillation (ENSO) Diagnostic Discussion” published by the Climate PredictionCenter,theNationalCentersforEnvironmentalPrediction(NCEP),theNational WeatherService(NWS),andtheInternationalResearchInstituteforClimateandSocietyat Columbia University on June 5, 2014. The discussion states that over “the last month, the chance of El Niño and its ultimate strength weakened slightly in the models (Fig. 6). Regardless,theforecastersremainjustasconfidentthatElNiñoislikelytoemerge.IfElNiño forms, the forecasters and most dynamical models, such as NCEP CFSv2, slightly favor a moderate-strengtheventduringtheNorthernHemispherefallorwinter(3-monthvaluesof theNiñ o-3.4indexbetween1.0°Cand1.4°C).However,significantuncertaintyaccompanies thisprediction,whichremainsinclusiveofaweakerorstrongereventduetothespreadofthe models and their skill at these lead times.” (CPC, 2014). In particular, the last sentence indicatesthatthepredictionwasaccompaniedbysubstantialuncertaintyaboutthestrength ofthepredictedElNiñoevent.Incontrast,amorerecentreportissuedonOctober8,2015 shows that the ENSO forecast at that time was considered to be associated with less uncertainty:“AllmodelssurveyedpredictElNiñotocontinueintotheNorthernHemisphere spring2016,andallmulti-modelaveragespredictapeakinlatefall/earlywinter(Fig.6).The forecaster consensus unanimously favors a strong El Niño, with peak 3-month SST departuresintheNiño3.4regionnearorexceeding+2.0°C.Overall,thereisanapproximately 95% chance that El Niño will continue through Northern Hemisphere winter 2015-16, graduallyweakeningthroughspring2016(clickCPC/IRIconsensusforecastforthechanceof eachoutcomeforeach3-monthperiod).”(CPC,2015b). These examples illustrate that there is significant variation in the uncertainty that accompanies the predictions of ENSO conditions. And this uncertainty is clearly communicated in the weekly and monthly bulletins published by the CPC in the form of quantitativegraphsandcomprehensivediscussions. 66 FigureB.1 GlobaleffectsofENSO ThisfigureillustratestheglobalimpactoffluctuationsinElNiñoSouthernOscillation(ENSO)activityon theclimateandweatherconditionsaroundtheworld.Source:ClimatePredictionCenter,NationalOceanicand AtmosphericAdministration(NOAA). FigureB.2 SeasurfacetemperaturedeviationsintheTropicalPacific This figure illustrates the monthly departures of sea surface temperatures from historical averages throughout the Tropical Pacific for the period between October 11, 2015 and November 7, 2015. This informationisprovidedonaweeklybasisbytheClimatePredictionCenter.Source:ClimatePredictionCenter, NationalOceanicandAtmosphericAdministration(NOAA). 67 FigureB.3 ProbabilisticforecastofENSOevents ThisfigureillustratestheprobabilisticforecastsofENSOeventsfortheperiodbetweenSeptember,2015 and July, 2016. This information is provided on a weekly basis by the Climate Prediction Center. Source: ClimatePredictionCenter,NationalOceanicandAtmosphericAdministration(NOAA). FigureB.4 ProbabilisticforecastofseasurfacetemperatureanomaliesinthePacific ThisfigureillustratesconsolidationforecastsofseasurfacetemperatureinthePacificOceanfortheperiod from November, 2015 to January, 2017. This information is provided on a monthly basis by the Climate PredictionCenter.Source:NationalOceanicandAtmosphericAdministration(NOAA). 