PolicyResearchWorkingPaper10436TheEffectsofEnergyPricesonFirmCompetitivenessEvidencefromChileJuergenAmannArtiGroverFinance,CompetitivenessandInnovationGlobalPracticeMay2023PublicDisclosureAuthorizedPublicDisclosureAuthorizedPublicDisclosureAuthorizedPublicDisclosureAuthorizedProducedbytheResearchSupportTeamAbstractThePolicyResearchWorkingPaperSeriesdisseminatesthefindingsofworkinprogresstoencouragetheexchangeofideasaboutdevelopmentissues.Anobjectiveoftheseriesistogetthefindingsoutquickly,evenifthepresentationsarelessthanfullypolished.Thepaperscarrythenamesoftheauthorsandshouldbecitedaccordingly.Thefindings,interpretations,andconclusionsexpressedinthispaperareentirelythoseoftheauthors.TheydonotnecessarilyrepresenttheviewsoftheInternationalBankforReconstructionandDevelopment/WorldBankanditsaffiliatedorganizations,orthoseoftheExecutiveDirectorsoftheWorldBankorthegovernmentstheyrepresent.PolicyResearchWorkingPaper10436ThispaperanalyzestheimpactofchangesinenergypricesonthecompetitivenessofmanufacturingfirmsinChile.UsingtheChileanAnnualNationalIndustrialSurveydata,thepaperillustratesthat,first,increasesinenergypricesgenerallydonothurtfirmcompetitiveness.Second,theimpactofenergypricesdependsonthefueltype—whileelectricitypriceincreasesarenegativelycorrelatedwithfirmoutcomes,fossilfuelpriceincreaseshaveapositiveassocia-tionwithinvestmentandfirmproductivity,aresultthatisconsistentwiththestrongversionofthePorterhypothe-sis.Third,theseeffectsareheterogeneousandvarybyfirmattributessuchassize,ownershipandlocation.Fourth,investigatingnon-linearpatternsinfirmoutcomesbasedonthelevelofenergyprices,thefindingsshowthatthepositivecorrelationbetweenfossilfuelpriceincreasesandcapitalupgradingisparticularlypronouncedwhenenergypricesareatrelativelylowlevels.ThispaperisaproductoftheFinance,CompetitivenessandInnovationGlobalPractice.ItispartofalargereffortbytheWorldBanktoprovideopenaccesstoitsresearchandmakeacontributiontodevelopmentpolicydiscussionsaroundtheworld.PolicyResearchWorkingPapersarealsopostedontheWebathttp://www.worldbank.org/prwp.Theauthorsmaybecontactedatagrover1@ifc.org.TheEffectsofEnergyPricesonFirmCompetitiveness:EvidencefromChile∗JuergenAmann(UniversityofNottingham)†ArtiGrover(InternationalFinanceCorporation,WorldBankGroup)‡JELClassification:D21;D24;L60;O14;Q41;Q55.Keywords:Energyprices;Fossilfuels;Firmproductivity;Firm-leveltechnologyadoption,Competitiveness.∗ThisresearchwascompletedundertheguidanceofMarianneFay,CountryDirectorforChile.TheauthorsaregratefultoDoerteDoemelandandYiraJ.Mascarofortheirstrategicguidance,andtoEsperanzaLasagabaster,VincentPalmade,RichardRecord,ErikvonUexkullandseveralcolleaguesfromtheMinistryofFinanceinChilefortheirusefulcomments.†email:amannj.work@gmail.com‡Correspondingauthor,email:agrover1@ifc.org1IntroductionGreengrowthrequiresgreentechnologies,thatis,productiontechniquesthatecon-omizeonexhaustibleresourcesandemitfewergreenhousegases.Yet,investmentinsuchtechnologicalchangesisinhibitedbecausecarbon,usuallyusedasshorthandforgreenhousegases(GHGs),ishighlymispriced.1Forinstance,fossilfuelssub-sidiesandthefailuretoimplementtaxesorcontrolsforinternalizingtherisksofclimatechangeresultinasubstantiallylowerusercostofcarbonthanisappropriatefromalong-termsocietalperspective.Pricingcarbontoreflectitstruecostremainschallengingduetopoliticalreasons.2Inaddition,assessingtheeffectsofenergypricevariationonfirm-leveloutcomesisdifficultduetoextensivedatarequirementsandidentificationissues.Consequently,empiricalworkanalyzingtheimpactofen-ergypricesonfirmcompetitivenessisstillrelativelyscarcerindevelopingcountriescomparedtodevelopedeconomies(Burkeetal.,2016).Thispapercontributestounderstandingtheeffectsofenergypriceincreasesonmanufacturingfirm-leveloutcomesinChile,anemergingeconomy,whichhasundergonefast-pacedeconomicdevelopmentinrecentyears.3Chileisaninterestingcaseasthecountryhasexperiencedasubstantialvariationinprimaryenergyinputs,includinghydro,coalandgas,asaprimarysourceofenergyproductioninthepast;seeFigureA.1.4Thecountryhasalsocommittedtocarbonneutralityby2050,butitsindustrycontinuestodependheavilyonfossilfuelsdespitesomeprogress.Inthecontextofitsbroadertaxreformagenda,theBoricadministrationthattookofficeinMarch2022isconsideringcorrectivetaxesandsubsidyreformstoincentivizetheuseoffossilfuel-savingtechnologies.Atthesametime,thegovernmenthasannounced1Notethatinthisstudy,wecombinevarioustypesoffossilfuelswhichmayvarynotablyintheamountofGHGemissions.2Industrialpolicieshavehistoricallysupportedtheprovisionofsubsidizedenergy,amajorin-putinmanufacturing,especiallyinmarketswherecost-competitivenessisparticularlyimportant(El-KatiriandFattouh,2017).Consequently,policymakersmayfeartheadversereactionsofin-terventionstargetingareductionofgreenhousegasemissionsthroughchangesintheprice-settingmechanism.Climatechangemitigationinemergingandemergingeconomiesmightposeachallengefromadevelopmentperspectiveifitrequiresmorecostly,low-carbonenergysources(JakobandSteckel,2014).3FollowingthehistoricWorldBankCountryandLendingGroupsclassification(lastaccess:December2022),Chilewasanuppermiddle-incomecountryatthebeginningofoursampleperiodin1995andbecameahigh-incomecountryin2012.4Between1995-2013,morethan95%ofChileanmanufacturingfirmsusedelectricity,followedbydiesel,with35%-54%usingthisfueltype.TheconsumptionofotherfuelssuchasLPGandpetrolislower;onaverage20%-30%firmsusethesefuels.Arelativelysmallershareoffirmsusesnaturalgas,coke,keroseneandpipegas.2variousmeasurestopromoteinvestment,innovationandproductivitygrowth.Well-designedpolicyinterventionthatcombinescarbonpricinginstrumentswithtargetedadjustmentsupporttoovercomeinitialobstaclestoadoptingnewtechnologiescanyieldawin-winoutcomeofhigherproductivityandloweremissions.However,thecurrentevidencebaseforpolicydesignislimited.Thispapercontributespreciselytothisgapbyassessingtheeffectsofenergypricechangesonfirmoutcomes.Tothisend,thepaperusesthepubliclyavailableversionoftheChileanAnnualNationalIndustrialSurveytostudytheeffectsofenergypricesonfirmoutcomes.Thedatacontains100,803observationson12,169plants,withanaverageof4,200uniquefirmsobservedacrosssurveyyearsfrom1995to2015.Weestimateseparateelasticitiesofelectricityandfossilfuelforvariousfirm-leveloutcomessuchaspro-ductivity,energyefficiency,employment,wages,andprofitsemployingfixed-effects(FE)panelestimations.5Wealsoestimatetheheterogeneityinresponsesbyfirmattributesandnon-linearitiesconditionalonenergypricelevels.Weaddtothero-bustnessofourresultsthroughanInstrumentalVariableapproachwhereweutilizetheregionalenergydistributioncosttoaccountforunobservedreversecausality.Thispapermakesthefollowingbroadcontributions:First,itfindsthatoverallincreasesinenergypricesgenerallydonothurtfirmcompetitiveness.Second,itcon-firmsnotableheterogeneityinhowenergypricesurgesaffectfirmperformanceoftheChileanmanufacturingsectorinlinewithotherrecentworks.Thesedependonthetypeofenergyinput,i.e.electricityvsfossilfuels,(Amannetal.,2021;Cal`ıetal.,2022,amongothers)andhintatvariousdegreesofsubstitutabilitybetweendistinctproductioninputsconditionalonfirmcharacteristics(BrucalandDechezleprˆetre,2021,amongothers).Forinstance,whileelectricitypriceincreasesnegativelycorre-latewithproductionandemployment,fossilfuelpriceincreasesarenotnecessarilydetrimentaltofirmperformance.Forexample,a10%riseinfossilfuelpricescor-relatessignificantlywithacapitalinvestmentincreaseof3.1%andanincreaseinoutputperworkerof0.8%.Inotherwords,ourresultssupportthestrongversionofthePorterHypothesis(Porter,1991;PorterandVanderLinde,1995)forfossilfuelincreasesbutnotforelectricitypriceincreases.Third,itpointsoutextensivehetero-geneitiesofenergypricehikesalongmanydimensions.Wefindheterogeneouseffectsbyfirmattributessuchassize,ownershipandlocation.Theheterogeneityanaly-sissuggeststhatthestrongversionofthePorterHypothesiscanonlybeobservedforlargefirms(with50employeesorabove).Bycomparison,smallfirmsobserveanegativecorrelationwithsurgesinenergypricesingeneralandelectricitypricesinparticular.OurresultsalsoindicatethatthePorter-typeadoptionmechanismis5Weconstructaharmonizedunitpriceindexforfossilfuelsbycombininginformationoncoal,gas,diesel,petrol,LPGandkeroseneuseofChileanmanufacturingfirms.3morepronouncedforexportingfirmsandfirmsunderforeignownership,whilefirmsthatonlyoperatedomesticallyandareunderdomesticownershiparemoreadverselyaffectedbyhikesinelectricityprices.Fourth,itpresentsexplanatoryevidencethatissuggestiveofanotablechangeintherelationshipbetweencompetitivenessindi-catorsandfossilfuelpricechangesconditionalonthepricelevel.Wefindthatthepositivecorrelationbetweenfossilfuelpriceincreasesandcapitalupgradingismorepronouncedwhenfuelpricesarelow.Athigherfuelpricelevels,thiseffectbecomessmallerandeventuallyinsignificant.Similarnon-lineareffectscannotbeobservedforelectricityprices.Thestructureofthispaperisasfollows:section2,providesanoverviewoftheempiricalfirm-levelliteratureontheenergyprices-productivitynexuswhilesection3setsabackgroundoftheinstitutionalchangespertainingtoenergypricesforChileanmanufacturingfirms.WethenintroducetheEstructuradelaindustriamanufactur-eradata(INE,2015)insection4,afterwhichweturntoourempiricalstrategyandtheresultsinsections5and6,respectively,beforeconcludinginsection7.2LiteratureTheliteratureidentifiesfourmaintransmissionchannelsforhowfirmsmayrespondtoenergypriceinterventions(Rentschleretal.,2017;Costeetal.,2018):Onestrat-egy,wherefirmsinnovatebyadoptingmoreefficienttechnologiesandreapeconomicbenefit,isoftenreferredtoasthePorterHypothesis.6Additionally,firmsmaypursueothercopingstrategies,suchascostpass-ontocustomersandotherfirms,absorptionoradjustingproductioninputs,i.e.,substitution.Evaluatingthesetransmissionchannels,theempiricalfirm-levelliteraturefindsbroadsupportforthePorter-typeinnovationhypothesis,wherepolicy-ledpricein-terventionsleadtoinnovationandnotablebusinessupgrading.BrucalandDeche-zleprˆetre(2021)showthatlargeandenergy-intensivesectorsinIndonesianmanu-facturingtendtoreduceenergyconsumptionmostsignificantlyinresponsetoanenergypricehike.Plantsreacttohigherenergypricesbyupdatingtheircapitalstockandinvestinginnewandpresumablymoreenergy-efficienttechnology.Usingfirm-leveldataofsmallandmicroenterprisesfromIndonesiafor2013,RentschlerandKornejew(2018)exploretheimpactofinputpricechangesofelectricityand6TheliteraturedistinguishesbetweendifferentformsofthePorterHypothesis(Ambecetal.,2013):Underits”weak”form,environmentalregulationsspurinnovations,whileunderthe”strong”version,environmentalregulationswillleadtoincreasesinfirmcompetitiveness.The”narrow”in-terpretationstatesthatmoreflexibleregulationsmaybebetteratprovidingthepreferableincentivestructurestofirmsthanmoreprescriptiveforms.4variousfossilfuelsonfirmperformance.Theauthorsfindthathigherpricesforallenergytypesareassociatedwithhigherenergyefficiency.Cal`ıetal.(2023)evaluateenergyprices’directandindirectimpactonfirms’economicperformanceforelevendevelopingcountriesbetween2002and2013.UsingWorldBankfirm-levelsurveydata,theauthorsfindthathigherenergypricesdonotnecessarilyhampereconomicperformance.Theyalsoreportconsiderableheterogeneities,asenergy-intensivefirmsreportasmallerperformanceeffectinresponsetoariseinenergypricesthantheirlessenergy-intensivecounterparts.Cal`ıetal.