Journal of Innovation
& Knowledge
https://www.journals.elsevier.com/journal-of-innovation-and-knowledge
Productivity and employment effects of digital complementarities
María Teresa Ballestar
a,∗, Ester Cami ˜ na
b, Ángel Díaz-Chao
c, Joan Torrent-Sellens
daESICBusiness&MarketingSchoolandUniversidadReyJuanCarlos,CaminoValdenigriales,s/n,28223PozuelodeAlarcón,Madrid,Spain
bUniversidadComplutensedeMadridandICAE.Av.Séneca,2,28040Madrid,Spain
cUniversidadReyJuanCarlos,PaseodelosArtilleross/n,28032Vicálvaro,Madrid,Spain
dFacultyofEconomicsandBusiness,OpenUniversityofCataloniaRambladelPoblenou,156,08018Barcelona,Spain
a rt i c l e i nf o
Articlehistory:
Received19September2020 Accepted24October2020 Availableonline27December2020
JEL:
J23 J24 Keywords:
Robotisation
Productivitydigitalisation
a b s t ra c t
Moderneconomicgrowthisnolongerfoundintotalfactorproductivity(TFP)becausetherearegainsfrom technologicalchangethatareneverrecordedinthereturnsfrominnovationorintheNationalAccounts.
Theexistenceofcomplementaritiesamongtechnologiesderivedfromtheuseofrobotics,electronic commerce,orinnovationisdifficulttoassessthroughcountry-levelrecords.Becausetheliteraturehas mainlyfocusedonrobotisationatanaggregateorindustrylevel,researchfocusingonafirmleveland complementaritiesanalysishavebeenlimited.Tofillthegap,inthispaper,weintendtoprovidenewevi- denceregardingtheeffectsofrobotisation,digitisation,andinnovationonproductivityandemployment infirms,byusingalargesampleof5511Spanishmanufacturingfirmsfortheperiod1991–2016.This datacapturesthepayofftothehighratesofinvestmentnecessarytoupgradetheproductiontechnology forfirmsinanewgloballycompetitiveframework.
©2020JournalofInnovation&Knowledge.PublishedbyElsevierEspa ˜na,S.L.U.Thisisanopenaccess articleundertheCCBY-NC-NDlicense(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Introduction
Theincreaseintheuseofautomatisationhashadadifferen- tialeffectacrossoccupationsandsectorsthathasnotbeenfully explored(OECD,2019).Researchhasfocusedmainlyonrobotisa- tionatanaggregateorindustrylevel,andtheresearchfocusingon firmlevelandcomplementaritiesanalysishasbeenscarce,leav- ingmanyquestionsunanswered(Seamans&Raj,2018).Tofillthe gap,inthispaper,weintendtoprovidenewevidenceregardingthe effectsofrobotisation,digitisation,andinnovationonproductivity andemploymentinfirms.
Thegrowingliteratureontheeffectsoftechnologiessuchas robotisation, artificial intelligence (AI), or big data attempts to explain thetransformation experiencedbyfirms sincethe90’s.
However,noclearaccountofwhyandhowthechangesinpro- ductivityareoccurringandthecomplementaritiesamongsomeof thosetechnologieshavebeenprovided.Explanationsrangefrom themismeasurementofcreativedestruction,suggestingthatnew products offerhigher qualityand utilityto consumers(Aghion, Bergeaud,Boppart,Klenow,&Li,2019),tomiscalculationsofthe priceindexesofe-commercegoods,especiallywhenconsidering
∗ Correspondingauthor.
E-mail addresses: [email protected] (M.T. Ballestar), [email protected](E.Cami ˜na),[email protected](Á.Díaz-Chao),[email protected] (J.Torrent-Sellens).
freebiesinonlinemarkets(Brynjolfsson&Oh,2012;Goolsbee&
Klenow,2018)orthatthestoppageoccursmerelybecauseof a reversalinthegrowthoffirms’productivityaftertheinitialgainsof theinformationandcommunicationstechnologies(ICT)revolution (Syverson,2011).
Thereisagreementamongtheliteraturethatthisproductiv- itygrowthinWesterneconomiesisduetotheinnovationofICT firmsthatstartedattheendofthelastcenturyandpersisteddur- ingtheGreat Recession,whichallowedfortheimprovement of livingstandards(Jorgenson,Ho,&Samuels,2011).Theimpactwas bothdirect,throughtheinnovativefirms,andindirect,throughthe effectsthattechnologicalinnovationhadontraditionalsectors,and theeffectwaslargerwhentherewasadigitaltransformationofthe firm(Kijek&Kijek,2019;Kraus,Palmer,Kailer,Kallinger,&Spitzer, 2019),whichrequiresagooddealofcomplementarities,bothin newandestablishedfirmsinwhatSoriano,Martinez-Climent,and Tur-Porcar(2018))calla‘virtuouscircle’thatimproveswelfareand growthbyimprovingorganisationwithinthefirm.
Theassessmentofthis ‘virtuouscircle’,includingtheimpact ofentrepreneurship,theadoptionof oneoftheseforms ofdig- italtransformation,theuseofindustrialrobotics,isincreasingly importantinthecurrentcontextofthenewindustrialrevolution.
Althoughmanycompaniesareeagertoadoptthesetechnologies asawaytoincreaseproductivity,someconcernshavebeenraised aboutthecostimpactofthetransformation,anditseffectonthe workforceduetotraining and newforms ofworkorganisation (Acemoglu,Makhdoumi,Malekian,&Ozdaglar,2019;Díaz-Chao,
https://doi.org/10.1016/j.jik.2020.10.006
2444-569X/©2020JournalofInnovation&Knowledge.PublishedbyElsevierEspa ˜na,S.L.U.ThisisanopenaccessarticleundertheCCBY-NC-NDlicense(http://
creativecommons.org/licenses/by-nc-nd/4.0/).
Sainz-González,&Torrent-Sellens,2015;Pi ˜neiro-Chousa,López- Cabarcos,Romero-Castro,&Pérez-Pico,2020).
Oneofthemissingpieceswhenevaluatinggrowthistherole ofcomplementarities.AsCarlawandLipsey(2002)posited,new technologiesinvigorategrowthbycreatingtechnologicalcomple- mentarities, whichare notadequately measuredby totalfactor productivity (TFP). Gains from technological change will occur throughindirecteffectsandexternalitiesthatarenoteasilyshown in national statistics.To bridgethat gap,weusedata fromthe EncuestasobreEstrategiasEmpresariales(BusinessStrategySur- vey,ESEE).TheESEEprovidesapaneldatarepresentingSpanish fabricmanufacturingbusinesses.Thisannualsurveyisconducted bytheSpanishMinistryofFinanceandPublicAdministrationand coversabroadperiodfrom1990to2015,includingtheeconomic recession and the subsequentrecovery period (Torrent-Sellens, 2018).