68 C Return-basedUncertaintyMeasuresandPotentialBiases Theexistingempiricalliteratureoninvestmentunderuncertaintytakesmanydifferent approachestomeasuringuncertainty.Themostcommonlyuseduncertaintymeasuresare basedonvolatilityindicessuchastheVIXfromtheChicagoBoardofOptionsExchangeoron firm-levelstockreturnvolatility(see,e.g.,Bloometal.,2007;KimandKung,2014;Leahyand Whited,1996).Thesearegenerallyrathernoisyproxiesforuncertaintyandoftencapture changesinthefirstmomentofinvestmentopportunitiesratherthanchangesinuncertainty. Without using an exogenous measure of uncertainty or appropriate instruments for returnvolatility,itisdifficulttocontrolforfactorsthatarecorrelatedwithreturnvolatility andatthesametimeaffectfirminvestment39.Therearemanyexamplesthatcanbeusedto emphasize the potential endogeneity of return volatility as an uncertainty proxy in an investment regression. However, to illustrate this point, I will specifically focus on two possible confounding factors that will lead to biased OLS estimates when using return volatilityasanuncertaintyproxy.Inparticular,thesebiaseswillbedifferentdependingon thelevelofproductcompetitionthatafirmfaces. A growing body of empirical research has linked product market competition to stock returnbehaviorandvolatility.Variousempiricalstudiesdocumentthatidiosyncraticreturn volatilityusuallyincreasesinresponsetohigherproductmarketcompetition (Gasparand Massa,2006;IrvineandPontiff,2009;Peress,2010).Inaddition,FrésardandValta(2013) illustrate the effect of tariff reductions and the associated increase in competition from foreign rivals on firm-level investment and financing choices. They find that higher competition is associated with substantial decreases in capital expenditure and R&D. This effect is economically large and heterogeneous depending on the ex-ante competitive structure of the product market. While more competition decreases investment in competitive industries, similar changes in more concentrated industries have virtually no effect on investment. Combined, these results present several empirical challenges to identifyingacausallinkbetweenuncertaintyandinvestmentwhenusingidiosyncraticstock returnvolatilityasanuncertaintyproxy.Ifreturnvolatilityispositivelycorrelatedwiththe degreeofproductmarketcompetitionandproductmarketcompetitionleadstodecreasesin 39Itcaneasilybeshownthatgivenatruemodel,V = [& \&2, + [% \%2, + Q2, ,omittingvariable\%2, willleadtoabiasedOLSestimate,J lim [& = [& + J lim : [% ,if[% ≠ 0and!NU(\& , \% ) ≠ 0. 2, .→a .→a 69 investment for firms in highly competitive industries, OLS estimates of the coefficient on returnvolatilitywillhaveastrongdownwardbiasincompetitiveindustries.UsingOLSin such a setting may lead to the inference that firms in competitive industries reduce investmentmoresharplyinresponsetoanincreaseinuncertaintythanisactuallythecase. Moreover,HobergandPhilips(2010)relatestockreturnsandoperatingperformancein industryboomsandbuststoindustry-levelcompetition.Thebasicintuitionbehindtheiridea relates to information gathering costs. It is costlier to collect information about a large number of competitors compared to few rivals and therefore firms in highly competitive industries rely more on signals conveyed by industry-level stock returns when taking investmentdecisions.Inconcentratedindustries,monopolypowerdecreasesthedispersion ofearningsforecasts(GasparandMassa,2006)andallowsforfasterincorporationofprivate informationintostockprices(Peress,2010).Thesefirmswillthusrelymoreonfirm-level information.Theauthorsfindthatfirmsincompetitiveindustriesoverinvestaboveoptimal levels in response to positive industry-level shocks and experience sharp decreases in operating cash flow and stock returns in the following periods. This relationship is much weakerandmostlyinsignificantformoreconcentratedindustries.Inasimilarargumentto above,ifreturnvolatilitycapturesthecyclicalnatureofindustryboomsandbustsandiffirms incompetitiveindustriesfacelargerdecreasesinoperatingperformanceinbustperiodsthan theircounterpartsinmoreconcentratedindustries,theOLSestimateoftheuncertaintyeffect willagainbedownwardbiasedforcompetitivefirms. Itisthereforeessentialtoidentifyexogenousvariationinuncertaintytodrawanycausal inferencesabouttheinvestment-uncertaintyrelationshipanditsinteractionwithindustry competition. In this paper, I exploit the unique empirical setting of the El Niño Southern Oscillation(ENSO)cycletoidentifyexogenouschangesinuncertaintythatarerelevantfora wide array of firms in different industries. See Section 4 for a detailed description of the specificidentificationstrategyfollowedinthisstudy. 