(2022)focusonmedium-tolarge-sizedmanufacturingfirmsintwoemergingeconomies(MexicoandIndonesia).Theyfindthatwhilesurgesinelectricitypriceharmplants’performance,increasesinfuelpricesyieldpositiveoutcomesforlaborandtotalfactorproductivityandprofits.Theeffectsareparticularlypronouncedforcapital-intensivemanufacturingsectors.Inthesamevein,Amannetal.(2021)analyzetheeffectsofenergypricehikesonmanufacturingfirmsinOman.Theyshowthatincreasesinfossilfuelpricesarecausallylinkedtoimprovementsinproductivityandefficiencyandleadtonotablebusinessupgrading.AsCal`ıetal.(2022),theOmanstudyonlyobservessuchinnovationeffectsforfossilfuelpriceincreasesbutnotforelectricityhikes,whicharemoredetrimentaltofirmperformance.Theextenttowhichenergyinputscanserveassubstitutesdependsonthecountryandpolicycontext.RentschlerandKornejew(2018)findthatwhileelectricitycanbesubstitutedawaybyablendofotherenergysources,itplaysaminorroleinreplacingfossil-basedenergysources.Amannetal.(2021)evaluatethesubstitutabilityofelectricityandaggregatedfossilfuelsandfindthatOmanifirmssignificantlyincreasetheirkWhconsumptionofelectricityinresponsetorisingfuelprices.Firm-levelevidencesuggeststhattheeffectivenessoftheabsorptionchannelsdependsonfirmcharacteristicsandsectoralattributes.Particularlyfirmsinenergy-intensivesectorsrespondmorestronglytopricehikes.RentschlerandKornejew(2018)highlightthatfirm-levelresponsepatternsofIndonesianmicro-firmsareverydifferentdependingontheanalyzedsector.BrucalandDechezleprˆetre(2021)reportthatforIndonesianmanufacturingfirms,energypriceincreasesarecausallylinkedtoupticksinplantexitandemploymentcontraction,particularlyforenergy-intensiveandlargefirms.Theauthorsalsoreportjobreallocationfromenergy-intensivetoenergy-efficientfirmsandsectors.KumarandPrabhakar(2020)analyzehowIndianenergy-intensiveexportsreacttochangesinenergypricesandfindadversetradeeffects,assertingthatcarbonleakageconcernsarenotnecessarilyunfounded.Fur-thermore,energy-intensivesectorsareaffectedmoreseverelybyenergypricehikes,mostnotablythenon-ferrousmetalandmachineryindustry.Inturn,Cal`ıetal.(2023)donotfindstatisticalevidencethatpricehikesleadtoemploymentlosses5acrosselevendevelopingeconomiesbetween2002and2013.Numerousstudiesidentifynotablecostpass-throughfromfirmstotheconsumerindicatingfirmsonlybearpartofthecarboncosts(LinandJiang,2011;Arlinghaus,2015;JoltreauandSommerfeld,2019).However,firm-levelevidenceonthistrans-missionchannelremainsrelativelyscarceforemergingeconomies.RentschlerandKornejew(2018)confirmthathigherpricesforallenergytypesareassociatedwithhigherlong-runsalesinIndonesia.Cal`ıetal.(2023)doonlyfindsignificantevidenceforanaggregatedcostpass-througheffectforfirmsexperiencingpoweroutages.3InstitutionalSettingIn1998,Chilesufferedaseveredrought,whichhadanotableimpactontheenergygenerationoftheCentralInterconnectedSystem(SIC)network.7Thisevent,inconjunctionwithdisruptionsinthreenaturalgasplants,causedblackoutsbetweenNovember1998andApril1999,forcingthegovernmenttorationsupply(Serra,2022).Thesubsequentperiodofthelate-1990stomid-2000swascharacterizedbyheavyuseofnaturalgasinelectricitygenerationandmassivenaturalgasimportsfromArgentina(Cansinoetal.,2018).However,thegas-heavyenergyproductionepisodeabruptlystoppedin2005whenArgentinabeganrestrictingandtaxingnat-uralgasexportsduetodomesticdeficitsleadingtoanotablepricesurgeinChile.Inresponse,Chilesawtheconstructionofnewcoalplantstoclosethewedgebetweenreducedenergysupplyandgrowingdemand.Coalplantsplayedacrucialroleinthisendeavorgiventheircomparablylowercostofelectricitygeneration(Serra,2022).Thecountryalsodiversifieditsenergymixintermsofenergytypeanddestination.Forexample,in2009/2010,ChileimportednaturalgasfromAlgeria,EquatorialGuinea,theArabRepublicofEgyptandIndonesiaandconstructedlarge-scaleliquefiednaturalgasterminals(Mundacaetal.,2015),leadingtoanoverallreductionintherelativeweightofhydropower(Cansinoetal.,2018).Technologicaladvancementsandgrowingenvironmentalconcernshaveledtoastrongerfocusongreenandrenewableenergyinrecentyears.Bythe2010s,increas-ingpublicoppositionandstricterenvironmentalrequirementshaltedtheconstruc-tionofadditionalcoal-firedplants(Serra,2022),andnewpolicymeasurespromotingnon-conventionalrenewableenergieswereputintopractice(MEFR,2008;Cansino7Throughouttheanalyzedperiod,Chilehadfourmainelectricitygrids:theNorthernInter-connectedSystem(SING),theCentralInterconnectedSystem(SIC),theAysenSystem(SEA),andtheMagallanesSystem(SEM),withtheformerbeingintegratedintotheNationalElectricitySystem(SEN).Withinthesegrids,numerousstate-ownedandprivateenergyproducershavebeeninchargeofenergygeneration(Serra,2022).6etal.,2018).Theyear2014alsosawthepassingofcarbontaxlegislationwiththeGeneralTaxReformBill.84DataTheChileanAnnualNationalIndustrialSurvey(ENIAfromnowon)ismadeavail-ablebytheNationalStatisticalInstitute(INE,2015).ENIAdataispubliclyavailableandspanstheperiod1995-2019.From1995to2007,ENIAoffersaharmonizedpaneldataset(referredtoasthecombineddata)whichidentifiesthesamemanufacturingfirmwithauniqueIDlinkingplantsovertime.Atthetimeofthisstudy,uniquefirmIDslinkingpaneldatawereunavailableand,untilrecently,ENIAhadmicrodataavailablefrom2016-2019inseparateannualwaves(withoutauniquefirmIDovertime).WerefertotheseseparatewavesastherawversiononENIA.9ThecombinedENIAdatasetsuffersfromalackofregionalandmanufacturingsector-levelidentifiers.TorecoverinformationonthelocationandtheISIC4-digitcodesforeachplant,weresorttotherawdataversionofENIA,whichcontainsbothdesiredvariables.Wematchplantsinthecombinedandrawdatabytheircommoncharacteristicsandapplythesamematchingtechniquestoexpandthepanelstructureinthepost-2007period.Thisprocedureyieldsafinaldatasetcontaining100,803observationson12,169plants,withanaverageof4,200uniquefirmsobservedacrosssurveyyears.WeprovideamoreextensivediscussiononthematchingalgorithminAppendixC.AsillustratedinFigureA.2,thematchingalgorithmachievesastablefirmsampleovertheentiresamplehorizonandalowattritionrate.Thecombineddataalsoexhibitirregularitiespotentiallystemmingfrommis-reportedelectricityandfossilfuelprofiles.Weaddressthisissuebyemployingasimplecleaningalgorithm,whichprovidesarule-basedproceduretoidentifyandcorrectinter-temporalchangesinreportedquantitiesandunitpricesofpre-definedmagnitudes.WeprovideamoreextensivediscussiononthecleaningalgorithminAppendixD.8Withitsinceptionin2014,thetaxreform(Ley20.780)fallstooclosetotheendofthesampleperiodtomeritanempiricalanalysis.Atthetimeofwriting,theimplementationofthepolicyremainsanongoingprocess;seetheWorldBank’sResultsBrief15February2021andprojectMarketInstrumentsforClimateChangeMitigationinChile(lastaccess:September2022).9Theauthorsareworkingonarevisionoftheresultsusingthenewlyavailabledata.75EmpiricalMethodology5.1BaselineEstimationModelStructure.OurempiricalsetupissimilartothatofotherworksintheliteraturesuchasAmannetal.(2021);Cal`ıetal.(2022).Weestimate:yit=βixit+εit,(1)wherexit=upELit,upFFit′denotetheunitpricesforelectricity(EL)andfossilfuels(FF)forfirmiandperiodt,respectively.Thecalculationoftheelectricityunitpricesisstraightforward:upmit:=valuemitquantitymit,(2)wherevaluedenotesthepurchasedvalue(inrealLCU)andquantitydenotesthepur-chasedquantityofenergytypem={EL}.10Forthefossilfuelaggregate,wecalcu-latekWh-equivalentquantitiesforalln={coal,gas,diesel,petrol,LPG,kerosene}fossilfueltypesaccordingtotheconversionratesinTableB.1asfollows:upFFit=nwnit×upnit,wnit=˜qnitn˜qnit,(3)where˜qnitdenotesthequantity(inkWhequivalents)offossilfueltypenforfirmiandperiodt,andwnitcorrespondstoitsfixed-quantityweight,respectively.ThecalculationoftheindividualupnitseriesisidenticaltoEquation2form=n.ThetwoenergyunitpriceserieswederiveinthiswayarevisualizedinFigureA.3.DependentVariables.Weanalyzetheeffectofenergypricevariationsonfirm-leveloutcomes,yit.Tothisend,wecompartmentalizetheoutcomevariablesbyapplyingtheframeworkintroducedinsection2totheextenttowhichthisispos-siblewiththeopen-accessversionofENIA.11Wedefinethesedomainsas:invest-10AllmonetaryseriesaredeflatedusingISIC2-digitindustry-levelfollowingHaraguchiandAmann(2023).11AmoredetailedanalysisofadditionalvariablesofinterestsuchasTFPQorothercopingmechanisms,particularlythecostpass-throughdimension,wasnotpossibleduetocurrentdata8mentandproductivity/survival,substitutionandabsorption.Onthecompetitive-ness/innovationeffects,weevaluatethestrongandweakformofthePorterHy-pothesisbycomputingvariousmeasuresofproductivity,includinglaborproductiv-ityandTotalFactorProductivity(TFP)usingtheAckerberg-Caves-Frazer(ACF)method(Ackerbergetal.,2015).Substitutioneffectismeasuredbytheelasticityofsubstitutionacrossvariousenergysourcesduetochangesintheirprices.Absorptioneffectismeasuredthroughtheimpactofenergypricesonfirmsize(employment),otherinputcosts(e.g.,wages),profitmarginsandreturnonsales,normalizedbynetincome,costsandthechangesoffixedassets.Weprovideacompletelistofanalyzedfirm-leveloutcomevariablesyitalongsidetheirdefinitionanddescriptivestatisticsinTableB.2aswellasalistofdependentvariablesanalyzedbyotherpapersinthisliteratureinTableB.3.5.2HeterogeneityinResponsesToanalyzeheterogeneityintheresponsepatterns,weadjustEquation1employingadummyidentifieroftheform:yit=(β′ixit)×I(zit∈Z)+εit,(4)whereI(·)isanindicatorfunctionwhichtakesthevalueoneifconditionZismetforaparticularobservationinourdatazit.Weanalyzeheterogeneitiesinfirm-levelresponsesbybreakingdownthesamplebasedonthefirm-levelcharacteristicsoffirmsize(small,medium,large;seepanelIinTableB.4forasummarystatisticsanddefinitionoftherespectivecut-offs),ownership(foreignanddomestic;seepanelII),exportstatus(exporterandnon-exporter;panelIII)andenergyintensity(intensiveandnon-intensive;panelIV).5.3Non-linearityinResponsesWeinvestigatenon-linearpatternsbyextendingthemodelinEquation1andesti-matingapanelquantileregressionmodelasproposedbyMachadoandSilva(2019)toinvestigatehowthecorrelationbetweenfirm-leveloutcomesandenergypricechangesvariesdependingontheactualobservedpricelevel.Theimplicationisthatafirm’slimitations.Thisleavesnumeroushighlypolicy-relatedquestionsunanswered.Forexample,oneofthemostdirectmeasuresofcostpass-throughistheunitpriceoffinalproductssoldtothemarket;however,thisisnotobservedintheopen-accessversionofENIA.Amoreextensiveanalysiscouldbeconductedinthepresenceofaccesstothefulldata.9optimalresponsetoenergypricehikesmayvarydependingontheenergypriceitobserves.Forexample,firmsmaybemorewastefulintheirenergyuseatlowerenergypricelevelsandretainlessefficientproductionprocessesbecauseoflowop-portunitycosts.Conversely,theymightfinditmoredifficulttoadjusttheirpro-ductionathigherpricelevelsastheymayhavealreadyexhaustedsomeofthemorestraightforwardenergy-savingstrategieswhentransitioningawayfromaninitiallylowenergy-priceregime.Thelinearspecificationfortheconditionalquantileofde-pendentvariableyandexplanatoryvariablesxitisgivenby:Q(τxit,β(τ))=β(τ)xit,(5)whereβ(τ)isavectorofthecoefficientsrelatedtotheτthquantile.5.4EndogeneityIssuesEndogeneityinestimatingtheimpactofenergypricesonfirmoutcomescanemergeduetoomittedvariablesbias,selectioneffectsandreversecausality.Weaddressthevarioussourcesasmuchaspossibleinthecurrentsetup.First,omittedvariablebiascanbeduetotime-invariantvariables(e.g.,accesstospecificenergysourcesduetoregionalorindustry-specificattributesorpoliciescontingentonsuchcriteria).Itcanalsovaryovertimeandbeeitheridiosyncratic(e.g.,managerialabilitytoadjustenergyusebasedonpricesoptimally)ormoresystemic(e.g.,time-trendssuchasstructuralchangesduetoglobalmarketshocksortechnologicalchange).Issuesofreversecausalityariseinourcontextbecauseenergypricesaremeasuredatthefirmlevel.Firmswithhigherproductivityorotheroutcomesarelikelytheoneswithbettercapabilities(e.g.,management).Inturn,thesecanaffectenergypricesatthefirmlevelthroughseveralchannels.Forexample,byeasinginformationandcoordinationfrictions,managementmayhelpreduce“hidden”administrativeandtimecostsassociatedwithenergyusethroughefficiency-enhancinginvestmentsinprogramsorequipmentthatcanaffectfirm-levelenergypricesperunit(GillinghamandPalmer,2014).Managementmayalsohelpfirmsovercomeuncertainty,forinstance,inenergypricefluctuationsorthemagni-tudeoffuturecostsavings,byenablingfirmstotargetandtrackhistoricalenergyuseandaccuratelyforecastfutureneedsandpayoffs.Inparticular,savvyfirmsmaybeabletooptimallytimeenergypurchaseswhenpricesarelow,e.g.,byemployingforecastingtoolsorusingfinancialvehicles.12Productivefirmsalsoimplementmore12Evendifficult-to-storeenergycarrierssuchaselectricityareoftensubjecttodifferencesinfixed10energy-centricmanagementpractices(GroverandKarplus,2020),whichmayraiseenergyefficiencyandtherebyreducetheper-unitprice.Fixed-effectsSpecification.First,weuseafixed-effects(FE)setuptocaptureendogeneityarisingfromvariationsintime-invariantfirmattributes,regionalandindustry-specificmarketconditionsandtechnology-relateddevelopments(Mundlak,1978;Wooldridge,2005).Tothisend,theerrorterminEquation1hasthefollowingstructure:εit=αi+Dst+τrt+eit,(6)whereαidenotesfirm-specificinterceptswhileindustry-yearinteractions(Dst)cap-tureindustry-specificmarketconditionsandtechnology-relateddevelopments.Wealsoincluderegion-yearinteractions(τr)tocapturetheevolutionofregion-specificlong-rundevelopmentdynamicsinlocalgovernance,infrastructure,development,innovation,etc.Second,weruleoutendogeneityresultingfromautocorrelationorheteroscedastic-itybyusingclusteredstandarderrorsattheplant-yearlevelasresidualsinEquation6(Petersen,2009),whichcanbeestimatedconsistentlywithasufficientlylargenumberofclusters(Wooldridge,2010;Cameronetal.,2011).WeemploythisFEspecificationforallbaselineestimations,includingtheheterogeneityandnon-linearityanalysis.InstrumentalVariables.Weaddresstime-varyingunobservedvariationattheplantlevelremainsasourceofendogeneitybyemployinganInstrumentVariable(IV)designcommonlyusedinthisfield(e.g.,Amannetal.,2021;Cal`ıetal.,2022).ThefirststageIVestimatingplant-levelenergypricesis:upmit=α0+α1instmit+Dst+τrt+ηit,(7)whereinstmitreferstotheinstrumentofenergyunitpriceupitofenergysourcem.Byemployingaspatial/leave-one-outinstrument,weintendtoexploitthegeographicalvariationinenergypricesresultingfromthecostofdistributingenergytoaparticularprovince.Forthevalidityofourinstrument,werequireittobecorrelatedwithplant-levelenergypriceswithoutdirectlyaffectingperformance.However,thevalidityofspatialinstrumentshasbeencriticizedforvariousreasons(Betzetal.,2018).costsbasedonuseandtime-of-daypricing,includingindevelopingcountries,creatingopportunitiesforarbitragebyshiftingproductiontooff-peakhours.11Spatialinstrumentsareeffectiveifunitpricesofotherplantsdonotdirectlyorindirectlyaffectplanti’soutcomevariable.Asenergypricesdrivefirmperformance,theycanindirectlyfeedintotheoutcomesofotherplantsthroughspillovereffects.Tominimizethiseffect,ourspatialinstrumentconsidersenergypricesfori′plantswithinaregionthatarenotoperatinginthesameISIC2-digitsectorasplantiinIVspecification1(IV1).Atthesametime,energypricesoffirmswithinthesamesectorcanbealignedduetopoliciespertainingtosubsidiesinagivensector.IfweconsiderpricesoffirmsinotherISIC2-digitsectorsasinstruments,theymaynotbestronglycorrelatedwithenergypricesofafirminanothersector,whilespillovereffectsmaybemostprominentinindustriesthataremostcloselyrelatedintermsofactivityandlocation.Consequently,andasarobustnesscheck,weprovideanalternativeIVspecification(IV2),whichconsiderstheenergypricesfori′plantswithinaregionthatarenotoperatinginthesameISIC4-digitsectorasplantibutcanoperateinthesamewiderISIC2-digitsector.Werecoverthecostofregionalenergydistributionasfollows:Instepone,forfirmiunitpriceindexforfueltypem,upmit,wecalculatetheaverageunitpricespaidbyallotherfirmi′inthesameregionrinagivenyear.Theinstrumentedenergyunitpricesarecalculatedas:IV1:¯upm,2it=Nrtsi∈r,i̸=i′,s2i̸=s2i′upmit(Nrt−1),(8)IV2:¯upm,4it=Nrtsi∈r,i̸=i′,s4i̸=s4i′upmit(Nrt−1),(9)whereNrtsdenotesthenumberofplantsinperiodtpopulatingthesameregionasfirmi,i∈rthatarenotengagedinthesamemanufacturingactivity,i.e.,thesameISICk-digitsectorsk,ski̸=ski′,k={2,4}.Insteptwo,wenormalizethisunitpriceinequations8and9bythenationalaverageofthesameyeartominimizethepotentialcorrelationofplant-leveloutcomeswithaggregateshocksaffectingeachenergytype:instm,kit=¯upm,kit¯upm,kt,¯upm,kt=Nkiupm,kitNk,(10)12where¯upm,ktistheaverage,nationalunitpriceofenergysourceminyeartandinstrumentoftypek.Thisinstrumentcanbeinterpretedasaproxyforthecostofenergydistributiontoaparticularregion:itiscorrelatedwithplant-levelenergypriceswhilemitigatingtheeffectofidiosyncraticplant-levelattributesbyexcludingtheplantitselfwhencalculatingregionalaveragesinequations8and9.Someconcernsaboutthevalidityofspatialinstrumentsmayremain.First,itispossiblethattheenergyunitpriceoffirmican,atleastpartially,beinfluencedbytheunitpriceofotherplants.GiventhemarketstructureoftheChileanenergysector,wearguethatplantstypicallyenterthemarketaspricetakers.Consequently,theycannotnegotiatelowerpricesbasedonotherplants’energybills,eveniftheyknowtheenergybilloftheircompetitors.Yet,notallofouranalyzedoutcomevariablesaresubjecttothesameendogeneityconcerns.Forexample,ifplantsi′becomemoreefficientandconsumelesscoal,itisunlikelythatthiswillaffectfirmi’scoalconsumption.Second,spatialinstrumentsmaybemoreofaconcernincaseswherethesubstitutabilityofinputsishighlydependentonlocalconditionsinthelabormarket.Forinstance,thehiringpracticesofplantsi′mayinfluencethebargainingpowerofjobseekersand,therefore,affectthewagebillofotherplants.Whilewecannotruleoutallsuchdynamics,theconsistencyinourfindingsacrossmodelspecifications(includingbothIVdesigns)mayindicatethatourempiricalsetupaccountsformostoftheseeffects.Finally,asanadditionalcheckforourinstrumentalvariabledesign,wefollowtherecommendationsinChaoandSwanson(2005)andprovideteststatisticsaspartoftheregressiontablestoillustratethatnoneofourmodelssuffersfromweakidentification.6Results6.1BaselineFEEstimatesTable1presentresultsfrombaselineestimation,whichwediscussalongwiththeidentifiedfirm-levelcopingmechanism.Investment&productivity/survival.Inlinewithglobalevidence,ourFEestima-tionssuggestapositivecorrelationbetweenproductivityandinvestmentwithfossilfuelprices(panelI)andasmalltrade-offbetweenproductivityandenergyprices(panelII).Therearesomenoteworthydifferencesinfirm-levelresponsebyfueltypes.Higherelectricitypricesdonotsignificantlycorrelatewithareductionininvestmentinassets(panelI)andlowerproductivity(panelII).However,increasesinfossilfuelpricescorrelatepositivelywithinvestmentsinmachinery(column1),thebalanceofnetassets(4),outputperworker(5)andhigherwages(13),andnegativelywithexit13probability(8).Overall,theresultsonfossilfuelpricehikespertainingtoinvest-mentssupportaweakversionofthePorterHypothesis,whilethoseonproductivityareinfavorofthestrongversion.Theseresultsalignwithotherstudiesdifferentiat-ingfirm-levelcopingstrategiesbyenergytypeandunderscoretheimportanceofadisaggregatedanalysisbyfueltype.Substitutioneffect.PanelIIIinTable1suggeststhatincreasesinpricesofelec-tricitycorrespondtoasignificantreductioninconsumedquantitiesofelectricity(9).Intermsofmagnitude,thequantifiedeffectsforfossilfuelarecomparableinmagni-tudetootherstudieswithasimilarestimationsetup(Amannetal.,2021).However,theresultshintatmorepronouncedpriceelasticitiesofbothenergytypes.Wealsoobservesimilarresponsesacrossallenergytypesinresponsetoasurgeinfossilfuelhikes,whicharesignificant.Firms’consumptionoffossilfuelscorrelatespositivelywithapricehikeinelectricity(9),butthedegreeofsubstitutionbetweenfossilfuelsandelectricityvariesbyfueltype(11vs12).Absorptioneffect.PanelIVofTable1suggeststhatfossilfuelincreasespositivelycorrelatewithwages(13)butnotemployment(15&16).Wageeffectsemanatepri-marilyfromthenon-productionworkersub-sample,suggestingsomereallocationofworkerstowardshigherskillusageorpremiumduetochangesinfossilfuelprices(BrucalandDechezleprˆetre,2021;Dechezleprˆetreetal.,2020).Bycomparison,elec-tricitypricehikesseemtoforcefirmstoshedemployment(15)and,moreso,produc-tionworkers(16).PanelVofTable1confirmsthatenergypriceincreasesnegativelycorrelatewithfirms’profitmargins.However,absorptionandpass-throughagainvarybyenergytype.Whilefossilfuelpricehikesareassociatedwithcostincreases,electricitysurgesaffectprofitmarginsthroughareductioninsales(perhapsthroughareductioninfirms’productionworkers).Theresultsforreturnonsalesarecon-sistentwiththefindingsonsalescostsandassetevolution,showingaconsistentnegativecorrelationwithincreasesinenergyprices.Toconclude,whileelectricitypriceincreasesareassociatedwithreducedpro-ductionandadeclineinfirmsize,fossilfuelpricehikeshaveapositiveassociationwithcapitalinvestment,possiblymanifestinginimprovedproductivity,therebysup-portingthestrongversionofthePorterHypothesis.However,theseproductivityincreasesmaynottranslateintohigherprofitability(atleastintheshort-run),in-dicatingthatthepass-throughofsuchpricehikesislimitedandatleastpartiallyabsorbedbyfirms.1313Evidencesuggestsfirmspassatleastpartsoftheenergycostsontoconsumers(Chateauetal.,2018),whichmaynotbedesirablefromthepointofthepolicydesignandmayhavenotablewelfareimplications(Maruejolsetal.,2022).Giventhecurrentdatalimitations,itisimpossibletoquantifythiseffectinthisstudy.14Table1:Baselineresults-FEPanelIMachinery/outputICT/outputVehicles/outputAssetbalance/outputInvestments(1)(2)(3)(4)Electricity-0.009-0.0530.027-0.024(0.026)(0.132)(0.033)(0.023)FossilFuel0.031∗∗0.102∗0.042∗0.030∗∗∗(0.015)(0.048)(0.021)(0.008)RMSE1.22860.958370.983440.71812R20.472070.690990.696400.72267AdjustedR20.343710.346590.559300.67233F-test,p-value1.00001.00000.999140.99853Observations39,9222,14318,71963,920#firms9,0971,1536,60211,609PanelIIOutput/workerValue-added/workerTFPExitProductivity/survival(5)(6)(7)(8)Electricity-0.017∗-0.018∗-0.004∗∗0.010∗∗∗(0.009)(0.009)(0.002)(0.004)FossilFuel0.008∗∗0.0040.001-0.003∗∗∗(0.003)(0.005)(0.0007)(0.0009)RMSE0.363760.565560.032510.22088R20.884660.747160.956060.305195AdjustedR20.864770.702210.938080.18521F-test,p-value0.292760.992030.990890.98324Observations67,98265,50213,47053,887#firms12,22812,2324,3639,207PanelIIIQnt.ElectricityQnt.FossilFuelQnt.DieselQnt.LPGSubstitution(9)(10)(11)(12)Electricity-0.626∗∗∗0.0040.0250.082∗(0.181)(0.034)(0.039)(0.043)FossilFuel0.173∗∗∗-0.584∗∗∗-0.512∗∗∗-0.501∗∗∗(0.014)(0.044)(0.043)(0.073)RMSE0.852001.00770.957250.85209R20.858200.763430.768350.75485AdjustedR20.833740.718680.718570.69437F-test,p-value0.541800.979900.969350.98297Observations68,24257,56142,92531,197#firms12,28412,2857,5386,074PanelIVWagesallworkersWagesproductionworkersEmp.allworkersEmp