The ESEE contains yearly information for approximately 1800 Spanish firms and covers three main areas: the strate- gic decision-making on prices, costs, markets,and investment;
the value process, which involveshuman capital, organisation, entrepreneurialinnovation,Research&Development(R&D),and ICT;andthemostimportantindicatorsand ratiosfrombalance sheetsandprofitsandlosses.
Withthisdata,andthroughamodelderivedfromVanReenen (1997), Kromann, Skaksen, and Sørensen (2011), and DeCanio (2016),weestablishtheeffectofcomplementaritiesintermsof labourproductivityandfirmperformance.Ourfindingsarerele- vanttounderstandingtheimpactofthetransformationtorobotics andallowforthedevelopmentofentrepreneurialstrategiesthat boosttheefficiencyofcompaniesandprovidethemtoolstoprotect themfromnegativefinancialevents,leadingtoanoptimalsizing oftheirworkforce.
Theremainderofthepaperisstructuredasfollows:Section2 presentsaliteraturereviewofdigitalcomplementarities.InSection 3,wederiveamodelthathelpscapturethoseeffectsfromthedatais profuselydescribeit.Section4presentstheeconometricestimates ofthemodel,whichwediscussindetailinSection5.Finally,Section 6presentsaconcisesetofconclusionsandfurtherdirectionsfor research.
Literaturereview
CarlawandLipsey(2002)describethatcomplementarityexists in‘...insituationinwhichthepastorpresentdecisionsoftheiniti- atingagentswithrespecttotheirowntechnologiesaffectthevalue of thereceivingagents’ existingtechnologiesand/ortheirfirms which tend tofollowproactiveinnovation strategies,especially intermsofprocessandtechnologyadoption,likemarketingnew products,improvingtheirquality,increasingcapacityorinvesting ininnovativeprocessesproducethosecomplementarities’.
Buttotakeadvantageofcomplementarities,firmsneedawider varietyoftechnologyresourcesthatimprovetheirperformance.
Currently,robotics,AI,machinelearning,andbigdataareinsep- arablylinkedwithinthatinnovationintheoveralleconomy,but especiallyintheindustrialsector.Ascompanieshaveevolved,they have focused on thenecessity of offeringbetter productswith higherquality,makingtechnologyimprovetheirperformanceand openingspaceforfurtherdevelopments(Dickson&Hadjimanolis, 1998;Kijel&Kijek,2019).
Theeffectsofthesecomplementaritiesarediverseandnotfully understood. Someof the issues, suchas theeffect on employ- ment,haveattractedsubstantialinterestbecausetheproblemis not simplya matter of thoseworkers who willbereplaced by automation,definedastheactionthatreplaceshumanlabouractiv- itywithworkdonebymachines.Thegoalofautomationtrough
complementaritiesistwofold;toincreasequalityandreduceunit cost(Muro,Maxim,Whiton,&Hathaway,2019).Theimpactwill beobservable: TheOrganisation forEconomicCo-operationand Development(2019)estimatesthat14%ofcurrentjobsmaydisap- pearbecauseofautomationand35%maybeseriouslyaffectedby thesamephenomenon.
Theeffectonthelabourmarketwillbeabsolute,eitherdirectly orindirectly,becauseofthesuccessivedisplacementofhumancap- italexpelledfromitspreviouspoststootherpostswithaneed fortraditionalqualifications.Chui,Manyika,andMiremadi(2016)) estimatethatinpredictablephysicalactivities(e.g.productionlines packing),anddependingonthecomplementarities,substitution couldreach78%,andinthosethatarenotforeseeable(e.g.forestry workorfarmers),thereplacementlevelis25%,andsubstitutionon serviceswilldependontheircharacteristicsandontheirsources ofinnovation(Ballestar,SainzandSoriano,2018;Philipson,2020).
GraetzandMichaels(2018)focusonproductivitybyanalysing datafromthe InternationalFederation ofRobotics (IFR)onthe roleofrobots.Theirestimationspresentnochangeinthenumber ofhoursworkedafterincreasingthedensityofrobots;however, withouta changein theircomposition, there is atechnological biasinfavourofemployeeswithhighandmediumqualifications.
Faber(2018)obtainsasimilarresult:theincreaseinthedensity ofautomationintheUnitedStateshasanegativeeffectontheoff- shoringofproductiontoMexicoandaffectsknowledgeasymmetry.
AcemogluandRestrepo(2020)alsousedatafromIFRandesti- matetheeffectofcompetitionbetweenrobotsandhumanlabour onemploymentandwagesintheUnitedStates.Theyfindanet reductioninemploymentbysubstitutionbetweentherobotsand labour,whichmaybelinkedtodifferencesinusebetweensectors.
Withsimilardata,FreyandOsborne(2017)pointtoareduction inemploymentnotonlyinsectorsthataretraditionallyusersof robots,suchastheautomotivesector,butalsoinothersthusfar sheltered,suchasservices.
DoraszelskiandJaumandreu(2018)usefirmdataandashorter samplethanoursfromthesamedatabaseandfindbiasintech- nologicalchange and that there is indeeda technological bias;
however,contrarytowhatwehavecommentedinprecedingpara- graphsroboticshasapositivesigntowardsemployment.Withthe samedata,Ballestar,Díaz-Chao,Sainz,andTorrent-Sellens(2020) confirmasignificant(5%),increasingfactorinproductivitygains linkedtoagreaterpresenceofqualityhumancapitalinthecom- panybutdonotprovidedetailsonhowthosegainsareachieved.
AlsoconfirmingtheintuitionofSeamansandRaj(2018)that firmdatawillgathermoreinsightsintoautomation,Blanas,Gancia, andLee(2019))advancethisideabyusingtheEUKLEMSdatapanel and,througha relativelysimplemodelofjobdemand,findthat workerswhoperformroutineandlow-skilledtasks,whichnor- mallycorrespondstowomenandyoungadultswillbethemost affectedbytheintroductionofrobots,becauseofgainsinproduc- tivity.
Theseanalysesadvance,butdo notdetail,thecharacteristics ofthischangethatcorrespondstothetrainingofemployees,how companieshavechangedtheirmanagementoftheknowledgeof theirhumancapital,andtheeffectsofthesechangeslinkedtotrain- ing.Toachievethis,wemustdeterminethecomplementaritiesof changeinemploymentandproductivitytodeterminethepolicies oftrainingnecessarytoavoidunemploymentandtofocussupport onthegroupsmostdisadvantaged.
Contrary to Graetz and Michaels (2018) or Acemoglu and Restrepo (2020); Autor and Salomons (2017); Doraszelski and Jaumandreu(2018),andBallestaretal.(2021)havedemonstrated thatproductivitygainsareoneofthemajorbenefitsintheexpan- sion,concurringwiththeresultsthatindustriescanbenefitfrom thegrowthof labour-augmentingproductivitythroughcomple- mentarities,andasemploymentseemstofallwithinanindustry,
asindustry-specific productivityincreases,positive spilloversto somesectorsmorethanoffsetthenegativeeffectsonotherindus- triesorsectors.SeamansandRaj(2018)suggestempiricalwork thatprimarilyusesstatisticsaggregatedbyindustryorcountrybut do investigatehowcomplementarities hideunder,for example, marketstructuresandlaboursubstation.