70 D AdditionalResultsandRobustnessTests D.1 AdditionalSpecificationsofModel(4) Oneofthekeyresultsofthispaperisthatuncertainty,onaverage,hasapositiveeffecton investmentforfirmsincompetitiveindustriesandanegativeimpactoninvestmentforfirms inconcentratedindustries.TheseresultsarediscussedinSection5.1andpresentedinPanel BofTable2.Theestimationsarebasedonspecificationsofmodel(4)thatincludeanumber ofdifferentcontrolvariables,andfirmandyearfixedeffects. In order to confirm the robustness of these findings, Table D.1 presents results for estimationsofsimplerspecificationswithoutcontrolvariables.Columns1to3,columns4to 6,andcolumns7to9showestimatesforallfirmsintheanalysissample,forthesubsample offirmsincompetitiveindustries,andforthesubsampleoffirmsinconcentratedindustries, respectively.Thecoefficientsincolumns3,6and9representthesameestimatesillustrated in Panel B of Table 2. The findings in Table D.1 confirm that excluding control variables (columns1,4and7)oraverageTobin’sq(columns2,5and8)hasnosignificantimpacton thecoefficientestimatesfortheuncertaintymeasure. D.2 FullSampleofCompustatFirms In Section 5, I document the effect of uncertainty on investment using a multi-step procedure. Before estimating the instrumental variables specification, I identify firms that -./0 are sensitive to ENSO by estimating the industry specific coefficient)*,, in model (2). Subsequently, I drop all industries from my sample for which this coefficient is not statistically significant (at the 5 percent level). This step is meant to reduce possible measurementerrorintheinstrument.IndustriesnotsensitivetoENSOfluctuationsshould theoretically have a sensitivity of zero, which effectively eliminates all the variation in uncertaintyforthesefirms.Therefore,identificationoftheuncertaintycoefficientintheIV estimation will only come from firms with non-constant uncertainty values. However, in practice,smallsamplesizesincertainindustriesthatarenotsensitivetoENSOmayleadto animpreciseestimationofthesensitivitycoefficientsinmodel(2).Whilethecoefficientfor such an industry is likely not statistically significant, the point estimate might still be substantially positive or negative if standard errors are large. This would introduce measurement error in the instrument and would lead to biased IV estimates of the 71 uncertaintycoefficient.Nevertheless,forrobustnesspurposesTableD.2presentstheresults oftheIVestimatesforcapitalinvestmentusingthefullsampleoffirmswithoutdiscarding industries that are not sensitive to changes in ENSO conditions. Firms on average reduce investmentandthiseffectisstrongerforfirmsinconcentratedindustries.Incontrast,firms facing substantial product market competition increase investment. The findings are qualitativelyconsistentwiththeresultspresentedinPanelBofTable2,butthemagnitudeof theuncertaintyeffectissomewhatsmaller.ThesmallercoefficientsinTableD.2areinline withanerrors-in-variablesargumentsuggestingthatmeasurementerrorintheENSObased uncertaintyproxymightbiasthoseestimatestowardszero. D.3 ENSOSensitivityCalculationBasedonFour-FactorModel Icalculateindustry-levelsensitivitiestochangesinENSOconditionsbyestimatingmodel (2).Thisinvolvesregressingmonthlyfirm-levelreturnsonreturnsoftheS&P500indexand -./0 monthlyseasurfacetemperaturemeasurements.Theresultingsensitivityestimates,)*,, , arekeyingredientstotheconstructionofmyinstruments. Totesttheirrobustness,Ialsoestimatethefollowingversionofmodel(2)inthesameway asdescribedinSection4.1.1 -./0 -./0 E2,f = )'2 + gih jk + )*,, ∙ Df + m2,f (D.1) wherejk