.productionworkersAbsorption-workers(13)(14)(15)(16)Electricity0.002-0.014-0.026∗∗∗-0.035∗∗∗(0.006)(0.020)(0.008)(0.011)FossilFuel0.005∗∗0.00080.0040.005(0.002)(0.006)(0.004)(0.005)RMSE0.236440.568240.335270.39471R20.867340.753670.911870.89357AdjustedR20.844480.702880.896670.87515F-test,p-value0.462090.998050.086280.21396Observations67,96745,78268,00467,575#firms12,2078,95512,28512,285PanelVProfitmarginSales/outputCosts/outputReturnonSalesAbsorption-businessmetrics(17)(18)(19)(20)Electricity-0.016∗-0.009∗∗0.003-0.043∗(0.008)(0.003)(0.005)(0.023)FossilFuel-0.010∗∗0.0010.005∗-0.018∗(0.004)(0.002)(0.002)(0.010)RMSE0.483430.230650.241690.78155R20.489550.607310.549860.82711AdjustedR20.397860.538550.472230.77826F-test,p-value1.00001.00001.00000.79504Observations64,95964,74268,22835,097#firms12,27711,72812,27712,255Fixed-effectsPlantYesYesYesYesIndustry-YearYesYesYesYesRegion-YearYesYesYesYestNote:Estimatesaccordingtofixed-effectsmodeldescribedinequations1and6includefirm-effectsandindustry-yearandregion-yeareffects.Clustered(plant&year)standard-errorsinparentheses.Signif.Codes::0.01,:0.05,:0.1156.2HeterogeneityinResponsesThissectionhighlightsthesignificantheterogeneityinresponsetoenergypricehikesacrossvariousfirmattributesbyestimatingEquation4.FirmSize.Consideringthreecategoriesoffirm-sizes:large(morethan50em-ployees),medium(between20and49)andsmall(fewerthan20),14wedrawthefollowingconclusionsontheheterogeneityinresponsesbasedontheresultspre-sentedinTable2:First,theoverallpositivecorrelationsoffossilfuelpricechangeswithfirmupgradingandproductivityareprimarilydrivenbylargefirms(panelI).Inresponsetothewithdrawaloffossilfuelsubsidies,largefirmsupgradebothma-chineryandICTequipment,consistentwiththeresultsfromOman,MexicoandIndonesia(Amannetal.,2021;Cal`ıetal.,2022).Second,theaggregatenegativeproductivityeffectsobservedforelectricitypricehikesareprimarilydrivenbysmallandmedium-sizedfirms(panelII).Largefirmsseemtobeunperturbedbyelectricitypriceincreases,bothintermsoftheirproductivityandprofitability.However,largefirmsalsoobservenegativecorrelationsoffossilfuelpricehikeswithprofitmargins.Third,firmsofallsizesaresensitivetoenergypricesandsubstituteawayfromtheenergysourcewhosepriceincreasesandtowardsthealternativesource.Intra-fuelelasticitiesdonotnotablydifferbyfirmsize(panelIII).ForeignOwnership.Wedifferentiatebetweennationally-owned(domestic)andforeignfirms.15ResultsfromTable3suggestthatalthoughinvestmentcorrelatespositivelyandsignificantlywithfossilfuelfuelhikesforbothfirmtypes,themag-nitudeisgenerallyhigherforforeignfirms(panelI).Itisnotsurprising,then,thatfuelpriceincreasescorrespondmorestronglytoproductivityincreasesforforeign-ownedfirmsthandomesticones(panelII).Furthermore,electricitypricehikesarenegativelyandsignificantlycorrelatedwithdomesticfirms’productivityandprofitmargins.Regardingsubstitutionandabsorptioneffects,foreignfirms’substitutemorestronglyinresponsetoincreasesinfossilfuelprices(panelIII)whiletheysharplyreduceemploymentfollowingelectricitypricehikes(panelIV).14Weprovidesummarystatisticsforthesizecut-offsusedintheanalysisinTableB.4.InENIAtheclassificationbyfirmsizevariesbetweentherawandcombineddatasetsandcomplicatesthefirmmatchingalgorithmdescribedinAppendixCslightly;seeTableC.7formoreinformationontherespectivecut-offsinthetwoENIAdatasets.15Wedefineafirmasdomesticifitrecordsaforeigncapitalformationofzeroinallperiods;seeTableB.4panelIIforthecorrespondingsummarystatistics.16Table2:Resultsbyfirmsize-FEPanelIMachinery/outputICT/outputVehicles/outputAssetbalance/outputInvestments(1)(2)(3)(4)Electricity×Large0.003-0.0680.002-0.046(0.035)(0.139)(0.040)(0.043)Electricity×Medium-0.065-0.0500.045-0.027(0.041)(0.286)(0.068)(0.025)Electricity×Small0.059-1.170.0470.004(0.062)(0.723)(0.070)(0.022)FossilFuel×Large0.042∗∗∗0.096∗0.0390.040∗∗∗(0.014)(0.047)(0.025)(0.011)FossilFuel×Medium0.0070.2110.0510.032∗∗(0.030)(0.154)(0.035)(0.013)FossilFuel×Small0.015-0.3180.059-0.005(0.038)(0.513)(0.050)(0.017)RMSE1.22800.955300.981430.71779R20.472460.693010.697710.72292AdjustedR20.344050.347960.561020.67259F-test,p-value1.00001.00000.999480.99920Observations39,9012,14218,70963,920#firms9,0971,1536,60211,609PanelIIOutput/workerValue-added/workerTFPProfitmarginProductivity/survival(5)(6)(7)(8)Electricity×Large-0.0130.010-0.0020.013(0.013)(0.016)(0.002)(0.014)Electricity×Medium-0.032∗∗-0.036∗∗-0.003-0.020(0.013)(0.014)(0.003)(0.012)Electricity×Small-0.030∗∗-0.054∗∗-0.008-0.053∗∗∗(0.012)(0.020)(0.005)(0.016)FossilFuel×Large0.016∗∗0.0070.01∗-0.024∗∗∗(0.006)(0.008)(0.007)(0.006)FossilFuel×Medium0.012∗∗0.0070.002-0.007(0.005)(0.007)(0.002)0.006FossilFuel×Small-0.0040.004-0.0030.015∗(0.007)(0.010)(0.002)0.008)RMSE0.358740.563110.032210.48318R20.887830.749420.956860.49022AdjustedR20.868470.704820.939170.39854F-test,p-value0.300550.994570.001101.0000Observations67,94065,46213,47064,920#firms12,22812,2324,36312,277PanelIIIQnt.ElectricityQnt.FossilFuelsQnt.DieselQnt.LPGSubstitution(9)(10)(11)(12)Electricity×Large-0.652∗∗∗-0.077∗-0.103∗0.062(0.210)(0.043)(0.053)(0.048)Electricity×Medium-0.579∗∗∗0.0620.0230.105∗∗(0.180)(0.036)(0.038)(0.049)Electricity×Small-0.609∗∗∗0.120∗∗0.0680.138∗∗(0.139)(0.047)(0.052)(0.055)FossilFuel×Large-0.154∗∗∗-0.596∗∗∗-0.468∗∗∗-0.535∗∗∗(0.015)(0.045)(0.038)(0.072)FossilFuel×Medium-0.174∗∗∗-0.550∗∗∗-0.496∗∗∗-0.467∗∗∗(0.023)(0.048)(0.054)(0.071)FossilFuel×Small-0.223∗∗∗-0.517∗∗∗-0.517∗∗∗-0.372∗∗∗(0.020)(0.051)(0.059)(0.067)RMSE0.844991.00140.950580.84883R20.860550.766310.771410.75677AdjustedR20.836460.722040.722210.69667F-test,p-value0.584480.985080.976670.98821Observations68,20057,52542,89531,180#firms12,28412,2857,5386,074PanelIVWagesallworkersWagesproductionworkersEmp.allworkersEmp.productionworkersAbsorption-workers(13)(14)(15)(16)Electricity×Large0.013-0.005-0.0060.004(0.014)(0.005)(0.008)(0.035)Electricity×Medium-0.020-0.014∗∗0.001-0.022(0.012)(0.006)(0.007)(0.029)Electricity×Small-0.053∗∗∗-0.0090.017∗∗-0.075∗∗(0.016)(0.006)(0.006)(0.032)FossilFuel×Large-0.024∗∗∗0.0010.010∗∗-0.008(0.006)(0.005)(0.003)(0.017)FossilFuel×Medium-0.0070.0020.002-0.011(0.006)(0.003)(0.003)(0.016)FossilFuel×Small0.015∗0.00040.0008-0.019(0.008)(0.006)(0.004)(0.023)AdjustedR20.398540.538770.472580.78191WithinAdjustedR20.003510.009980.004520.02166F-test,p-value0.552070.999550.025930.12069Observations64,61242,43364,64264,218#firms12,1588,90412,23512,235Fixed-effectsPlantYesYesYesYesIndustry-YearYesYesYesYesRegion-YearYesYesYesYesNote:Estimatesaccordingtofixed-effectsmodeldescribedinequations1and4andincludefirm-effectsandindustry-yearandregion-yeareffectsaswellasinteractionsandlevelsoftherespectivefirm-levelheterogeneityvariables.Clustered(plant&year)standard-errorsinparentheses.Signif.Codes::0.01,:0.05,:0.117Table3:Resultsbyownership-FEPanelIMachinery/outputICT/outputVehicles/outputAssetbalance/outputInvestments(1)(2)(3)(4)Electricity×Domestic-0.013-0.1130.050-0.020(0.028)(0.182)(0.040)(0.012)Electricity×Foreign0.0320.204-0.124-0.064∗∗(0.058)(0.336)(0.085)(0.032)FossilFuel×Domestic0.022∗0.1090.045∗∗0.025∗∗∗(0.013)(0.073)(0.019)(0.006)FossilFuel×Foreign0.078∗∗∗0.0550.0290.065∗∗∗(0.026)(0.166)(0.036)(0.013)RMSE1.22850.957230.983200.71800R20.472170.691720.696540.72276AdjustedR20.343770.346210.559410.67242F-test,p-value1.00001.00000.999390.99891Observations39,9222,14318,71963,920#firms9,0971,1536,60211,609PanelIIOutput/workerValue-added/workerTFPProfitmarginProductivity/survival(5)(6)(7)(8)Electricity×Domestic-0.021∗∗-0.024∗∗-0.004∗∗-0.017∗(0.009)(0.010)(0.002)(0.009)Electricity×Foreign0.0220.0510.001-0.008(0.033)(0.039)(0.003)(0.028)FossilFuel×Domestic0.007∗0.0040.0007-0.009∗∗∗(0.003)(0.004)(0.0008)(0.001)FossilFuel×Foreign0.016∗0.0040.003∗∗-0.014(0.009)(0.016)(0.001)(0.010)RMSE0.363610.565500.032490.48342R20.884750.747210.956110.48955AdjustedR20.864880.702260.938130.39783F-test,p-value0.315090.993780.001051.0000Observations67,98265,50213,47064,959#firms12,22812,2324,36312,277PanelIIIQnt.ElectricityQnt.FossilFuelsQnt.DieselQnt.LPGSubstitution(9)(10)(11)(12)Electricity×Domestic-0.634∗∗∗0.0250.0110.085∗(0.176)(0.035)(0.041)(0.042)Electricity×Foreign-0.516∗∗-0.157∗0.183∗0.070(0.245)(0.076)(0.088)(0.121)FossilFuel×Domestic0.173∗∗∗-0.558∗∗∗-0.477∗∗∗-0.472∗∗∗(0.015)(0.044)(0.045)(0.072)FossilFuel×Foreign0.161∗∗∗-0.666∗∗∗-0.566∗∗∗-0.631∗∗∗(0.019)(0.051)(0.055)(0.073)RMSE0.851901.00720.956840.85131R20.858230.763680.768550.75529AdjustedR20.833770.718960.718790.69489F-test,p-value0.569860.983630.974780.98616Observations68,24257,56142,92531,197#firms12,28412,2857,5386,074PanelIVWagesallworkersWagesproductionworkersEmp.allworkersEmp.productionworkersAbsorption-workers(13)(14)(15)(16)Electricity×Domestic-0.003-0.019-0.024∗∗-0.032∗∗(0.006)(0.017)(0.009)(0.012)Electricity×Foreign0.048∗∗0.043-0.043∗∗-0.063∗∗∗(0.019)(0.054)(0.016)(0.020)FossilFuel×Domestic0.006∗∗0.0030.0030.003(0.002)(0.007)(0.004)(0.005)FossilFuel×Foreign-0.002-0.0130.0150.016(0.008)(0.020)(0.009)(0.010)RMSE0.236350.568180.335250.39468R20.867440.753730.911880.89359AdjustedR20.844590.702920.896680.87516F-test,p-value0.488890.998530.096250.23316Observations67,96745,78268,00467,575#firms12,2078,95512,28512,285Note:Estimatesaccordingtofixed-effectsmodeldescribedinequations1and4andincludefirm-effectsandindustry-yearandregion-yeareffectsaswellasinteractionsandlevelsoftherespectivefirm-levelheterogeneityvariables.Clustered(plant&year)standard-errorsinparentheses.Signif.Codes::0.01,:0.05,:0.118ExporterStatus.Weseparatefirmsbasedontheirinternationalmarketengage-ment.16TheresultspresentedinTable4canbesummarizedasfollows:Capitalupgradingamongexportersismorestronglyandpositivelyassociatedwithfossilfuelpriceincreases(panelI).Non-exportingfirms,inturn,experienceanegativecorrelationbetweenproductivityindicatorsandelectricitypriceincreases(panelII).Thesameeffectisabsentforexporters.Exportershaveapositiveassociationwithproductivityandfossilfuelhikes.Consideredjointly,ourresultssuggestthatthestrongversionofthePorterHypothesisholdsforexportingfirms.Fornon-exporters,theempiricalevidencelendssupporttotheweakversionofthePorterHypothesisinstead.17Energysubstitutionandabsorptionarecomparableacrossbothfirmtypes(panelsIIIandIV).EnergyIntensity.Finally,weevaluatetheimpactofenergypricevariationscon-ditionalonenergyconsumption.Wedosobyclassifyingthemanufacturingsectorsaseitherenergy-intensiveornon-intensive.18TheresultspresentedinTable5sug-gestthatcapitalupgradingismoreprevalentforenergy-intensivesectors;however,evidencefortheweakPorterHypothesisexistsforenergy-intensiveandnon-intensivesectorsinresponsetofossilfuelpricesurges(panelI).FurtherevidenceinsupportofthestrongversionofthePorterHypothesisisagainobserved,primarilyfortheenergy-intensivesub-sectors(panelII).Intra-fuelsubstitutionandabsorptionofen-ergypricehikesarealsomoreprevalentforenergy-intensivesectors(panelsIII)andIV).16Wedefineafirmasanexporterifitreportsapositivenetvalueofexportearningsofself-madeproductsduringanyperiod.FormoreinformationandsummarystatisticseeTableB.4panelIII.17Theestimatedsignsfornon-exportersarealsoinlinewiththestrongversionofthePorterHypothesisyetremaininsignificant.18Energy-intensivesectorsarefood;pulpandpaper;basicchemicals;refining;ironandsteel;non-ferrousmetals;non-metallicmineralsfollowingthesectorclassificationofEIA(2022).SeeTableB.4panelIVforacross-tabulationbygroups.19Table4:Resultsbyexporterstatus-FEPanelIMachinery/outputICT/outputVehicles/outputAssetbalance/outputInvestments(1)(2)(3)(4)Electricity×Exporter-0.0004-0.264-0.029-0.035(0.034)(0.213)(0.043)(0.034)Electricity×Non-exporter-0.0140.0760.058-0.020(0.035)(0.151)(0.048)(0.022)FossilFuel×Exporter0.045∗∗0.150∗0.0400.042∗∗∗(0.021)(0.072)(0.027)(0.014)FossilFuel×Non-exporter0.0200.0640.0430.025∗∗(0.019)(0.065)(0.025)(0.009)RMSE1.22850.958550.982980.71808R20.472160.690870.696690.72270AdjustedR20.343760.344410.559620.67234F-test,p-value1.00001.00000.999380.99892Observations39,9222,14318,71963,920#firms9,0971,1536,60211,609PanelIIOutput/workerValue-added/workerTFPProfitmarginProductivity/survival(5)(6)(7)(8)Electricity×Exporter-0.0200.003-0.00080.026(0.014)(0.018)(0.002)(0.016)Electricity×Non-exporter-0.015-0.024∗∗-0.006∗∗-0.029∗∗∗(0.010)(0.011)(0.002)(0.010)FossilFuel×Exporter0.016∗∗0.0050.002∗-0.023∗∗∗(0.006)(0.009)(0.0008)(0.007)FossilFuel×Non-exporter0.0040.0040.0003-0.004(0.004)(0.005)(0.0010)(0.005)RMSE0.363360.565490.032430.48328R20.884910.747220.956270.48986AdjustedR20.865060.702270.938350.39819F-test,p-value0.313540.993770.001031.0000Observations67,98265,50213,47064,959#firms12,22812,2324,36312,277PanelIIIQnt.ElectricityQnt.FossilFuelsQnt.DieselQnt.LPGSubstitution(9)(10)(11)(12)Electricity×Exporter-0.725∗∗∗-0.097-0.132∗0.037(0.172)(0.058)(0.067)(0.056)Electricity×Non-exporter-0.592∗∗∗0.0420.0050.104∗∗(0.183)(0.034)(0.038)(0.042)FossilFuel×Exporter0.154∗∗∗-0.608∗∗∗-0.491∗∗∗-0.550∗∗∗(0.012)(0.043)(0.038)(0.075)FossilFuel×Non-exporter0.180∗∗∗-0.555∗∗∗-0.491∗∗∗-0.461∗∗∗(0.018)(0.049)(0.052)(0.070)RMSE0.850801.00850.956720.85131R20.858600.763050.768610.75530AdjustedR20.834200.718210.718860.69489F-test,p-value0.566400.985230.974720.98616Observations68,24257,56142,92531,197#firms12,28412,2857,5386,074PanelIVWagesallworkersWagesproductionworkersEmp.allworkersEmp.productionworkersAbsorption-workers(13)(14)(15)(16)Electricity×Exporter0.0040.025-0.032∗∗-0.044∗∗(0.009)(0.030)(0.014)(0.016)Electricity×Non-exporter0.001-0.026-0.023∗∗-0.032∗∗(0.007)(0.019)(0.009)(0.013)FossilFuel×Exporter0.007-0.0070.0010.006(0.004)(0.009)(0.006)(0.007)FossilFuel×Non-exporter0.0040.0040.0050.004(0.003)(0.007)(0.004)(0.005)RMSE0.236350.568150.334270.39391R20.867440.753760.912400.89400AdjustedR20.844590.702960.897280.87565F-test,p-value0.488880.998530.093450.22952Observations67,96745,78268,00467,575#firms12,2078,95512,28512,285Note:Estimatesaccordingtofixed-effectsmodeldescribedinequations1and4andincludefirm-effectsandindustry-yearandregion-yeareffectsaswellasinteractionsandlevelsoftherespectivefirm-levelheterogeneityvariables.Clustered(plant&year)standard-errorsinparentheses.Signif.Codes::0.01,:0.05,:0.120Table5:Resultsbyenergyintensity-FEPanelIMachinery/outputICT/outputVehicles/outputAssetbalance/outputInvestments(1)(2)(3)(4)Electricity×Energy-intensive-0.008-0.0240.004-0.022(0.033)(0.135)(0.036)(0.053)Electricity×Non-intensive-0.010-0.1180.089∗0.003(0.041)(0.227)(0.046)(0.031)FossilFuel×Energy-intensive0.037∗∗0.0920.048∗∗0.034∗∗∗(0.015)(0.058)(0.023)(0.005)FossilFuel×Non-intensive0.0150.1260.0210.041∗∗∗(0.029)(0.106)(0.035)(0.011)RMSE1.22860.958280.983310.71808R20.472100.691050.696480.72270AdjustedR20.343680.344770.559310.67234F-test,p-value1.00001.00000.999390.99892Observations39,9222,14318,71963,920#firms9,0971,1536,60211,609PanelIIOutput/workerValue-added/workerTFPProfitmarginProductivity/survival(5)(6)(7)(8)Electricity×Energy-intensive-0.015∗-0.011-0.003∗-0.033∗∗(0.009)(0.010)(0.002)(0.014)Electricity×Non-intensive-0.022∗-0.036∗∗-0.005-0.009(0.011)(0.015)(0.003)(0.010)FossilFuel×Energy-intensive0.009∗0.0100.002∗∗∗-0.014∗(0.004)(0.007)(0.0008)(0.007)FossilFuel×Non-intensive0.003∗-0.010-0.003-0.009(0.002)(0.007)(0.002)(0.005)RMSE0.363560.565500.032480.48336R20.884780.747210.956140.48968AdjustedR20.864910.702260.938170.39798F-test,p-value0.314810.993780.001051.0000Observations67,98265,50213,47064,959#firms12,22812,2324,36312,277PanelIIIQnt.ElectricityQnt.FossilFuelsQnt.DieselQnt.LPGSubstitution(9)(10)(11)(12)Electricity×Energy-intensive-0.624∗∗∗0.002-0.0070.085∗(0.181)(0.039)(0.037)(0.045)Electricity×Non-intensive-0.628∗∗∗-0.005-0.1020.074(0.182)(0.042)(0.062)(0.058)FossilFuel×Energy-intensive0.172∗∗∗-0.565∗∗∗-0.514∗∗∗-0.491∗∗∗(0.015)0.112)(0.046)(0.065)FossilFuel×Non-intensive0.171∗∗∗-0.516∗∗∗0.402∗∗∗-0.499∗∗∗(0.020)(0.111)(0.048)(0.088)RMSE0.850981.15550.956680.85076R20.858540.813960.768630.75561AdjustedR20.834130.781900.718890.69529F-test,p-value0.566970.878920.974700.98596Observations68,24268,32642,92531,197#firms12,28412,2857,5386,074PanelIVWagesallworkersWagesproductionworkersEmp.allworkersEmp.productionworkersAbsorption-workers(13)(14)(15)(16)Electricity×Energy-intensive-0.0040.037∗-0.050∗∗∗-0.063∗∗∗(0.007)(0.019)(0.010)(0.013)Electricity×Non-intensive0.004-0.032-0.016∗-0.025∗∗(0.007)(0.023)(0.009)(0.012)FossilFuel×Energy-intensive-0.0030.003-0.0008-0.005(0.003)(0.009)(0.006)(0.007)FossilFuel×Non-intensive0.008∗∗∗-0.0020.0060.008(0.003)(0.008)(0.005)(0.006)RMSE0.236400.566460.334670.39363R20.867380.755220.912180.89415AdjustedR20.844520.704720.897030.87583F-test,p-value0.489510.998400.094590.22821Observations67,96745,78268,00467,575#firms12,2078,95512,28512,285Note:Estimatesaccordingtofixed-effectsmodeldescribedinequations1and4andincludefirm-effectsandindustry-yearandregion-yeareffectsaswellasinteractionsandlevelsoftherespectivefirm-levelheterogeneityvariables.Clustered(plant&year)standard-errorsinparentheses.Signif.Codes::0.01,:0.05,:0.1216.3Non-linearityinResponsesNext,weturntotheissueofpotentialnon-linearitiesinthefirm-levelresponsestoenergypriceincreases.Weexamineiffirms’optimalresponsetoenergypricehikesmaydependontheenergypriceitobserves.Firmsmayresorttodifferentenergy-savingstrategiesatlowenergypricelevelscomparedtohigherprices.Forexample,firmsmayinvestinupdatedproductiontechnologiestoincreaseefficiencyatlowerpricelevels.Still,thiscopingmechanismmaynotbeaccessibletofirmsoncetheyhaveexhaustedthesavingpotentialstemmingfromcapitalupgrading,forexample,athigherenergypricelevels.Itis,therefore,crucialtoanalyzetowhatextentthepreviouslyestablishedcorrelationpatternsarethemselvesdeterminedbytheactualpricelevelsofenergy.TheresultsofthisanalysisaresummarizedinFigure1andshowthatthepos-itivecorrelationbetweenfossilfuelpriceincreasesandcapitalupgradingismorepronouncedwhenfuelpricesarelow(panelI).Athigherfuelpricelevels,thisef-fectbecomessmallerandeventuallyinsignificant.Similareffectscannotbeobservedforelectricityprices.19Thesepatternsmaybesuggestiveofthecostofupgradingelectricity-basedtechnologyrelativetofuel-basedmachinery(panelI,Figure1a).Furthermore,thepreviouslyidentifiedPorter-typeinnovationhypothesisforfossilfuelsseemsmostpronouncedatlow(er)energypricelevelsbutbecomesstatisticallyinsignificantathigherfuellevels(panelII).Bycomparison,thesubstitutionchanneldoesnotexhibitanon-linearrelationshipascorrelationsbetweenelectricityandfos-silfuelpricesandconsumedquantitiesofeitherenergysourcearesimilarduringlow-andhigh-priceenergyregimes(panelIII).Lastly,thepositivewageeffectcorrelatingwithasurgeinfossilfuelpricesbecomessmallerandinsignificantathigherfuelpricelevels(panelIV).Thesefindingssuggestthatfirmsmayresorttodifferentcopingstrategiesde-pendingonenergypricelevels.Atlowerenergypricelevels,adjustmentisnotablythroughtheinnovationchannel,whileothermechanisms,suchasabsorption,maybecomemorerelevantatevenhigherenergypricelevels.19Thecorrelationsbetweenmachineryinvestmentandelectricitypriceincreasesfollowalesspronouncedyetstillnon-linearpattern.Apositivecorrelationwithinnovationisobservedathigherelectricityprices,whilelowerelectricitypricescorrespondtoanegativecorrelationwithinvestmentinmachinery.22Figure1:Non-linearCorrelationsPanelI:InvestmentPanelII:Productivity/SurvivalElectricityFossilFuel123456789123456789−0.10.00.1PercentilesEstimatesMachinery/outputScatteranderrorbar:PercentileFEestimatesand90%confidenceinterval.Horizontallineandshadedarea:estimatesand90%confidenceinterval.Dashedhorizontalline:Zeroline.(a)Machinery/outputElectricityFossilFuel123456789123456789−0.10−0.050.000.05PercentilesEstimatesAssetbalance/outputScatteranderrorbar:PercentileFEestimatesand90%confidenceinterval.Horizontallineandshadedarea:estimatesand90%confidenceinterval.Dashedhorizontalline:Zeroline.(b)Assetbalance/outputElectricityFossilFuel123456789123456789−0.04−0.020.000.02PercentilesEstimatesOutput/workerScatteranderrorbar:PercentileFEestimatesand90%confidenceinterval.Horizontallineandshadedarea:estimatesand90%confidenceinterval.Dashedhorizontalline:Zeroline.(c)Grossoutput/workerElectricityFossilFuel123456789123456789−0.02−0.010.000.01PercentilesEstimatesTFPScatteranderrorbar:PercentileFEestimatesand90%confidenceinterval.Horizontallineandshadedarea:estimatesand90%confidenceinterval.Dashedhorizontalline:Zeroline.(d)TFPPanelIII:SubstitutionElectricityFossilFuel123456789123456789−1.0−0.50.00.5PercentilesEstimatesQnt.ElectricityScatteranderrorbar:PercentileFEestimatesand90%confidenceinterval.Horizontallineandshadedarea:estimatesand90%confidenceinterval.Dashedhorizontalline:Zeroline.(e)QuantityElectricityElectricityFossilFuel123456789123456789−0.50−0.250.00PercentilesEstimatesQnt.FossilFuelScatteranderrorbar:PercentileFEestimatesand90%confidenceinterval.Horizontallineandshadedarea:estimatesand90%confidenceinterval.Dashedhorizontalline:Zeroline.(f)QuantityFossilFuelPanelIV:Absorption-workersElectricityFossilFuel123456789123456789−0.010.000.010.02PercentilesEstimatesWagesallworkersScatteranderrorbar:PercentileFEestimatesand90%confidenceinterval.Horizontallineandshadedarea:estimatesand90%confidenceinterval.Dashedhorizontalline:Zeroline.(g)WagesallworkersElectricityFossilFuel123456789123456789−0.050−0.0250.0000.025PercentilesEstimatesWagesproductionworkersScatteranderrorbar:PercentileFEestimatesand90%confidenceinterval.Horizontallineandshadedarea:estimatesand90%confidenceinterval.Dashedhorizontalline:Zeroline.(h)WagesproductionworkersElectricityFossilFuel123456789123456789−0.10−0.050.000.05PercentilesEstimatesEmp.allworkersScatteranderrorbar:PercentileFEestimatesand90%confidenceinterval.Horizontallineandshadedarea:estimatesand90%confidenceinterval.Dashedhorizontalline:Zeroline.(i)EmploymentallworkersElectricityFossilFuel123456789123456789−0.04−0.020.00PercentilesEstimatesEmp.productionworkersScatteranderrorbar:PercentileFEestimatesand90%confidenceinterval.Horizontallineandshadedarea:estimatesand90%confidenceinterval.Dashedhorizontalline:Zeroline.