Thisresearchisthefirsttoestimatesomeofthecomplementar- itiesbyusingfirm-leveldataandwhetherthegainsarelimitless, orreachasaturationpoint, assumingthatswitchingtorobotics causesproductivitytoincreasebutdoesnotincreaseitindefinitely, regardlessofthecomplementarities,whichaffectstheTFPgrowth effects of automationonboth hoursworkedand onthelabour share,whichshouldbereflectedinwages,leadingtoanincreasein humanlabourcosts.
Methodology Modelandhypothesis
By adapting VanReenen (1997); Kromannet al.(2011),and DeCanio(2016),ourbaselinemodelconsidersaperfectlycompet- itivefirmoperatingunderconstantreturnstoscale.Weassumea constantelasticityofsubstitutionproductionfunctionwiththree perfect substitutableinputs(labour, capital,andhumancapital) andwiththreeknowledge-basedtechnologies(robotisation,inno- vation,anddigitisation)oftheform
Q =[(RN)(−1)/+(IN)(−1)/+(DN)(−1)/
+K(−1)/+H(−1)/]/(−1) (1) whereQisfirmoutput,andN,K,andHarefirmlabour,firmcapital, andfirmhumancapital,respectively.R(robotisation),I(innova- tion),andD(digitisation)arelabour-augmentingHarrod-neutral technologies,andistheelasticityofsubstitutionbetweenlabour, capital,andhumancapital.
We assume that robotisation, innovation, and digitisation mainlyresultinanincreaseinR,I,andD.AnincreaseinR,I,andD impliesthatthesameamountoflabourservices(RN),(IN),and(DN) requirelessinputfromlabour(N);assumingthat,intotallycom- petitiveenvironments,realwage
W/P
isequaltothemarginal productoflabour,andthefirst-orderconditionforlabourcanbe writtenas
logQ−logN= +log
WP
−(−1) logR−(−1) logI−(−1) logD (2)
Inthesamemanner,inperfectcompetitiveenvironments,we canassumethatthemarginalproductofcapitalequalsthecostof capital (C).Therefore,thefirst-orderconditionforcapitalcanbe writtenas
logQ−logK= logC (3)
Finally,andfollowinghumancapitaltheory,wecanassumethat themarginalproductofhumancapitalequalsthecostofemployee educationandtraining (T).Therefore,thefirst-orderconditionfor humancapitalcanbewrittenas
logQ−logH=logT (4)
Combiningthesethreeexpressions,wecanobtainouremploy- mentdemandfunction:
logN= −log
WP
+(−1) logR+(−1) logI+(−1) logD+logK+logC+logH+logT (5)
Alternatively,thelabourproductivityfunctioncanbeexpressed as
logQ−logN= +log
WP
−(−1) logR
−(−1) logI−(−1) logD (6)
where the second equation terms −(−1) logR,
−(−1) logI, and−(−1) logD, refer to the TFP based on technological changes (or TFP), and the first equation term log
W/P
referstolabourdeepening.Whentheelasticityofthe substitution betweenlabour, capital, and humancapital is low (<1)andforthegivenrealwages,labourproductivityincreases in R, I,and D.Consequently, employment decreasesas long as output (the given level of production) and real wages remain constant.Thedecreaseinemploymentoccursbecausetheincrease inR,I,andDimpliesthatlesslabourisnecessarytoachieveagiven leveloflabourservices(RN),(IN),and(DN)becausethelowdegree ofsubstitutionbetweenlabour,capital,andhumancapitalimplies asmallincreaseintheuseoflabourservices.Bycontrast,when theelasticityof substitution is high(>1), labourproductivity decreasesinR,I,andD(forgivenrealwages), andemployment increases(forgivenQ).Thereasonforthisisthatthedecreasein R,I,andDimpliesthatmorelabourisnecessarytoachieveagiven leveloflabourservices(RN),(IN),and(DN),andduetotheshift fromcapitalservicestolabourservices,morelabourisnecessary toachieveagivenleveloflabourservices.
WefollowEq.(5)andobservethatforagivenlevelofcapital andhumancapital,therealwagesandusercostofcapital,human capital,andemploymentincreasesinR,I,andDiftheelasticityof substitutionishigh(>1),butdecreasesiftheelasticityofsub- stitutionislow(<1).Therefore,theimplicationsofrobotisation, innovation,anddigitisationonemploymentareequaltothecaseof agivenlevelofproduction(Eq.6).Empiricalevidencesupportsthat thevalueofisbelow1inthecaseofrobotisation(León-Ledesma, McAdam,&Willman,2010)andabove1inthecaseofinnovation (Harrison,Jaumandreu,Mairesse,&Peters,2014).
Thus,wecanconcludethatforagivencapitalandhumancap- italstockandforagivenoutputstock(i.e.intheshortterm),the impactofrobotisation,innovation,anddigitisationonproductivity oremploymentdependsonthesizeoftheelasticityofthesubsti- tutionbetweenlabour,capital,andhumancapital.However,inthe longterm,outputandcapitalsareendogenous.Thus,itispossi- bletoexpectthatrobotisation,innovation,anddigitisationwould reducethemarginalcostsofproduction,whichcouldencourage investment,productivity,andoutput.Dependingontheelasticity ofdemand,thisimprovementineconomicactivitycouldincrease employmentinthelongterm.Hence,althoughrobotisation,inno- vation,anddigitisationtendtoreduceemploymentintheshort term,thistrendmayreverseinthelongterm(Acemoglu&Restrepo, 2019;Autor&Salomons,2018).Iftheincreaseinoutputissuffi- cientlyhigh,thenetlong-termeffectofrobotisation,innovation, anddigitisationonemploymentcouldbepositive.Inthissense,we proposeHypotheses1and2:
H1. Robotisation,innovation,anddigitisationincreaselabourpro- ductivityin the longterm (ie: non-givencapitals ornon-given outputstock).Thishypothesisonlyrequiresthattheelasticityof substitutionbetweenlabour,capital,andhumancapitalisbelow1.
H2. Robotisation,innovation,anddigitisationincreaseemploy- ment in the long run. This hypothesis suggests significant employmentcreationinthelongterm,whichwouldcompensate fortheshort-termreductioninjobs.