isavectorofobservableriskfactors.Iemployafour-factormodelusingthethree factors of Fama and French (1992), which I augment with the momentum risk factor in -./0 Carhart(1997).Usingtheresultingestimates)*,, intheconstructionofmyinstruments describedinSection4andestimatingmodel(4)yieldstheresultspresentedinTableD.3.The coefficients throughout this table are not statistically significantly different from the coefficientsinPanelBofTable2.Thus,themainfindingsofthepaperarerobusttousing model(D.1)toestimateindustryENSOsensitivities:inresponsetoheighteneduncertainty, firms in competitive industries increase investment and firms in concentrated industries decreaseinvestment. D.4 TimeInvariantENSOSensitivities -./0 Theindustry-levelsensitivities,)*,, ,arecalculatedbyestimatingmodel(2)onafive-./0 yearrollingbasis(asdescribedinSection4.1.1).Foreachyeart,)*,, iscalculatedbyusing 72 monthlyreturnsofthepreviousfiveyears(fromt-6tot-1).Oneconcernmightbethatfirms inagivenindustrysystematicallychooseinvestmentdecisionstochangetheirindustry-level sensitivitiestoENSO.Tomitigatethisconcern,inanadditionalrobustnesstest,Icalculate industry-levelENSOsensitivitiesbyestimatingmodel(2)usingmonthlyreturndatainthe years 1980 to 1984, which is prior to my main estimation period (1985 to 2014). The resulting time invariant sensitivities,)*-./0 , are then used to construct the instruments as describedinSection4andestimatemodel(4).TheresultsarepresentedinTableD.4.The main conclusion that firms in competitive industries increase investment and firms in concentrated industries decrease investment in response to heightened uncertainty still holds. D.5 CompetitionasanAdditionalRiskFactorintheCalculationofENSO Sensitivities Icalculateindustry-levelsensitivitiestochangesinENSOconditionsbyestimatingmodel (2) and using stock return data for firms and the S&P 500 as well as monthly sea surface temperaturemeasurements.Oneconcerncouldbethatanindustry’ssensitivitytoachange in ENSO conditions might systematically capture the level of competition in that industry. Thus,thedifferentialeffectofuncertaintyacrossfirmsinindustrieswithvaryingdegreesof competitioncouldbearesultofsystematicdifferencesinthesensitivitiesoftheseindustries to changes in ENSO conditions. To mitigate this concern, I include the Compustat HHI measureinmodel(2)asanadditionalriskfactor.Estimatingmodel(4)usingtheresulting -./0 sensitivities,)*,, ,intheconstructionoftheinstrumentsyieldsresultspresentedinTable D.5.Column2showsthatuncertainty,onaverage,hasapositiveeffectonfirminvestmentin competitive industries, while it has a negative effect on investment in concentrated industries.ThesefindingsareconsistentwiththeresultsinPanelBofTable2.However,the magnitudesoftheeffectsaresmaller. D.6 Industry-LevelENSOSensitivitiesBasedonTwo-DigitSICCodes Section4.1.1describeshowIestimatemodel(2)tocalculatetime-varying,industry-level sensitivitiestochangesinENSOconditions.Theindustryclassificationisbasedonthree-digit SICcodes.However,forrobustnesspurposes,Icalculatethesesensitivitiesforeachindustry based on two-digit SIC codes. The results are presented in Table D.6. The signs of the coefficientsareconsistentwiththeestimatesinPanelBofTable2.However,thenegative 73 effectofuncertaintyinconcentratedindustryisstronger,whiletheaveragepositiveeffectof uncertaintyoninvestmentincompetitiveindustriesisslightlysmaller. D.7 UnpredictableComponentofReturnVolatilityasUncertaintyProxy In model (4), I use firm-level stock return volatility as an uncertainty proxy. Due to its endogenous nature, I construct instruments based on ENSO prediction data to identify exogenous variation in return volatility. Instead of examining all of the variability in firmlevelstockreturns,Icanremovetheforecastablevariationbyestimatingthefollowinglinear factormodel(Gilchristetal.,2014) o E2,n − En = p2 + qih jr + Q2,n o whereE2,n