(j)EmploymentproductionworkersNote:Y-axisplotthecoefficientsofthenon-linearFEregressionasdescribedinEquation5andbaselineFEregression(Equation1).Scattersanderrorbars:PercentileFEestimatesand90%confidenceintervalofnon-linearFEregressions.Coloredhorizontallinesandshadedareas:Pointestimatesand90%confidenceintervalofbaselineFEregressions.Dashedhorizontalline:Zeroline.6.4RobustnessChecksInstrumentalVariablesEstimation.Weaddresspotentialendogeneitybyem-ployingaspatial/leave-one-outinstrumenttoexploitthegeographicalvariationinenergypricesfromthecostofdistributingenergytoaparticularprovincetoelim-inatetime-varyingunobservedvariationintheenergypricesfacedbyfirmi.Aswediscussinsection5.4,weprovidetwovariationsofthespatialinstrumentwhereeitherexcludingthesameISIC2-digitsector(IV1)or4-digitsector(IV2)toaddresspotentialspillovereffectsthatcouldpotentiallyinvalidateourIVdesign.20TheresultsoftheIVestimationsareinlinewiththatofthebaselinemodelandsimilarinmagnitude.ResultsfrominstrumentalvariableIV1andIV2,presentedinTable6andTa-ble7generallyconfirmtherobustnessofourOLSresultsinbaselinemodel.ThemostnotabledifferencesarethattheeffectoffossilfuelincreasesonICTcapitalupgrading(2)andwages(13)arenowinsignificantinthetwoIVspecifications.Thesameistrueforthenegativecorrelationofelectricitypricehikeswiththeprofitmar-gin(17).Incontrast,theIVestimatorsreportaslightlyhigherpriceelasticityofelectricity.Finally,wagesbyproductionworkersarenowfoundtosignificantlycon-tractinresponsetoanelectricitypricehike(14)withIV1.ThiseffectisstatisticallyinsignificantinthebaselinemodelandintheIV2specification.20WeprovidetheFirstStageestimatesforbothvariationsoftheinstrumentintablesB.5andB.6,respectively.Acrossallestimatedmodels,theF-statisticsliebetween118and148,suggestingthattheselectedinstrumentisstrong.Inbothcases,theinstrumenthasapositiveassociationwithelectricitypricesconsistentwiththeresultsreportedforIndonesiainCal`ıetal.(2022).Sinceplants’electricityisgeneratedbyutilizingfuel,higherfuelpricesareimplicitlyincludedinelectricityprices.24Table6:InstrumentalVariableResults-IV1PanelIMachinery/outputICT/outputVehicles/outputAssetbalance/outputInvestments(1)(2)(3)(4)Electricity-0.020-0.0620.013-0.026(0.027)(0.186)(0.034)(0.023)FossilFuels0.028∗0.0510.045∗-0.032∗∗∗(0.015)(0.066)(0.022)(0.009)FitstatisticsAdjustedR20.351210.375440.565740.67408WithinAdjustedR20.001150.001390.002240.01570Observations39,9222,14318,71963,920Weakid.test59.4102.069.8110.1PanelIIOutput/workerValue-added/workerTFPExitProductivity/survival(5)(6)(7)(8)Electricity-0.022∗∗-0.021∗∗-0.005∗0.010∗∗(0.009)(0.010)(0.003)(0.003797)FossilFuels0.008∗∗0.003-0.0004-0.003∗∗∗(0.004)(0.005)(0.001)(0.0007308)FitstatisticsAdjustedR20.869300.711510.818880.18521WithinAdjustedR20.003970.002810.002090.00622Observations67,98265,50213,47053,887Weakid.test92.683.743.743.9PanelIIIQnt.ElectricityQnt.FossilFuelsQnt.DieselQnt.LPGSubstitution(9)(10)(11)(12)Electricity-0.860∗∗∗0.0300.00060.070(0.064)(0.035)(0.042)(0.043)FossilFuels-0.183∗∗∗-0.543∗∗∗-0.451∗∗∗-0.491∗∗∗(0.012)(0.044)(0.039)(0.063)FitstatisticsAdjustedR20.880090.754310.769210.70066WithinAdjustedR20.126240.098200.074390.08329Observations68,24257,56142,92531,197Weakid.test81.973.8116.373.5PanelIVWagesallworkersWagesproductionworkersEmp.allworkersEmp.productionworkersAbsorption-workers(13)(14)(15)(16)Electricity-0.003-0.029∗-0.025∗∗-0.037∗∗(0.006)(0.015)(0.010)(0.014)FossilFuels0.0040.0020.0020.005(0.002)(0.007)(0.004)(0.004)FitstatisticsAdjustedR20.852800.695450.900480.87769WithinAdjustedR20.001390.008480.001790.00182Observations67,96745,78268,00467,575Weakid.test111.555.938.662.9PanelVProfitmarginSales/outputCosts/outputReturnonSalesAbsorption-businessmetrics(17)(18)(19)(20)Electricity-0.014-0.009∗∗0.002-0.043∗(0.008)(0.003)(0.005)(0.024)FossilFuels-0.010∗∗0.00060.005∗-0.019∗(0.004)(0.003)(0.003)(0.010)FitstatisticsAdjustedR20.402140.544100.477550.77830WithinAdjustedR20.000980.000910.000640.00137Observations64,95964,74268,22835,097Weakid.test119.590.795.281.2Fixed-effectsPlantYesYesYesYesIndustry-YearYesYesYesYesRegion-YearYesYesYesYestNote:EstimatesarebasedonFEmodeldescribedinequations1and7andincludefirm-effectsandindustry-yearandregion-yeareffects.Clustered(plant&year)standard-errorsinparentheses.Signif.Codes::0.01,:0.05,:0.1WeakidentificationteststatisticforinstrumentalvariablesfollowingChaoandSwanson(2005).25Table7:InstrumentalVariableResults-IV2PanelIMachinery/outputICT/outputVehicles/outputAssetbalance/outputInvestments(1)(2)(3)(4)Electricity-0.028-0.1130.027-0.030(0.026)(0.195)(0.036)(0.023)FossilFuels0.027∗0.0680.052∗∗-0.034∗∗∗(0.015)(0.070)(0.020)(0.008)FitstatisticsAdjustedR20.350310.299920.568270.67441WithinAdjustedR20.001060.002350.001660.01534Observations39,9222,14318,71963,920Weakid.test42.652.997.3107.4PanelIIOutput/workerValue-added/workerTFPExitProductivity/survival(5)(6)(7)(8)Electricity-0.017-0.016-0.0040.010∗∗(0.010)(0.011)(0.002)(0.003797)FossilFuels0.007∗0.0040.0002-0.003∗∗∗(0.004)(0.005)(0.0009)(0.0007308)FitstatisticsAdjustedR20.868840.710600.888520.18521WithinAdjustedR20.003750.002670.005050.00622Observations67,98265,50213,47053,887Weakid.test102.063.242.043.9PanelIIIQnt.ElectricityQnt.FossilFuelsQnt.DieselQnt.LPGSubstitution(9)(10)(11)(12)Electricity-0.900∗∗∗0.0320.0040.077∗(0.034)(0.035)(0.042)(0.044)FossilFuels-0.189∗∗∗-0.567∗∗∗-0.478∗∗∗-0.503∗∗∗(0.010)(0.047)(0.043)(0.062)FitstatisticsAdjustedR20.881450.737190.744280.69602WithinAdjustedR20.124600.099270.074390.08509Observations64,86654,77640,56730,069Weakid.test59.355.267.766.7PanelIVWagesallworkersWagesproductionworkersEmp.allworkersEmp.productionworkersAbsorption-workers(13)(14)(15)(16)Electricity0.003-0.019-0.025∗∗-0.037∗∗(0.007)(0.018)(0.010)(0.014)FossilFuels0.0040.0020.007∗0.009∗(0.002)(0.007)(0.004)(0.004)FitstatisticsAdjustedR20.850100.695630.898630.87635WithinAdjustedR20.001500.008510.001590.00173Observations67,96745,78268,00467,575Weakid.test73.595.959.850.3PanelVProfitmarginSales/outputCosts/outputReturnonSalesAbsorption-businessmetrics(17)(18)(19)(20)Electricity-0.013-0.012∗∗∗0.001-0.040(0.009)(0.004)(0.005)(0.023)FossilFuels-0.010∗∗0.00080.004-0.017(0.004)(0.003)(0.002)(0.010)FitstatisticsAdjustedR20.399030.540650.476330.77868WithinAdjustedR20.000920.000880.000520.00108Observations64,95964,74268,22835,097Weakid.test96.3111.972.9106.3Fixed-effectsPlantYesYesYesYesIndustry-YearYesYesYesYesRegion-YearYesYesYesYestNote:EstimatesarebasedonFEmodeldescribedinequations1and7andincludefirm-effectsandindustry-yearandregion-yeareffects.Clustered(plant&year)standard-errorsinparentheses.Signif.Codes::0.01,:0.05,:0.1WeakidentificationteststatisticforinstrumentalvariablesfollowingChaoandSwanson(2005).26ContemporaneousversusMedium-termEffects.Weareconcernedthatitmighttaketimeforfirmstomakedecisionsontheirinvestmentsthatwould,inturn,translateintoproductivityeffectswithfurtherlags.Bycomparison,substitutionandabsorptioneffectscouldpotentiallybeinstantaneous,althoughtheymayshowfurtheradjustmentsoverthemediumtolongrun.21Tothisend,weexperimentwithlagsofenergypriceschangesinEquation1toanalyzetheintertemporalcomponentoftheresponsepatternsininvestmentandproductivityeffects.Weestimatethemodel:yit=βixi,t−ℓ+εit,(11)whereℓ={1,3,5},i.e.,wecomparethecontemporaneouseffectwiththeone-,three-andfive-yearlaggedobservationsoftheelectricityandfossilfuelprices,respectively.TheresultspresentedinFigure2suggestthatthenegativecorrelationofelectric-itypricehikesoninvestmentindicatorsisnolongersignificantforlaggedmodels.Atthesametime,productivityeffectsarestrongerforfiveyearperiodwhenconsideringelectricityprices(panels2aand2b).22SensitivitytoSamplePeriod.Finally,weconfirmtherobustnessofourresultsbyprovidingasensitivityanalysiswhereweremovethefinalthreeyearsofthesamplewhereweobserveastrongerdeviationofthereportedrawdatasamplevis-a-visourreconstructeddataset.Wedothistoensurethattheresultsofouranalysisarenotdrivenbyourmatchingalgorithmwhileacknowledgingthatthecorrelationsbetweenthevariablesmaychangeovertime.Thisimpliesthatafurthertruncationofthedatasetmay,therefore,pickupintertemporalvariationsratherthaneliminateanypotentialcontagionstemmingfromourdatageneration).TheresultsinTable8confirmthepreviouslyestablishedrelationshipsbetweenenergypricesandfirm-leveloutcomevariables.21Historically,laggedvaluesoftheexplanatoryvariablesarealsousedasvalidinstrumentsal-thoughsubjecttosomecriticism.Theliteratureonwhetherlaggedexplanatoryvariablesareeffectiveinsurmountingendogeneityconcernsisscarce.Amongthefewcontributions,Reed(2015)andBellemareetal.(2017)evaluatethepracticeofreplacinganendogenousvariablewithitslag.Reed(2015)focusesonsituationswhereendogeneitystemsfromsimultaneitybetweenyandx.Inturn,Bellemareetal.(2017)identifythelackofserialcorrelationinthepotentiallyendoge-nousexplanatoryvariableandserialcorrelationamongtheunobservedsourcesofendogeneityasconditionsthatcouldleadtoincorrectinferenceswhenemployinglaggedvariablesasinstruments.Consequently,werefrainfromemployingsuchalaggedIVsetupinthisstudy.22Resultspertainingtosubstitutionandabsorptioneffectsareinstantaneousinmostcaseswithstrongeradjustmentsinfive-yearperiods.27Figure2:Contemporaneousvsmedium-termEffectsEffectsofelectricitypriceEstimateand90%Conf.Int.−0.2−0.10.00.10.20.3Grossaddition/outputVehicles/outputMachinery/outputAdvertisement/outputInvestmentLags1−period3−period5−period(a)InvestmenteffectselectricitypricesEffectsoffossilfuelspriceEstimateand90%Conf.Int.−0.050.000.050.10Grossaddition/outputVehicles/outputMachinery/outputAdvertisement/outputInvestmentLags1−period3−period5−period(b)InvestmenteffectsfossilfuelpricesEffectsofelectricitypriceEstimateand90%Conf.Int.−0.020.000.020.040.060.080.10TFP(ACFCD)Value−added/workerOutput/workerProductivityLags1−period3−period5−period(c)ProductivityeffectsofelectricitypricesEffectsoffossilfuelspriceEstimateand90%Conf.Int.−0.015−0.0050.0050.0100.015TFP(ACFCD)Value−added/workerOutput/workerProductivityLags1−period3−period5−period(d)ProductivityeffectsoffossilfuelpricesNote:EstimatesbasedonEquation11.Scattersanderrorbars:FEpointestimatesand90%confidenceintervals.28Table8:ResultsSub-samplePanelIMachinery/outputICT/outputVehicles/outputAssetbalance/outputInvestments(1)(2)(3)(4)Electricity-0.027-0.0800.002-0.029(0.029)(0.173)(0.036)(0.022)FossilFuels0.034∗∗0.0740.048∗-0.029∗∗∗(0.015)(0.081)(0.025)(0.010)FitstatisticsRMSE1.23590.958290.978300.66044R20.481960.691040.712420.75643AdjustedR20.337750.346700.561070.70543F-test,p-value1.00001.00000.999730.99859Observations29,2042,14313,55946,779#firms7,2841,1535,1469,427PanelIIOutput/workerValue-added/workerTFPExitProductivity/survival(5)(6)(7)(8)Electricity-0.022∗∗-0.026∗∗-0.003∗0.009∗∗(0.009)(0.010)(0.002)(0.005)FossilFuels0.007∗0.0020.001-0.004∗∗∗(0.003)(0.006)(0.0007)(0.001)FitstatisticsRMSE0.348470.551340.031770.22433R20.899780.767860.960450.40215AdjustedR20.879960.720520.943210.26693F-test,p-value0.432450.995810.009741.0000Observations49,81447,87510,66036,817#firms9,9589,9603,5358,079PanelIIIQnt.ElectricityQnt.FossilFuelsQnt.DieselQnt.LPGSubstitution(9)(10)(11)(12)Electricity-0.857∗∗∗0.0360.0150.082∗(0.068)(0.034)(0.042)(0.039)FossilFuels-0.178∗∗∗-0.518∗∗∗-0.433∗∗∗-0.458∗∗∗(0.013)(0.047)(0.041)(0.068)FitstatisticsRMSE0.920191.00230.959350.88167R20.850040.760640.761840.75881AdjustedR20.820360.707780.702640.68901F-test,p-value0.847340.996890.995680.99631Observations50,05742,37332,67722,207#firms10,01510,0166,4474,885PanelIVWagesallworkersWagesproductionworkersEmp.allworkersEmp.productionworkersAbsorption-workers(13)(14)(15)(16)Electricity-0.009-0.031∗-0.020∗-0.030(0.008)(0.016)(0.011)(0.022)FossilFuels0.0030.00070.008∗0.009(0.003)(0.007)(0.004)(0.006)FitstatisticsRMSE0.228530.568240.323920.35748R20.873310.753670.920320.91481AdjustedR20.848240.702880.904540.89792F-test,p-value0.688040.998050.218