However,asinthefirst digitalwave where aclearlinkwas establishedbetweenintangibleassets,suchashumancapitaland workplaceinnovation,bothwithICTuses(Acosta,Sainz,&Salvador,
2006;Venturini,2015),weexpecttheuseofroboticstolinkwith thetypesoffirmknowledgeflows.Therefore,andextendingour model, we alsoaimto evaluatethecomplementarity effects of robotisation,innovation,anddigitisationonfirmproductivityand employment. Theconditionsof (long-term)flexibilityofoutput, capitals, real wages, and usercosts of capital and human cap- ital are established as in theprevious model. Additionally, we expectthatthecomplementarityeffectofrobotisation,innovation, and digitisationreducesthemarginal costsofproduction, rein- forcesproductivity,andincreasesthedemandofafirm’soutput.In thispositivesituation,thedisplacingeffectonemploymentinthe shorttermwouldbeclearlyacceleratedbytheincreasesinout- putandhumancapitaloverthelongterm(Brynjolfsson,Rock,&
Syverson,2018;Doraszelski&Jaumandreu,2018).Thus,wepro- poseHypotheses4and5:
H3. Thecomplementarityeffectamongrobotisation,innovation, anddigitisationstrengthensproductivityinthelongterm.
H4. Thecomplementarityeffectbetweenrobotisation,innova- tion,digitisation,andhumancapitalincreasesemploymentinthe longterm.
Estimationfunctionsandmethods
Weaimtoestimatetherelationshipbetweenrobotisation,inno- vation,digitisation,labourproductivity,andemploymentbyusing twotypesofmodels.Thefirstmodelestimatestheindividualeffects ofunitlabourcost,robotisation,innovation,anddigitisationonpro- ductivityandemployment(togetherwithcapitalperworkerand humancapital).Thesecondmodelestimatesthecomplementarity effectsofrobotisation,innovation,anddigitisationonproductiv- ity(togetherwiththeindividualeffectofunitlabourcost)andthe complementarityeffects ofrobotisation,innovation,digitisation, andhumancapitalonemployment(togetherwiththeindividual effectsofunitlabourcostandcapitalperworker).
Assuggestedinthemodelandthehypotheses,theproductivity andemploymenteffectsoftechnologydependonthe.Long-term effects areestimatedinloglevels.Firmdifferencesin loglevels reflectthedifferencesintheproductivityoremploymentexplana- toryvariablesoverthelongterm.Thestochasticformofequations ofproductivity(derivedfromEq.6)andemployment(derivedfrom Eq.5)inthelongtermare
qit−it= ˇ(w−p)it+˛Rit+Iit+Dit+se+si+uit (7)
it=−ˇ(w−p)it+kit+hit+˛Rit+Iit+Dit+se+si+εit
(8)
Lowercaselettersdenotelogs.Eqs(7)and(8)refertothemodels ofindividualeffects.seandsiaresectorandsizefirmdummies.
Thesedummiescontrolforunobservedheterogeneityinthemanu- facturingsectorandfirmsize.Rit,Iit,andDitaretheuseofindustrial robots,innovationactivities,anddigitisation,respectively,infirm iinperiodt. (w−p)itreferstothereallabourunitcost.Capital(kit) andhumancapital(hit)arealsocaptured.InEq.(5),theusesofthe costofcapital(logC)andtheusesofthecostofhumancapital(log T)affectlabourdemand.However,wheretherearedifferencesin theusercostofcapitalandhumancapitalacrossmanufacturing sectorsorfirmsize,thesefactorsarecapturedbythefixedeffects.
Thisphenomenonimpliesthattheirdifferencesareconstantover time.Additionally,uitanditarewhitenoiseterms.
TheEqs(9),(10),(11),(12),and(13)refertothemodelsofcom- plementarityeffects.Eqs(9)and(10)depicttheproductivityand employmentfunctionswithtwocomplementarityeffects.Eqs(11) and(12)depictproductivityandemploymentfunctionswiththe
three-complementarityeffects. Eq.(13)depicts an employment functionwithfour-complementarityeffects:
qit−it= ˇ(w−p)it+˛1RIit+˛2RDit+˛3IDit+se+si+uit(9)
it= −ˇ(w−p)it+kit+˛1RIit+˛2RDit+˛3RHit+˛4IDit +˛5IHit+˛6DHit++se+si+εit (10) qit−it= ˇ(w−p)it+˛RIDit+se+si+uit (11) it=−ˇ(w−p)it+kit+˛1RIDit+˛2RIHit
+˛3RDHit+˛4IDHit++se+si+εit (12)
it= −ˇ(w−p)it+kit+˛RIDHit+se+si+εit (13) Therearetwo-complementarity effects:RI (robotisation and innovation), RD(robotisation and digitisation), RH (robotisation andhumancapital),ID(innovationanddigitisation),IH(innova- tionandhumancapital),andDH(digitisationandhumancapital).
Thethree-complementarityeffectsareRID(robotisation,innova- tion,and digitisation),RIH(robotisation,innovation,andhuman capital),RDH(robotisation,digitisation,andhumancapital),and IDH (innovation, digitisation, and human capital). Finally, the four-complementarityeffectintheemploymentequationisRIDH (robotisation,innovation,digitisation,andhumancapital).
Theprovisionofannualseriesforanextendedperiodallows theestimationofexplanatoryfactorsforthemanufacturingfirms’
long-termproductivityand employment. Thus;thus,we elabo- ratethevariables’andindicators’arithmeticaverageoftheforthe twoestablishedestimationperiods.Thefirstperiodisfrom1991 to2016.Asaresultoftheprogressiveadaptationoftheinforma- tionsourcetothebusinesscontext,weuseasecondestimation periodfrom2000to2016thatincludestheindicatorsrelatedto themanufacturingfirms’digitisationprocess.
The estimation of the hypothesised functions is conducted usingordinaryleastsquares(OLS)withtheintroductionmethod.
OLS regression should be used only if somestandard require- mentsofthedataareachieved,suchasnormality,linearity,and homoscedasticity(Hairetal.,2010).Theskewnessandkurtosis- obtained values in all the estimated models suggest that the variablescanbeassumedtobenormallydistributed(belowthe thresholdof2.58).Multicollinearitydiagnoseshavebeenaddressed bytestingthetoleranceandvarianceinflationfactor(VIF)among theexplanatoryvariables.Becauseallthesevalues havethresh- oldtolerance=0.10andVIF=10.0,multicollinearitymaynotbea concernin ourregression models.Thecorrelationmatricesalso indicatetheabsenceofmulticollinearity.Finally,homoscedasticity isvisuallyexaminedandtestedinplotsofstandardisedresiduals againstthepredictedvalue(Durbin-Watsontest,1.5<DW<2.5).
Informationsource
Theinformationsourceused fortheanalysisistheESEE, an annualsurveyofaround1800Spanishmanufacturingfirmscon- ductedbytheSpanishGovernment’sMinistryofFinanceandPublic Administrationfrom1990to2016.ESSEprovidesdetailedinforma- tiononbusinessesintheareasofstrategicdecision-making(prices, costs,markets,andinvestment)andthevalueprocess(humancap- ital, organisation, innovation,R&D, and ICT use). In addition, it coversthemostimportantindicatorsandratiosfromfirms’balance sheetsandprofitandlossaccounts.Consequently,thispaneldata allowsadetailedstudyofthemicroeconomicsofproductivityand employment,andtheanalysisofchangesinSpanishmanufacturing firmsduringvariousstagesofthebusinesscycle(Torrent-Sellens, 2018).