aredailystockreturnsandEn istherisk-freerate.jr isavectorincludingthethree riskfactorsinFamaandFrench(1992)aswellasthemomentumfactorsuggestedbyCarhart (1997).Thevolatilityoftheunpredictablepartofthefirm-levelstockreturnreturnisthen calculatedas s2,, = 1 =, @t nu& % Q2,n − Q2,, wherethesamplemeanofdailyfirm-levelreturnsinyeartisdefinedasQ2,, = & @t @t nu& Q2,n . Theabovemeasure,s2,, ,representstheannualfirm-levelstockreturnvolatilityforfirmiin yeartvoidoftheforecastablevariationinexpectedreturns40.Instrumentingthismeasure insteadoftherawstockreturnvolatilityinmodel(4)yieldstheresultspresentedinTable D.7.ThecoefficientsshownarenotsignificantlydifferentfromtheonespresentedinPanelB ofTable2. D.8 TheEffectofUncertaintyonLeverage(Debt-EquityRatio) In Section 6.1.3, I test alternative explanations for the negative effect of uncertainty on investmentobservedinconcentratedindustries.Oneoftheseexplanationsisbasedonthe argument that under financing frictions, uncertainty might affect investment through borrowing costs. More specifically, Gilchrist et al. (2014) argue that a rise in uncertainty 40Toannualizethismeasure,theaveragedailyreturnvolatilityineachyearismultipliedbythe squarerootoftradingdaysinthatyear. 74 increasescreditspreadsandleadsfirmstodecreasetheirleverageandinvestmentspending. Examining external financing decisions of firms in concentrated industries in response to heighteneduncertainty,Ifindnoevidenceforthischannel.Columns7and8inPanelAof Table5illustratethatuncertaintyhasnoeffectonnetdebtissueandresultsindecreasednet equityissuesforsuchfirms.Asanadditionaltest,Iexaminetheeffectofuncertaintyonthe overall debt-to-equity ratio by estimating model (4) with the debt-to-equity ratio as a dependent variable. I define this ratio as the book value of long-term debt plus current liabilities divided by the market value of common equity plus the liquidating value of preferredequity.TheresultsforthistestarepresentedinTableD.8.Thecoefficientsreveal thatuncertaintyhasapositiveeffectonleverageinconcentratedindustriesandnoeffectin competitiveindustries. 75 TableD.1 Theeffectofcompetitionontheinvestment-uncertaintyrelationship–additionalspecificationsofmodel(4) Thistablepresentsresultsofestimatingmodel(4)usinganinstrumentalvariablesapproachwherethedependentvariableisinvestmentscaledbythepreviousyear’s capitalstock.ThesampleonlyincludesfirmsinindustriesthataresensitivetoENSOfluctuations.TheCompustatbasedmeasureoftheHerfindahl-HirschmanIndex(HHI) isusedtosplitthesampleintofirmsinhighlycompetitiveindustries(HighCompetition)andfirmsinlesscompetitiveindustries(LowCompetition).Controlsincludetotal assets,debt-to-equityratio,cashflowsscaledbycapitalstock,salesandthe12-monthSSTlevelforecast.Standarderrorsareclusteredatthefirmlevel.***,**and* indicate1%,5%,and10%significance,respectively. AllFirms HighCompetition LowCompetition (1) Returnvol. (2) 0.0330 (0.03) Tobin'sQ (3) -0.0438 -0.0439 (0.03) (0.03) 0.0198*** (4) (0.00) (5) 0.3005*** (6) 0.3267*** (0.05) (7) 0.2557*** (0.06) (0.06) -0.2268*** 0.0172*** (8) (9) -0.2842*** (0.04) -0.2360*** (0.05) (0.05) (0.00) 0.0192*** (0.00) N F-Stat.-1st Stage 51,203 51,203 51,203 24,663 24,663 24,663 26,540 26,540 26,540 483.69 421.11 325.36 254.13 257.87 170.94 196.51 199.15 144.78 Controls No Yes Yes No Yes Yes No Yes Yes YearFE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes FirmFE Yes Yes Yes Yes Yes 76 TableD.2 Theeffectofcompetitionontheinvestment-uncertaintyrelationship–withoutexcludingindustriesthat arenotsensitivetoENSO This table presents results of estimating model (4) using an instrumental variables approach where the dependent variable is investment scaled by the previous year’s capital stock. The sample includes firms in all industriesregardlessofwhethertheyaresensitivetoENSOfluctuations.TheCompustatbasedmeasureofthe Herfindahl-HirschmanIndex(HHI)isusedtosplitthesampleintofirmsinhighlycompetitiveindustries(High Competition)andfirmsinlesscompetitiveindustries(LowCompetition).Controlsincludetotalassets,debt-toequityratio,cashflowsscaledbycapitalstock,salesandthe12-monthSSTlevelforecast.Standarderrorsare clusteredatthefirmlevel.