070.27194Observations49,79545,78249,82649,566#firms9,9438,95510,01610,016PanelVProfitmarginSales/outputCosts/outputReturnonSalesAbsorption-businessmetrics(17)(18)(19)(20)FossilFuels-0.010∗∗0.00080.005∗-0.021∗(0.004)(0.003)(0.003)(0.011)Electricity-0.016∗-0.009∗∗0.002-0.046∗(0.008)(0.003)(0.005)(0.024)FitstatisticsRMSE0.478500.217940.232520.74621R20.509040.660510.572030.84141AdjustedR20.407620.592400.487390.79057F-test,p-value1.00001.00001.00000.89491Observations47,52947,23850,06626,255#firms10,0079,50610,0089,983PlantYesYesYesYesIndustry-YearYesYesYesYesRegion-YearYesYesYesYestNote:EstimatesaccordingtoFEmodeldescribedinequations1and6includefirm-effectsaswellasindustry-yearandregion-yeareffects.Clustered(plant&year)standard-errorsinparentheses.Signif.Codes::0.01,:0.05,:0.17ConclusionThispapercontributestounderstandingtheenergyprice-firmcompetitivenessnexusinanemergingeconomycontext.Firmsusefourmaincopingmechanismstonavi-gateenergypriceincreases:innovationandcompetitiveness,substitutionacrossfueltypesandotherinputs,andabsorptionorpass-through.Ourresultsindicatethatwhileelectricitypriceincreasesareassociatedwithreducedproductionandfirmsize,fossilfuelpricehikesresultinincreasedcapitalinvestment,manifestinginimprovedproductivity,therebysupportingthestrongversionofthePorterHypothesis.How-ever,theseproductivityincreasesdonottranslateintohigherprofitability(atleastnotintheshortrun),indicatingthatsuchpricehikesareatleastpartlyabsorbedbyfirms.Thesebroaderresultsmaskheterogeneitybyfirmattributes.Thestrongver-sionofthePorterHypothesiscanonlybeobservedinlargefirms.Smallfirmsaremorenegativelyaffectedbysurgesinenergyprices.Likewise,exportersandforeign-ownedfirmsarealsolessaffectedcomparedtodomesticallyorientedfirms.Moreover,energy-intensivefirmsarealsofoundtoexperiencemoreextensivePorter-typeinno-vations,andtheirbusinessmetricsandemploymentaretypicallymoreaffectedbyenergypricehikesthanfirmsinnon-energy-intensivesectors.Giventhefirmheterogeneityinoutcomes,policyreformsaffectingenergypricesandaccompanyingmeasuresarebesttargetedbasedonsolidmicro-levelanalyses.Themostvulnerablefirms,suchasthesmalleranddomesticallyorientedfirms,maynothavethemeanstoadjusttoenergypricefluctuationsandmayneedcomplemen-tarysupporttoundertakethenecessarychangesandinvestments.Abetterunder-standingofhowexistinginequalitiesinteractwiththerisksposedbyenergypricepolicies,takingintoaccountfirmcapabilitiesandmanagementskills,willinformamoreefficientpolicydesign.2323Awordofcautioniswarrantedhereinover-interpreting.Theseresultsremainbroadlyillustra-tiveastheybuildonthepubliclyavailableversionoftheENIAdataset,whichincludesfirm-levelpaneldataupto2015.Additionalworkisrequiredtomergepost-2015yearstothepaneldimensionofthedatasetandincludetheintroductionoftheexogenousCarbontax,whichbecameeffectivein2018,intheobservationperiod.30ReferencesAckerberg,D.A.,Caves,K.,andFrazer,G.(2015).Identificationpropertiesofrecentproductionfunctionestimators.Econometrica,83(6):2411–2451.Amann,J.,Cantore,N.,Cal´ı,M.,Todorov,V.,andCheng,C.F.C.(2021).Switchingitup:Theeffectofenergypricereformsinoman.WorldDevelopment,142:105252.Ambec,S.,Cohen,M.A.,Elgie,S.,andLanoie,P.(2013).ThePorterhypothesisat20:canenvironmentalregulationenhanceinnovationandcompetitiveness?Reviewofenvironmentaleconomicsandpolicy,7(1):2–22.Arlinghaus,J.(2015).ImpactsofCarbonPricesonIndicatorsofCompetitiveness:AReviewofEmpiricalFindings—OECDEnvironmentWorkingPapers—OECDiLibrary.Bellemare,M.F.,Masaki,T.,andPepinsky,T.B.(2017).Laggedexplanatoryvariablesandtheestimationofcausaleffect.TheJournalofPolitics,79(3):949–963.Betz,T.,Cook,S.J.,andHollenbach,F.M.(2018).Ontheuseandabuseofspatialinstruments.PoliticalAnalysis,26(4):474–479.Brucal,A.andDechezleprˆetre,A.(2021).Assessingtheimpactofenergypricesonplant-levelenvironmentalandeconomicperformance:Evidencefromindonesianmanufacturers.Burke,M.,Craxton,M.,Kolstad,C.D.,Onda,C.,Allcott,H.,Baker,E.,Barrage,L.,Carson,R.,Gillingham,K.,Graff-Zivin,J.,andOthers(2016).Opportunitiesforadvancesinclimatechangeeconomics.Science,352(6283):292–293.Cal`ı,M.,Cantore,N.,Iacovone,L.,Pereira-L´opez,M.,andPresidente,G.(2022).Toomuchenergytheperverseeffectoflowfuelpricesonfirms.JournalofEnvi-ronmentalEconomicsandManagement,111:102587.Cal`ı,M.,Cantore,N.,Marin,G.,Mazzanti,M.,Nicolli,F.,andPresidente,G.(2023).Energypricesandtheeconomicperformanceoffirmsinemergingcountries.StructuralChangeandEconomicDynamics,inprint.Cameron,A.C.,Gelbach,J.B.,andMiller,D.L.(2011).Robustinferencewithmultiwayclustering.JournalofBusiness&EconomicStatistics,29(2):238–249.31Cansino,J.M.,S´anchez-Braza,A.,andRodr´ıguez-Ar´evalo,M.L.(2018).HowcanChilemoveawayfromahighcarboneconomy?EnergyEconomics,69:350–366.Chao,J.C.andSwanson,N.R.(2005).Consistentestimationwithalargenumberofweakinstruments.Econometrica,73(5):1673–1692.Chateau,J.,Bibas,R.,andLanzi,E.(2018).Impactsofgreengrowthpoliciesonlabourmarketsandwageincomedistribution.OECDEnvironmentWorkingPapers,137.Coste,A.,Cal`ı,M.,Cantore,N.,andHeine,D.(2018).StayingCompetitive:Pro-ductivityEffectsofEnvironmentalTaxes.FiscalPoliciesforDevelopmentandClimateAction.Dechezleprˆetre,A.,Nachtigall,D.,andStadler,B.(2020).TheeffectofenergypricesandenvironmentalpolicystringencyonmanufacturingemploymentinOECDcountries:Sector-andfirm-levelevidence—OECDEconomicsDepartmentWork-ingPapers—OECDiLibrary.El-Katiri,L.andFattouh,B.(2017).ABriefPoliticalEconomyofEnergySubsidiesintheMiddleEastandNorthAfrica.Gillingham,K.andPalmer,K.(2014).Bridgingtheenergyefficiencygap:Policyinsightsfromeconomictheoryandempiricalevidence.ReviewofEnvironmentalEconomicsandPolicy.Grover,A.andKarplus,V.J.(2020).Theenergy-managementnexusinfirms.PolicyResearchWorkingPaper,9397.Haraguchi,N.andAmann,J.(2023).Expandedrealvalue-addeddataformanufac-turingsub-sectorsandpatternsofmanufacturingdevelopment.UNIDOInclusiveandSustainableIndustrialDevelopmentWorkingPaperSeries,2(2023).INE(2015).Estructuradelaindustriamanufacturera.Jakob,M.andSteckel,J.C.(2014).Howclimatechangemitigationcouldharmdevelopmentinpoorcountries.WIREsClimateChange,5(2):161–168.Joltreau,E.andSommerfeld,K.(2019).Whydoesemissionstradingundertheeuemissionstradingsystem(ets)notaffectfirms’competitiveness?empiricalfindingsfromtheliterature.Climatepolicy,19(4):453–471.32Kumar,S.andPrabhakar,P.(2020).Industrialenergypricesandexportcom-petitiveness:evidencefromIndia.EnvironmentalEconomicsandPolicyStudies,22(1):1–20.Lin,B.andJiang,Z.(2011).EstimatesofenergysubsidiesinChinaandimpactofenergysubsidyreform.EnergyEconomics,33(2):273–283.Machado,J.A.andSilva,J.S.(2019).Quantilesviamoments.JournalofEcono-metrics,213(1):145–173.Marin,G.andVona,F.(2021).Theimpactofenergypricesonsocioeconomicandenvironmentalperformance:Evidencefromfrenchmanufacturingestablishments,1997–2015.EuropeanEconomicReview,135:103739.Maruejols,L.,H¨oschle,L.,andYu,X.(2022).Vietnambetweeneconomicgrowthandethnicdivergence:Alassoexaminationofincome-mediatedenergyconsumption.EnergyEconomics,114:106222.MEFR(2008).MinisteriodeEconom´ıa,FomentoyReconstrucci´on.LeyNo.20.257.LeyGeneraldeServiciosEl´ectricosRespectodelaGeneraci´ondeEnerg´ıaEl´ectricaconFuentesDeEnerg´ıasRenovablesNoConvencionales.Mundaca,L.,Rom´an,R.,andCansino,J.M.(2015).TowardsaGreenEnergyEcon-omy?Amacroeconomic-climateevaluationofSweden’sCO2emissions.AppliedEnergy,148:196–209.Mundlak,Y.(1978).Onthepoolingoftimeseriesandcrosssectiondata.Econo-metrica:journaloftheEconometricSociety,pages69–85.Petersen,M.A.(2009).Estimatingstandarderrorsinfinancepaneldatasets:Com-paringapproaches.TheReviewoffinancialstudies,22(1):435–480.Porter,M.(1991).America’sgreenstrategy.ScientificAmerican,page96.Porter,M.E.andVanderLinde,C.(1995).Towardanewconceptionoftheenvironment-competitivenessrelationship.Journalofeconomicperspectives,9(4):97–118.Reed,W.R.(2015).Onthepracticeoflaggingvariablestoavoidsimultaneity.OxfordBulletinofEconomicsandStatistics,77(6):897–905.33Rentschler,J.andKornejew,M.(2018).Energypricevariationandcompetitiveness:FirmlevelevidencefromIndonesia.InFossilFuelSubsidyReforms,pages75–106.Routledge.Rentschler,J.,Kornejew,M.,andBazilian,M.(2017).Fossilfuelsubsidyreformsandtheirimpactsonfirms.EnergyPolicy,108:617–623.Serra,P.(2022).Chile’selectricitymarkets:Fourdecadesonfromtheiroriginaldesign.EnergyStrategyReviews,39(September2021):100798.Wooldridge,J.M.(2005).Fixed-effectsandrelatedestimatorsforcorrelatedrandom-coefficientandtreatment-effectpaneldatamodels.ReviewofEconomicsandStatistics,87(2):385–390.Wooldridge,J.M.(2010).Econometricanalysisofcrosssectionandpaneldata.MITpress.34AppendixASupplementaryFiguresFigureA.1:EnergyComposition025507510019952000200520102015%CokeDieselElectricityKeroseneLiqnatgasLPGMethanolNatgasPetrolPipegasPurchasedquantitiesbyenergytype,nationaldataChile.EvolutionenergyconsumptionYears1997,2006highlightedwithdashedlines.Note:Purchasedquantitiesbyenergytype.Years2007and2006highlightedwithdashedlines.Source:Authors’calculationsbasedonENIAdata(INE,2015).35FigureA.2:DistinctPlantsoverTime020004000600019952000200520102015CountedcasesPlantobservedAnyprev.yearLastyearNewAllfirms,countedcases.Distinctplantsovertime,ChileDashedline:numberofuniqueplantsasreportedintherawdata.Note:Dashedline:numberofuniqueplantsreportedinrawdata.Source:Authors’calculationsbasedonENIAdata(INE,2015).FigureA.3:HarmonizedUnitPricesbyEnergyTypeElectricityFossilfuels1995200020052010201519952000200520102015−2.50.02.55.0HarmonisedenergyunitpriceHarmonisedenergyunitpricebyISICsectorovertimeISICII−digitdeflatorsfollowingHA,2022.Dashedline:median.Shadedarea:10th/90th(light);25th/75th(darker)percentile.Note:ISIC2-digitindustry-leveldeflatorsfollowingHaraguchiandAmann(2023).Dashedline:median.Shadedarea:10th/90th(light);25th/75th(darker)percentile.Source:Authors’calculationsbasedonENIAdata(INE,2015).36BSupplementaryTablesB.1DescriptiveStatisticsTableB.1:FossilFuelConversionRatesUnitconversionFossilfuelUnitmeasure(um)um/kgum/m3um/kWhCoalkg18507.00Dieselliter0.850.0019.10Gascubicmeter(m3)0.510.2708.80Gasolineliter0.750.00110.00Keroseneliter0.800.00110.35LPGliter0.510.2706.90Source:Authors’calculationsbasedonENIAdata(INE,2015).37TableB.2:DescriptiveStatisticsVariableResponsesAbbreviationDescriptionMeanMed.LQUQN%IndependentvariablesElectricity(EL)Electricityunitprice0.10.10.10.19386897.17Fossilfuel(FF)Fossilfuelunitprice2.21.80.93.06958272.03InvestmentMachinery/outputInvestmentinfixedassets,machineryandequipment/grossvalueofnetproduction0.10.00.00.05127853.08ICT/outputInvestmentinfixedassets,softwareandICTequipment/grossvalueofnetproduction0.00.00.00.024922.58Vehicles/outputInvestmentinfixedassets,vehicles/grossvalueofnetproduction0.00.00.00.02278223.58Grossaddition/outputGrossadditiontofixedassets/grossvalueofnetproduction.Includesinvestmentinland;buildings;machineryandequipment;vehicles;furniture;softwareandICT;otherassets.0.10.00.00.16108263.23ProductivityOutput/workerGrossvalueofnetproduction/averagetotalcontractworkers68599.823556.513481.748014.49588399.26Value-added/workerNetvalue-added/averagetotalcontractworkers24098.28082.34409.215412.29589299.27TFPTotalFactorProductivityfollowingAckerbergetal.