TheESEEcontainssegmentedinformation formanufacturing firmswithmorethan200workers(largefirms)andfirmswith 10–200workers(SMEs).Asaresultofthedatacollection,theclassi- ficationofSMEsisdifferentthanthatoftheEuropeanCommission (EuropeanCommission,2012).InthecaseoftheESEE,thelimit usedtodefineanSMEis200employees,andtheEuropeanCom- missionusesamaximumsizeof250workers.Thisdifferenceis due tothe samplingprocedureused by theESEE, in which all largemanufacturingfirms(morethan200workers)areincluded inthesample.However,forSMEs(from10to200workers),strati- fied,proportional,andsystematicsamplingisusedbyindustries (nationaleconomic activitytwo-digitclassificationcode,NACE), andthesizeofthefirmisalsoused.Thesamplingexcludesmanu- facturingmicro-firms(i.e.firmswithlessthan10workers).
TheESEEalsoprovidesdetailed informationfor20manufac- turing branches of activity,which can beobserved in detail in AppendixA(dimension:TableA1;andindustries:TableA2)regard- ingthesampleoffirmsused.Theanalysisofthepaneldatasuggests agrowingpresenceofsmallerfirms(from63.3%ofSMEsin1991to 81.0%ofSMEin2016)andanotablereorientationoftheindustrial branchesofactivity.In1991,sixindustrieseachaccountedformore than7%ofthenumberoffirmsinthesample—textileandclothing (11.3%), foodand tobacco(10.4%),metalproducts(7.6%),chem- icalandpharmaceuticals(7.4%),non-metallicmineral(7.2%),and machineryandelectricalequipment(7.1%)—in2016,specialisation hadincreasedsignificantlyandonlythreeindustriesaccountedfor 7%ormoreofthetotalnumberoffirms:foodandclothing(13.5%), metalproducts(12.7%),andchemicalandpharmaceuticals(7.0%).
Variablesandindicators
Thedependentvariablesarelabourproductivityandemploy- mentinmanufacturingfirmsandareapproximatedbyusingthe logarithmofaddedvalueperhourworked(HPT)andthelogarithm ofthetotalstaffemployed(allcontracts)inthefirm(EMPL).
Tocapturetheuseofrobots(R),weuseadichotomousvariable thatis0whenthefirmdoesnotuserobots,and1whenafirmuses robots.Despitetheobviousrestrictionsoftheuseofadichotomous variable,itsincorporationintothepredictivemodelallowsusto assesstheimpactonfirmproductivityandemploymentwhena firmstartstouseindustrialrobotics.
Regardinginnovation,weusetwodichotomousvariablesthat take two values (value 1, innovation generation; value 0, no innovation):productinnovation(PRODI)whenthefirmhaspro- ducedtotallynewproductsorproductswithmodificationsthat aresorelevantthattheymakethemdifferentfromthoseprevi- ouslyproduced;andprocessinnovation(PROCI)whenthefirm hasintroducedsomeimportantmodificationintotheproduction and/ordistributionprocess.
Regarding digitisation, we built an additive indicator that reflects internet-based electronic commerce (D). This indicator takesfourvalues(0–3)andistheresultofthesumofthreedichoto- mous variables (value1,use; value0, nouse):purchasesfrom suppliersovertheinternet(PFSD),salestoendconsumersover theinternet(B2CD),andsalestofirmsovertheinternet(B2BD).
Realwagesareapproximatedbyusinganindicatoroflabour costsperworker(LCW),thecapitalstockofthefirmwasapproxi- matedbyusingthelogarithmoffinancialassetsperworker(KPTW), andthehumancapitalstockofthefirm(HKPTW)ismeasuredby usingthelogarithmofthepercentageofemployeeswithtertiary (university)education(bachelor’sdegreelevelandhigher).
Tocapturetheeffectofsectoral(se)andsize(si)dummies,we constructfouradditionalvariablesthatcapturethenon-observed heterogeneityinthemodels.Fromtheaveragevaluesofthepro- ductivity,exports,R&Dexpenditure,andproportionofemployees withauniversityeducation,weconstructfourdichotomousvari-
ablesthatassignthevalue1tothemanufacturingsectorswhen thevaluesofproductivity,exports,R&Dspending,anduniversity trainingofemployeesareabovetheaverage,andthevalue0in thealternativecase.Whenweobtainthesedichotomousvariables, wemultiplythembythefirmsizevariable,whichtakesvalue0for firmswith200employeesorlessonaverageinthereferenceperiod, andvalue1inthecaseoffirmswithmorethan200employeeson averageinthereferenceperiod.
Asaresultofthesecombinations,weobtainthefollowingfour variables:(1)LAREF(largeandefficientfirms)identifieslargefirms locatedinmanufacturingsectorswithabove-averageproductivi- ties;(2)LAREXP(largefirmswithexportintensity)identifieslarge firmslocatedinmanufacturingsectorswithabove-averageexport- ingvalues;(3)LARHC(largefirmswithintensityinhumancapital) identifies largefirms located in manufacturingsectors withan above-averagenumberofemployeeswithauniversityeducation;
and(4)SMER&D(SMEswithR&Dintensity)identifiessmalland medium-sizedfirms(SMEs)locatedinmanufacturingsectorswith above-averageR&Dexpenditure.
Allthevariablesandindicatorsexpressedinnominaltermshave beendeflatedbyusingaPaascheindexreferredtothevariation inpricesofintermediateconsumption.Thisindexhasbeenbuilt ontwogroupsofgoods:producergoodsandenergyandservices acquired.Becausewehavenorelativeweightsofproducergoods andenergy,weaddavariationofthesetwocomponentsbyusinga geometricmeanwithfixedweights.Thus,thepriceindexofinter- mediateconsumptiontakesthefollowingform:
PIINTCON(t)= VPGE(t)
VINTCON (t)PIPGE(t)+ VSER(t)
VINTCON (t)PISER(t) (14) where PIINTCON (t) is theprice index of intermediate consump- tion in period t (to be calculated); VPGE(t) is the value of the purchasesconsumedin period t;VINTCON(t) is thevalue ofthe intermediateconsumptioninperiodt;PIPGE(t) isthepricevari- ationofproducergoodsandenergybetweent-1andtobtained asPIPGE(t)= [(PIPG(t)]0,95x[(PIE(t)]0,5,wherePIPGandPIEarethe priceindicesofproducergoodsandenergyprovidedbythefirm;
VSER(t) isthevalueoftheservicesacquiredinperiodt;andPISER
isthepriceindexoftheservicesacquiredintheperiodt-1and t.AppendixB(TableB1)presentsthedescriptivestatisticsofthe variablesandindicatorsusedintheanalysis.
Results
Individualeffectsonproductivityandemploymentestimations Table1presentstheindividualeffectsofestimatingthelong- termtrendofproductivitylevel(worked-hourlyproductivity,HPT).
ConsistentwithEq.7,thefirstcolumn(Model1)analysestheeffects ofLCWonproductivity.Inthesecondcolumn(Model2),theeffects ofknowledge-basedtechnology(robotsandprocessinnovation) areincorporated.Inthethirdcolumn(Model3),sizeandindustry dummiesarealsointegrated.
Theuseofindustrialrobotsandprocessinnovationresultina significantincreaseinlabourproductivity.Theseindividualeffects arerobust(increasesinthechangeofadjustedR2 andmodelsp value=0.000)totheinclusionofLCWandsizeandindustrydum- miesasexplanatoryvariables,whichisevidentthroughcolumn analysis.Boththeunitlabourcostandthesizeandindustrydum- mieshavesignificantimpactsontheexpectedsigns.However,the unitlabourcostcoefficienttendstodecreaseasweincorporate morevariablesintotheregression(Models2and3comparedto Model1).Inthesamemanner,theroboticsuseandprocessinno- vationcoefficientstendtodecreasewhen thesizeand industry dummiesareincorporated(Model3comparedtoModel2).
Table1
Labourproductivity(addedvalueperhourworked,HPT),individualexplanatoryfactorsinSpanishmanufacturingfirmsin1991–2016and2000–2016.
1991–2016 2000–2016
HPT(addedvalueperhourworked) Model1 Model2 Model3 Model1 Model2 Model3
(Constant) −3.451*** −3.288*** −3.218*** −3.598*** −3.374*** −3.268***
(0.050) (0.051) (0.054) (0.071) (0.071) (0.076)
Labourcostperworker(LCW) 0.785*** 0.752*** 0.740*** 0.738*** 0.699*** 0.682***
(0.012) (0.012) (0.013) (0.016) (0.016) (0.017)
Useofrobots(R) 0.051*** 0.044*** 0.060*** 0.050***
(0.007) (0.007) (0.007) (0.007)
Processinnovation(PROCI) 0.090*** 0.089*** 0.119*** 0.117***
(0.007) (0.007) (0.008) (0.008)
Largeandefficientindustry(LAREF) 0.052** 0.049**
(0.013) (0.015)
LargeandHCintensiveindustry(LARHC) 0.055*** 0.066***
(0.01) (0.01)
SMEsintensiveinR&Dindustry(SMEsR&D) 0.086*** 0.081***
(0.012) (0.014)
Statistics
N(observations) 5,408 5,408 5,408 4,061 4,061 4,061
AdjustedR2 0.616 0.629 0.633 0.545 0.565 0.57
EstimationSE 0.177 0.174 0.173 0.174 0.17 0.169
ChangeofAdjustedR2 0.616 0.013 0.004 0.545 0.021 0.005
Fvalue 8,689 3,055 1,552 4,860 1,760 896.2
pvalue 0.000 0.000 0.000 0.000 0.000 0.000
Durbin-Watson 1.906 1.881
Note:Realmonetarydataareinloglevels.Regressionanalysis:Ordinaryleastsquareswiththeintroductionmethod.Estimatedcoefficients:Standardisedcoefficients.
Standarderrorsofthenon-standardisedeffectsareinbrackets.
***p<0.01;**p<0.05;*p<0.1.
Table2
EmploymentandindividualexplanatoryfactorsinSpanishmanufacturingfirmsin1991–2016and2000–2016.
1991–2016 2000–2016
Employment Model1 Model2 Model3 Model1 Model2 Model3
(Constant) −3.972*** −3.216*** −1.112*** −5.395*** −4.172*** −1.557***
(0.204) (0.187) (0.147) (0.255) (0.231) (0.194)
Labourcostperworker(LCW) 0.406*** 0.353*** 0.194*** 0.447*** 0.372*** 0.201***
(0.056) (0.051) (0.040) (0.066) (0.060) (0.050)
Capitalperworker(KPTW) 0.109*** 0.059*** 0.066*** 0.100*** 0.042** 0.047***
(0.018) (0.016) (0.012) (0.020) (0.018) (0.014)
Humancapitalperworker(HKPTW) −0.222*** −0.244*** −0.240*** −0.187*** −0.203*** −0.201***
(0.024) (0.022) (0.017) (0.027) (0.025) (0.020)
Useofrobots(R) 0.228*** 0.124*** 0.262*** 0.148***
(0.019) (0.015) (0.019) (0.016)
Processinnovation(PROCI) 0.103*** 0.078*** 0.105*** 0.083***
(0.023) (0.018) (0.024) (0.019)
Productinnovation(PRODI) 0.227*** 0.129*** 0.192*** 0.121***
(0.024) (0.019) (0.027) (0.021)
Digitisation(D) - - 0.058*** 0.048***
(0.013) (0.011)
Largeandefficientindustry(LAREF) 0.414*** 0.391***
(0.023) (0.027)
Largeandexportingindustry(LAREXP) 0.324*** 0.302***
(0.026) (0.03)
SMEsintensiveinR&Dindustry(SMEsR&D) 0.112*** 0.106***
(0.033) (0.037)
Statistics
N(observations) 4,480 4,480 4,480 3,502 3,502 3,502
AdjustedR2 0.199 0.354 0.624 0.213 0.377 0.605
EstimationSE 0.528 0.475 0.361 0.507 0.451 0.359
ChangeofAdjustedR2 0.199 0.155 0.271 0.213 0.165 0.228
Fvalue 371.1 409.3 827.9 316.8 303.9 536.8
pvalue 0.000 0.000 0.000 0.000 0.000 0.000
Durbin-Watson 1.817 1.881
Note:Realmonetarydataareinloglevels.Regressionanalysis:Ordinaryleastsquareswiththeintroductionmethod.Estimatedcoefficients:Standardisedcoefficients.
Standarderrorsofthenon-standardisedeffectsareinbrackets.
***p<0.01;**p<0.05;*p<0.1.
Thereductionintheunitlabourcostcoefficientisrelatedtothe typeofinvestmentandefficiencymodelofthetechnology-based firms.Robotisationandprocessinnovationimpliesasmallereffect ofthelabourcostintheproductivityexplanation,whichalready startssuggestingthatthereexistsalabour-share-displacingeffect.
Inthesamemanner,theintroductionofsizeandindustrydum- miesreflectstheheterogeneityofthefirm.Largefirmsinefficient andhumancapital-intensiveindustriesandSMEsinR&D-intensive industriestendtobemoreefficientbythemselves,reducingthe effectsoflabourcostandtechnologyonproductivity.
Table3
Labourproductivity(addedvalueperhourworked,HPT)andtwo-complementarityexplanatoryfactorsinSpanishmanufacturingfirmsin1991–2016and2000–2016.