***,**and*indicate1%,5%,and10%significance,respectively. AllFirms HighCompetition LowCompetition (1) (2) (3) Returnvolatility -0.0682** F-Stat.-1stStage (0.04) 0.0169*** (0.00) N -0.1023** (0.07) 0.0197*** 0.1377** (0.03) Tobin'sQ 0.0200*** (0.00) (0.00) 85,095 32,456 52,639 176.63 101.54 109.92 Controls Yes Yes Yes YearFE Yes Yes Yes FirmFE Yes Yes Yes 77 TableD.3 Theeffectofcompetitionontheinvestment-uncertaintyrelationship–ENSOsensitivitiescalculated using4-factormodel This table presents results of estimating model (4) using an instrumental variables approach where the dependentvariableisinvestmentscaledbythepreviousyear’scapitalstock.Thesampleincludesonlyfirmsin industries which are significantly affected by changes in ENSO conditions. These sensitivities are calculated accordingtomodel(D.1).TheCompustatbasedmeasureoftheHerfindahl-HirschmanIndex(HHI)isusedtosplit thesampleintofirmsinhighlycompetitiveindustries(HighCompetition)andfirmsinlesscompetitiveindustries (LowCompetition).Controlsincludetotalassets,debt-to-equityratio,cashflowsscaledbycapitalstock,salesand the12-monthSSTlevelforecast.Standarderrorsareclusteredatthefirmlevel.***,**and*indicate1%,5%,and 10%significance,respectively. AllFirms HighCompetition LowCompetition (1) (2) (3) Returnvolatility 0.0349 N F-Stat.-1stStage -0.2634** (0.06) 0.0191*** 0.2712*** (0.04) Tobin'sQ (0.12) 0.0194*** 0.0171*** (0.00) (0.00) (0.00) 51,203 24,663 26,540 321.63 144.29 32.02 Controls Yes Yes Yes YearFE Yes Yes Yes FirmFE Yes Yes Yes 78 TableD.4 Theeffectofcompetitionontheinvestment-uncertaintyrelationship–timeinvariantENSOsensitivities calculatedusingmonthlyreturndatafrom1980to1984 This table presents results of estimating model (4) using an instrumental variables approach where the dependentvariableisinvestmentscaledbythepreviousyear’scapitalstock.Thesampleincludesonlyfirmsin industries which are significantly affected by changes in ENSO conditions. These sensitivities are calculated accordingtomodel(2)usingmonthlyfirm-levelreturndatafrom1980to1984.TheCompustatbasedmeasure oftheHerfindahl-HirschmanIndex(HHI)isusedtosplitthesampleintofirmsinhighlycompetitiveindustries (HighCompetition)andfirmsinlesscompetitiveindustries(LowCompetition).Controlsincludetotalassets,debtto-equityratio,cashflowsscaledbycapitalstock,salesandthe12-monthSSTlevelforecast.Standarderrorsare clusteredatthefirmlevel.***,**and*indicate1%,5%,and10%significance,respectively. AllFirms HighCompetition LowCompetition (1) (2) (3) Returnvolatility -2.4350* N F-Stat.-1stStage -0.7865*** (0.40) 0.0019 2.6832*** (1.35) Tobin'sQ (0.30) 0.0293*** 0.0112*** (0.01) (0.00) (0.00) 51,203 24,663 26,540 3.44 42.81 6.81 Controls Yes Yes Yes YearFE Yes Yes Yes FirmFE Yes Yes Yes 79 TableD.5 Theeffectofcompetitionontheinvestment-uncertaintyrelationship–ENSOsensitivitiescalculatedby includingcompetitionmeasureasanadditionalriskfactorinmodel(2) This table presents results of estimating model (4) using an instrumental variables approach where the dependentvariableisinvestmentscaledbythepreviousyear’scapitalstock.Thesampleincludesonlyfirmsin industrieswhicharesignificantlyaffectedbychangesinENSOconditions.Thesesensitivitiesarecalculatedby estimatingmodel(2)augmentedwithameasureofcompetitionasanadditionalriskfactor.TheCompustatbased measure of the Herfindahl-Hirschman Index (HHI) is used to split the sample into firms in highly competitive industries(HighCompetition)andfirmsinlesscompetitiveindustries(LowCompetition).Controlsincludetotal assets,debt-to-equityratio,cashflowsscaledbycapitalstock,salesandthe12-monthSSTlevelforecast.Standard errorsareclusteredatthefirmlevel.