(2015)withCobbDouglasproductionfunction9.69.78.810.51618916.76ExitFirmexitprobability:1iffirmremainsinsampleforatleasttwoconsecutiveyearsanddropsfromthesample(withoutreturning)91.06...8704194.70SubstitutionQnt.ElectricityQuantityofelectricityconsumed10972.084.027.0414.09472498.06Qnt.DieselQuantityofdieselconsumed690.226.08.091.04402245.57Qnt.LPGQuantityofLPGconsumed155.25.02.018.03161232.72Absorption-workforceWagesallworkersAverageremunerationtotalcontractworkers6218.24857.23126.47632.89586399.24WagesproductionworkersAverageremunerationworkersassociatedwiththeindustrialprocess4060.82877.81155.55044.26055862.69Emp.allworkersAveragetotalcontractworkers68.826.015.062.09599299.37Emp.productionworkersAveragecontractworkersassociatedwiththeindustrialprocess51.819.010.046.09516698.52Absorption-businessmetricsProfitmargin(Totalnetincomefromthesaleofproductsandworkperformed/Totalnetcostofgoodsreceivedandworkperformedundercontract-1)0.30.40.20.59638199.77ReturnonSalesTotalnetincomefromthesaleofproductsandworkperformed-Totalnetcostofgoodsreceivedandworkperformedundercontract-Netbalanceoffixedassetsattheendoftheperiod-1016851.213684.2-205588.9129285.19374897.05Costs/outputTotalnetcostofgoodsreceivedandworkperformedundercontract/grossvalueofnetproduction0.60.60.50.89639399.79Note:LQ/UP:Lowerandupperquartile.Top/bottom0.1per-centtrimmed.Source:Authors’calculationsbasedonENIAdata(INE,2015).38B.2LiteratureSummaryTableB.3:RelatedLiteratureAmannetal.(2021)Data•Oman;AnnualIndustrialSurvey;3600manufacturingfirms(2012-2017).Variables•Dependent:value-added/employment;output/employment;TFP;grossprofitmargin;qm;machineryandICTsalesandpurchases•Independent:upm,varioussector-&region-controls.BrucalandDechezleprˆetre(2021)Data•Indonesia;manufacturingindustry;coveringallmediumandlargeenterprises(1980-2015).Variables•Dependent:energyuse;CO2emissions;output;employment;energyandCO2intensity;energy/worker;capitalandcapitalintensity;purchases/salesofland,buildings,machinery,vehicles.•Independent:energyprice(firm-levelaverageenergypriceacrossenergysourcesweightedbyfirm-levelconsumption-sharebyenergytype),varioussector-&region-controls.Cal`ıetal.(2022)Data•Indonesia;StatistikIndustri:manufacturingfirmswith20+employees(1998-2015).•Mexico;EncuestaAnualdelaIndustriaManufacturera;mfn.firms(2009-2015).Variables•Dependent:TFP;profitability;value-added/employment;value-added/kWh;energyefficiency;machineturnover•Independent:upm,varioussector-&region-controls.Cal`ıetal.(2023)Data•WorldBank’sEnterpriseSurveysfor11countriesbetweentheyears2002and2013Variables•Dependent:Totalemployment;sales/(totalemployment);value-added/(totalemployment);returnsonsales;exportshare.•Independent:Energypriceindex,interactedwithenergyexportshare;firmsize;ownership;energy-outagedummy;R&Ddummy.TheEnergypriceindex(EP)givenbyEPcst=logjθjcs,1995×pjctwhereθjcs,1995istheshareofenergysourcej(e.g.crudeoil,naturalgas,electricity,etc.)overtotalenergyuseofsectorsincountrycinyear1995andpjctistherealpriceofenergysourcejincountrycandyeart.39TableB.3continuedfrompreviouspageMarinandVona(2021)Data•France;manufacturingindustry(1997-2015).Variables•Dependent:energyconsumption;CO2emissions;employment;annualwage;employmentsharebyoccu-pationgroup.•Independent:energyprice(inkWh,firm-levelaverageenergypriceacrossenergysourcesweightedbyfirm-levelconsumption-sharebyenergytype),initialcapitalstockoffirmjint=0,varioussector-&region-controls.RentschlerandKornejew(2018)Data•Indonesia;41,402smallandmicrominingandmanufacturingfirms(2013).Variables•Dependent:costshare•Independent:pricesofelectricity,petrol,diesel,kerosene,LP,sector-&region-controls.Abbreviations:upm:unitpriceofenergytypem;qm:physicalquantityofenergytypem;m={Electricity,FossilFuel};j={electricity,naturalgas,petrol,...}.kWh:Kilowatthours;TFP:TotalFactorProductivity;FE:Fixedeffects;IV:InstrumentalVariable.40B.3HeterogeneityinResponsesTableB.4:SummaryStatisticsHeterogeneityAnalysisPanelI:Firm-sizeN%DefinitionResponsesLarge3225232.26Largefirms:≥50employees.Medium3239132.40Mediumfirms:≥20,<50employees.Small3523535.24Smallfirms:<20employees.Non-responsesNofirm-sizeinformation960.10PanelII:OwnershipN%DefinitionResponsesDomestic8721590.28Foreigncapitalformation=0LCUinallpe-riods.Foreign93859.72Foreigncapitalformation>0LCUinanype-riod.PanelIII:ExportengagementN%DefinitionResponsesExporter1971420.41Exporterifnetvalueofexportearningsofself-madeproducts>0LCUinanyperiod.Non-exporter7688679.59Exporterifnetvalueofexportearningsofself-madeproducts=0LCUinallperiods.PanelIV:EnergyintensityN%DefinitionResponsesEnergy-intensive6437964.4Sectors:Food;pulpandpaper;basicchemi-cals;refining;ironandsteel;nonferrousmet-als;nonmetallicminerals.Non-intensive3559535.6AllothersectorsaccordingtoEIA(2022).Source:Authors’calculationsbasedonENIAdata(INE,2015).41B.4InstrumentalVariablesTableB.5:FirstStage-IV1ElectricityFossilFuelModel:(1)(2)VariablesElectricityIV0.889∗∗∗-0.034(0.025)(0.026)FossilFuelIV-0.115∗∗∗0.809∗∗∗(0.037)(0.045)FitstatisticsAdjustedR20.603160.77699WithinAdjustedR20.052390.01972F-test134.43148.09Observations65,05865,058Fixed-effectsPlantYesYesIndustry-YearYesYesRegion-YearYesYesClustered(plant&year)standard-errorsinparenthesesSignif.Codes::0.01,:0.05,:0.1TableB.6:FirstStage-IV2ElectricityFossilFuelModel:(1)(2)VariablesElectricityIV0.785∗∗∗-0.041(0.023)(0.045)FossilFuelIV-0.106∗∗∗0.591∗∗∗(0.022)(0.042)FitstatisticsAdjustedR20.626910.67030WithinAdjustedR20.046940.00930F-test117.91142.66Observations65,05865,058Fixed-effectsPlantYesYesIndustry-YearYesYesRegion-YearYesYesClustered(plant&year)standard-errorsinparenthesesSignif.Codes::0.01,:0.05,:0.142CFirm-levelMatchingAlgorithmENIAismadeavailablebytheNationalStatisticalInstituteintwodistinctforms.Thecombineddatasetidentifiesthesamemanufacturingfirmbetween1995and2007withaconstant5-digitIDyetfailstoprovideinformationonthefirms’locationortheirISIC4-digitclassificationcode.Inturn,thisinformationisavailableinthecorrespondingrawdatafilesofENIA(availablefortheyears1995to2015);however,thesedatadouseafirm-levelidentifierwhichdiffers(a)fromthatofthecombineddatasetand(b)variesovertime.Weproposeasimplealgorithmthatexploitsfirm-levelcharacteristicstomatchfirmsbetweenthecombinedandrawdatasetstoextractinformationonthere-spectivefirms’locationand4-digitmanufacturingsector.Ourprocedureachievesasuccessfulmatchof>95%andfollowsthefollowingsteps:•Foreveryyear,checkifthefollowingcharacteristicsproduceanexactmatchbetweenthecombinedandrawdatasets,respectively:–Yearofthesurvey.–3-digitISICsector.–Shareforeign/domesticownership.–Firmsize.Thereissomeambiguitybetweenthecombinedandrawfirmdatawhenclassifyingfirmsizes,whichrequirestheharmonizationofeightsizegroupsintherawdatato4groupsaspertheharmonizeddata.Thisreclassificationis1-to-1forallbutoneoutoftheeightgroupsidentifiedintherawdata;seeTableC.7.Ifafirmisincategory6intherawdata,itcanbematchedtoeithercode3or4inthecombineddata.•Furthermore,weenforceafuzzymatchforthevalueandamountofelectricityconsumedbyfirmsacrossbothdatasetsasfollows:–Quantityelectricitypurchased(±5%).–Valueelectricitypurchased(±10%)•Becauseofthefuzzynatureofthematch,noexclusive1-to-1matchoutcomeistobeexpected.Consequently,incasemultipleIDscanbematchedbasedontheabovematchcriteria,weidentifythebestmatchby:–ThelongestnumberofmatchedyearsbetweentwoIDs;TableC.7:Firm-sizeClassificationsRawdataIDRawdatacut-offsCombineddataIDCombineddatacut-offs00<=tot.emp.>=040<10employed105<=tot.emp.>=090<10employed210<=tot.emp.>=19110-19employed320<=tot.emp.>=49220-49employed450<=tot.emp.>=99350-249employed5100<=tot.emp.>=1993.6200<=tot.emp.>=4993.5.7500<=tot.emp.>=9994250+employed8tot.emp.>=10004250+employedNote:Firm-sizeclassificationaccordingtorawandcombinedENIAdatasets.Source:Authors’calculationsbasedonENIAdata(INE,2015).–Thelowestaverageabsolutedifferencebetweenthereportedquantity(value)ofelectricitypurchasedthatfallswithinanacceptablerangebe-tweentwoIDs;–Inotherwords,wepickthematchedfirmthatisclosesttothereportedvaluesinthecombineddatasetaslongasthedifferencedoesnotex-ceed5%(10%)ofthereportedquantity(value)ofelectricitypurchased,respectively.•Thisway,wecanuniquelyidentify90percentofallfirmsinthecombineddatasetasdescribedinTableC.8.TableC.8:MatchRatesFirm-levelMatchingAlgorithmMatchesCasesPercentAtmost1uniqueIDmatch10,53890.3Atmost2IDsmatched1,0388.9Atmost3IDsmatched840.7UniqueIDmatchafter2ndround11,15695.6Note:FirmmatchesacrossrawandcombinedENIAdata.Source:Authors’calculationsbasedonENIAdata(INE,2015).•Finally,whiletheoutlinedalgorithmuniquelymatchesaround90%offirmIDs,thealgorithm’snatureissuchthatthesamematchedfirmmaysatisfythe“closestproximity”requirementofthepreviouschat.Thisisthecaseforaround9.5%ofallrecordedcases.Giventhese,were-iteratethepreviousstepbyevaluatingtheclosestproximityconditionalontheproximityoftheother44non-uniquematchesoftheidenticalfirmIDs.Throughthisfinalstep,thematchratebetweenthetwodatasetsisat95.6%;seelastlineinTableC.8.DEnergyUnitPriceAdjustmentAlgorithmTheenergyandelectricitycleaningalgorithmdescribedinthissectionisdesignedtoidentifyandeliminateartificialjumpsintherawenergydatathatarethepotentialresultofdataimputationorreportingissues.Considertheexamplewherequantitiesofanunspecifiedliquidwerereportedintonnesinperiodst0t1andt2,respectively,buttheunitofmeasurementinperiodt1wouldbewrongfullygiveninm3.Insuchacase,theconversionofm3int1totonneswouldleadtoaninter-temporalchangeofreportedquantitiesandunitprices(assuminganotnotablechangeinpurchasedvolumesoverthethreeyears)ofafactorclosetotheconversionfactorbetweentonnesandcubicmeters.Thealgorithmweproposeidentifiesjumpsofthesemagnitudesandoverwritethe(assumedtobewrongfullyreported)quantityinformationifthecorrectedmeasureswouldnotconstituteaminimum/maximumintheseries.Inourexample,thismeansthatthequantitymeasureint1willbeadjusted,iftheadjustedvaluewouldnotbesmaller(larger)thanmin(t0,t2)(max(t0,t2)),respectively.Morespecifically,thecleaningalgorithmappliedtotherespectiveenergyseriestakesthefollowingsteps:1.Foranysequence,i.e.,firmandarbitraryenergyunitprice(up)andarbitrarypointintime(t0)overthetimeintervalt=1,...,T,lookatprevious/nextperiod(t0±1)andobservechangeofunitprice:∆upt0±1=upt0±1upt0.2.Check,if∆upt0±1isoforderofmagnitude1×10k,k={...,−1,0,1,...}:•Ifitis,correctupt0byfactorkdefinedasvariableupkt0:=upt0×k.•Ifitisnot,movetonextsequence.3.Check,ifupkt0issmaller[bigger]thanmin(upt∈T,t̸=t0)[max(upt∈T,t̸=t0)]:•Ifitis,removeobservationupt0.•Ifitisnot,replaceupt0withupkt0.FigureD.1comparesthepre-vspost-adjustmentenergyunitprices.Highlighteddatapointsindicateadjustmentsmadebytheadjustmentalgorithm.Differentcolorshighlightdifferentquantityunites,e.g.,cubicmeters,tonnesetc.45FigureD.1:UnitPricesSeries,pre-andpost-adjustment(a)Electricity,pre-adjustment(b)Electricity,post-adjustment(c)Coal,pre-adjustment(d)Coal,post-adjustment(e)Diesel,pre-adjustment(f)Diesel,post-adjustment(g)Petrol,pre-adjustment(h)Petrol,post-adjustment(i)Kerosene,pre-adjustment(j)Kerosene,post-adjustment(k)LPG,pre-adjustment(l)LPG,post-adjustment(m)Gas,pre-adjustment(n)Gas,post-adjustmentNote:UnitpriceseriesfollowingenergyunitpriceadjustmentalgorithmdescribedinAppendixD.Highlighteddatapointsindicateadjustmentsmadebytheadjustmentalgorithm.Differentcolorsemphasisedifferentquantityunits,e.g.,cubicmeters,tonnesetc.Source:Authors’calculationsbasedonENIAdata(INE,2015).