1991–2016 2000–2016
2-complementarityfactors Model1 Model2 Model3 Model1 Model2 Model3
(Constant) −3.433*** −3.333*** −3.234*** −3.598*** −3.441*** −3.308***
(0.050) (0.051) (0.054) (0.069) (0.070) (0.075)
Labourcostperworker(LCW) 0.783*** 0.765*** 0.747*** 0.741*** 0.715*** 0.694***
(0.011) (0.012) (0.012) (0.015) (0.016) (0.017)
Robots(R)xProcessInnov.(PROCI) 0.077*** 0.067*** 0.075*** 0.064***
(0.009) (0.010) (0.011) (0.011)
Processinnov.(PROCI)xDigitisation(D) - - 0.049*** 0.046***
(0.008) (0.008)
Largeandefficientindustry(LAREF) 0.048** 0.042*
(0.013) (0.015)
LargeandHCintensiveindustry(LARHC) 0.070*** 0.078***
(0.010) (0.010)
SMEsintensiveinR&Dindustry(SMEsR&D) 0.083*** 0.075***
(0.013) (0.014)
Statistics
N(observations) 5,458 5,458 5,458 4,196 4,196 4,196
AdjustedR2 0.614 0.619 0.624 0.55 0.56 0.565
EstimationSE 0.177 0.176 0.174 0.175 0.173 0.172
ChangeofAdjustedR2 0.614 0.006 0.005 0.55 0.011 0.005
Fvalue 8,673 4,439 1,810 5,119 1,780 908.1
pvalue 0.000 0.000 0.000 0.000 0.000 0.000
Durbin-Watson 1.895 1.872
Note:Realmonetarydataareinloglevels.Regressionanalysis:Ordinaryleastsquareswiththeintroductionmethod.Estimatedcoefficients:Standardisedcoefficients.
Standarderrorsofthenon-standardisedeffectsareinbrackets.
***p<0.01;**p<0.05;*p<0.1.
Table4
Labourproductivity(addedvalueperhourworked,HPT)andthree-complementarityexplanatoryfactorsinSpanishmanufacturingfirmsin2000–2016.
3-complementarityfactors Model1 Model2 Model3
(Constant) −3.571*** −3.509*** −3.326***
(0.069) (0.069) (0.075)
Labourcostperworker(LCW) 0.738*** 0.728*** 0.698***
(0.015) (0.016) (0.017)
Robots(R)xProcessinnov.(PROCI)xDigitisation(D) 0.059*** 0.046***
(0.009) (0.010)
Largeandefficientindustry(LAREF) 0.055*
(0.015)
LargeandHCintensiveindustry(LARHC) 0.084***
(0.010)
SMEsintensiveinR&Dindustry(SMEsR&D) 0.076***
(0.014) Statistics
N(observations) 4,245 4,245 4,245
AdjustedR2 0.544 0.547 0.554
EstimationSE 0.177 0.176 0.174
ChangeofAdjustedR2 0.544 0.004 0.007
Fvalue 5,063 2,565 1,055
pvalue 0.000 0.000 0.000
Durbin-Watson 1.867
Note:Realmonetarydataareinloglevels.Regressionanalysis:Ordinaryleastsquareswiththeintroductionmethod.Estimatedcoefficients:Standardisedcoefficients.
Standarderrorsofthenon-standardisedeffectsareinbrackets.
***p<0.01;**p<0.05;*p<0.1.
Thecomparisonofresultsbetweenthetwoconstructeddata series:1991–2016and2000–2016capturethedifferentialeffects onproductivityfromthe2000s.For simplicity,wefocus onthe modelthatincorporatesalltheexplanatoryvariablesintotheanal- ysis(Model3).Theresultsobtainedconfirmthesignificanteffects fromrobotisationandprocess innovationonproductivity.From 1991to2016,themarginaleffectontheworked-hourlyproduc- tivitylevelofonemorerobotisedfirmis0.044percentagepoints, andanadditionalprocess-innovativefirmboostedworked-hourly productivityby0.089percentagepoints.Theseresultshaveaccel- eratedsince the2000s.From2000to2016,themarginal effect onworked-hourlyproductivitylevelofonemorerobotisedfirmis 0.050percentagepoints,andanadditionalprocess-innovativefirm boostedworked-hourlyproductivityby0.117percentagepoints.In
addition,thereisagreaterlabour-displacingeffect.Anadditional realeurooflabour costperemployee increasedworked-hourly productivityby0.682percentagepointsbetween2000and2016, comparedto0.740percentagepointsin1991–2106.Contraryto whatweexpected,digitisation(useofelectroniccommerce)had nosignificanteffectontheproductivityexplanation(neitherper workernorperhourworked).
Table 2 presents the individual effects from estimating the long-term trend of the employment level. Unlike productivity, industrialemploymenthasclearlyevolveddownwardsduringthe periodsanalysed.Therefore,apositivecoefficientimpliesapos- itive contribution tothe declining trend of employment (i.e. a positivecoefficientimpliesadecreaseinemployment),andaneg- ativecoefficientimpliesanegativecontributiontothedowntrend
Table5
Employmentandtwo-complementarityexplanatoryfactorsinSpanishmanufacturingfirmsin1991–2016and2000–2016.
1991–2016 2000–2016
2-complementarityfactors Model1 Model2 Model3 Model1 Model2 Model3
(Constant) −3.972*** −2.968*** −0.957*** −5.395*** −3.986*** −1.427***
(0.204) (0.188) (0.148) (0.255) (0.233) (0.194)
Labourcostperworker(LCW) 0.406*** 0.350*** 0.193*** 0.447*** 0.374*** 0.202***
(0.056) (0.051) (0.040) (0.066) (0.060) (0.050)
Capitalperworker(KPTW) 0.109*** 0.065*** 0.069*** 0.100*** 0.049** 0.051***
(0.018) (0.016) (0.012) (0.02) (0.018) (0.014)
Robots(R)xHumanCapital(HC) 0.241*** 0.129*** 0.272*** 0.151***
(0.018) (0.014) (0.017) (0.014)
Processinnovation(PROCI)xHumanCapital(HC) 0.112*** 0.084*** 0.111*** 0.089***
(0.021) (0.016) (0.021) (0.017)
Productinnovation(PRODI)xHumanCapital(HC) 0.223*** 0.127*** 0.187*** 0.118***
(0.021) (0.016) (0.023) (0.018)
Digitisation(D)xHumancapital(HC) 0.067*** 0.054***
(0.012) (0.009)
Largeandefficientindustry(LAREF) 0.415*** 0.393***
(0.023) (0.027)
Largeandexportingindustry(LAR EXP) 0.328*** 0.311***
(0.026) (0.03)
SMEsintensiveinR&Dindustry(SMEsR&D) 0.116*** 0.114***
(0.033) (0.037)
Statistics
N(observations) 4,480 4,480 4,480 3,502 3,502 3,502
AdjustedR2 0.199 0.348 0.622 0.213 0.374 0.604
EstimationSE 0.528 0.476 0.362 0.507 0.453 0.36
ChangeofAdjustedR2 0.199 0.15 0.274 0.214 0.16 0.23
Fvalue 371.1 400.2 821.3 316.8 297.9 532.7
pvalue 0.000 0.000 0.000 0.000 0.000 0.000
Durbin-Watson 1.722 1.786
Note:Realmonetarydataareinloglevels.Regressionanalysis:Ordinaryleastsquareswiththeintroductionmethod.Estimatedcoefficients:Standardisedcoefficients.