***,**and*indicate1%,5%,and10%significance,respectively. AllFirms HighCompetition LowCompetition (1) (2) (3) Returnvolatility -0.0272 F-Stat.-1stStage (0.06) 0.0169*** (0.00) N -0.1059* (0.06) 0.0191*** 0.1852*** (0.04) Tobin'sQ 0.0205*** (0.00) (0.00) 51,203 24,663 26,540 54.09 225.49 35.90 Controls Yes Yes Yes YearFE Yes Yes Yes FirmFE Yes Yes Yes 80 TableD.6 Theeffectofcompetitionontheinvestment-uncertaintyrelationship–industry-levelENSOsensitivities basedontwo-digitSICcodes This table presents results of estimating model (4) using an instrumental variables approach where the dependentvariableisinvestmentscaledbythepreviousyear’scapitalstock.Thesampleincludesonlyfirmsin industrieswhicharesignificantlyaffectedbychangesinENSOconditions.Thesesensitivitiesarecalculatedby estimatingmodel(2).Industryclassificationisbasedon2-digitSICcodes.TheCompustatbasedmeasureofthe Herfindahl-HirschmanIndex(HHI)isusedtosplitthesampleintofirmsinhighlycompetitiveindustries(High Competition)andfirmsinlesscompetitiveindustries(LowCompetition).Controlsincludetotalassets,debt-toequityratio,cashflowsscaledbycapitalstock,salesandthe12-monthSSTlevelforecast.Standarderrorsare clusteredatthefirmlevel.***,**and*indicate1%,5%,and10%significance,respectively. AllFirms HighCompetition LowCompetition (1) (2) (3) Returnvolatility -0.0540 N F-Stat.-1stStage -0.3504*** (0.07) 0.0184*** 0.1892*** (0.06) Tobin'sQ (0.11) 0.0154*** 0.0180*** (0.00) (0.00) (0.00) 51,203 24,663 26,540 35.31 134.83 17.46 Controls Yes Yes Yes YearFE Yes Yes Yes FirmFE Yes Yes Yes 81 TableD.7 Theeffectofcompetitionontheinvestment-uncertaintyrelationship–removingforecastablevariation inexpectedreturnsfromfirm-levelstockreturnvolatility This table presents results of estimating model (4) using an instrumental variables approach where the dependent variable is investment scaled by the previous year’s capital stock. Return volatility indicates the measuredescribedinAppendixSectionD.7.Thesampleincludesonlyfirmsinindustrieswhicharesignificantly affectedbychangesinENSOconditions.TheCompustatbasedmeasureoftheHerfindahl-HirschmanIndex(HHI) is used to split the sample into firms in highly competitive industries (High Competition) and firms in less competitiveindustries(LowCompetition).Controlsincludetotalassets,debt-to-equityratio,cashflowsscaledby capitalstock,salesandthe12-monthSSTlevelforecast.Standarderrorsareclusteredatthefirmlevel.***,**and *indicate1%,5%,and10%significance,respectively. AllFirms HighCompetition LowCompetition (1) (2) (3) Returnvolatility -0.0484 F-Stat.-1stStage (0.05) 0.0171*** (0.00) N -0.2368*** (0.06) 0.0197*** 0.2404*** (0.03) Tobin'sQ 0.0191*** (0.00) (0.00) 51,203 24,663 26,540 269.43 197.63 120.03 Controls Yes Yes Yes YearFE Yes Yes Yes FirmFE Yes Yes Yes 82 TableD.8 Theeffectofuncertaintyonleverage(debt-equityratio) This table presents results of estimating model (4) using an instrumental variables approach where the dependentvariableisthedebt-to-equityratio.Thesampleincludesonlyfirmsinindustrieswhicharesignificantly affectedbychangesinENSOconditions.TheCompustatbasedmeasureoftheHerfindahl-HirschmanIndex(HHI) is used to split the sample into firms in highly competitive industries (High Competition) and firms in less competitiveindustries(LowCompetition).Controlsincludetotalassets,debt-to-equityratio,cashflowsscaledby capitalstock,salesandthe12-monthSSTlevelforecast.Standarderrorsareclusteredatthefirmlevel.***,**and *indicate1%,5%,and10%significance,respectively. Returnvolatility HighCompetition LowCompetition (1) (2) (3) 0.0114 F-Stat.-1stStage AllFirms N -0.0146 0.0263** (0.01) (0.01) (0.01) 51,203 24,663 26,540 262.98 172.66 117.25 Controls Yes Yes Yes YearFE Yes Yes Yes FirmFE Yes Yes Yes 83