Standarderrorsofthenon-standardisedeffectsareinbrackets.
***p<0.01;**p<0.05;*p<0.1.
inemployment(i.e.anegativecoefficientimpliesanincrease in employment).ConsistentwithEq.8,thefirstcolumn(Model1) analysestheeffectsofLCW,capitalperworker (financialassets per employee), and human capital per worker (percentage of employeeswithtertiaryeducation)onemployment.In thesec- ondcolumn(Model2),theeffectsofknowledge-basedtechnology (robotisation,innovation—productandprocess—anddigitisation) areincorporated.Inthethirdcolumn(Model3),sizeandindustry dummiesarealsointegrated.
Theuseofindustrialrobots,innovation,anddigitisationsignif- icantlydecreasestheemploymentlevel(positivecontributionto thedowntrendinemployment).Theseindividualeffectsarerobust (increasesinthechangeofadjustedR2andmodelspvalue=0.000) to theinclusionof LCW,capital perworker, humancapital per worker,andsizeandindustrydummiesasexplanatoryvariables.
Boththeunitlabourcostandhumancapitalperworkerhavesignif- icantimpactsandwiththeexpectedsigns:aslabourcostincreases, employmentdecreases;andtertiaryeducationincreases,employ- mentincreases.However,thecapitalperworkerandthedummies donotbehaveasweexpected:asthecapitalperworkerincreases, theemploymentdecreases.Largefirmsinefficientandexporting industries andSMEsinR&D-intensiveindustriestendtobeless labour-intensive.
WecompareModel1withModel2,andrelevantconsiderations areobtained.Intheexplanationoftheemploymentreduction,the coefficientsofLCWandofcapitalperworkerevolvedownwards whenthetechnologicalvariablesareincorporatedintotheanalysis.
Investmentanduseoftechnologydisplacethelabourandcapital returns,whileanincreaseinthehumancapitalindicatesgreater educationalneedsforamoreappropriateuseoftechnology.The comparisonofModels2and3alsoprovidesrelevantconclusions.
Theintroductionofsizeandindustrydummiesreflectsfirmhet- erogeneity.Largefirmsinefficientandexportingindustriesand
SMEsinR&D-intensiveindustriestendtobelesslabour-intensive bythemselves,reducingtheeffectsoflabourcost,humancapital, andtechnologyonemployment.
The comparison of the results obtained for the 1991–2016 and 2000–2016 intervals allows us to evaluate the explana- tory factors of the observed destruction of employment and, more particularly, to analyse whether they have been accen- tuated since the2000s. Regardingthe technologicaldimension (Model 3), the results obtained confirm the significant and negative effects of robotisation, process, and product inno- vation and digitisation on employment. From 1991 to 2016, having one more robotised firm reduced employment levels (increased negative employment trend) by 0.124 percentage points,anadditionalprocess-innovativefirmdecreasedemploy- ment by 0.078 percentage points, and an additional product- innovative firm decreased employment by 0.129 percentage points.
Theseresultshaveacceleratedsincethe2000s.From2000to 2016,havingonemorerobotisedfirmreducedemploymentlevels by0.148percentagepoints,andanadditionalprocess-innovative firmlessened employment by0.083percentage points.Bycon- trast,thecontributionof productinnovationfell slightly (0.121 percentage points). Additionally, and in the 2000–2016 inter- val,digitisation also reduced employment by 0.048 percentage points. There are also greater labour-displacing and human- capital-displacing effects, and a lower capital-displacing effect.
Anadditional realeuro of labour cost per employee decreased employmentby0.201percentagepointsbetween2000and2016, comparedto0.194percentagepointsbetween1991and2106.An additionalemployee withtertiary educationincreased employ- ment by 0.201 percentage points in the 2000–2016 interval, comparedto0.240percentagepointsinthe1991–2016interval.An additionalrealeuroofcapitalperworkerdecreasedemployment
by0.047percentagepointsinthe2000–2016interval,compared to0.066percentagepointsinthe1991–2016interval.
Complementarityeffectsonproductivityandemployment estimations
Thecomplementarityeffectsobtainedfromestimatingthelong- termtrendoftheproductivitylevel(worked-hourlyproductivity, HPT)arepresentedinTable3(two-complementarities,Eq.9)and Table 4(three-complementarities, Eq. 11).Asin theestimation oftheindividualeffects,thefirstcolumn(Model1)analysesthe effectsofLCWonproductivity.Inthesecondcolumn(Model2),the effectsoftechnologycomplementarities(robots,processinnova- tion,anddigitisation)areincorporated.Inthethirdcolumn(Model 3),sizeandindustrydummiesarealsointegrated.
Thecomplementaritiesbetweenindustrialrobotsandprocess innovationandbetweenprocessinnovationanddigitisationresult inasignificantincreaseinlabourproductivity.Thesecomplemen- tarity effectsare robust(increasesin thechangeof adjustedR2 andmodels’pvalue=0.000)totheinclusionofLCWandsizeand industrydummiesasexplanatoryvariables.Boththeunitlabour costandthesizeandindustrydummieshavesignificantimpacts andtheexpectedsigns.Asinthecaseofindividualeffects,thetwo regressions’modelcomparisonsuggestsalabour-share-displacing effectandreflectsfirmheterogeneity.Theunitlabourcostcoef- ficientstendtobereducedwhenincorporatingtechnology-based complementaritiesand sizeand industrydummies(fromModel 1 to Models 2 and 3). In thesame manner, thecoefficients of technologicalcomplementaritiesarereducedbyintroducingthe dummieseffect(from Model2toModel 3).In thissense,tech- nologicalcomplementaritiesalsointroducechangesintothefirm efficiencymodels,withlowercontributionsfromthelabourfactor.
Additionally,thelocationofthefirminlarge,efficient,andhuman- capital-intensiveindustriesorinR&D-intensiveSMEsalsoreduces thecontributionofthelabourfactorandtechnologycomplemen- tarities.
Thecomparisonofthecoefficientsobtainedforthetwocon- structeddataintervals(1991–2016and2000–2016)suggeststwo notableresults.First,wecanconfirmthesignificanteffectsoftwo technologicalcomplementarities(betweenrobotisationandpro- cessinnovation,andbetweenprocessinnovationanddigitisation) andthreetechnologicalcomplementarities(betweenrobotisation, processinnovation,anddigitisation)onproductivity.From1991 to2016,havingonemorerobotisedandprocess-innovativefirm increasedworked-hourlyproductivitylevelby0.067percentage points. However,thiscomplementarity hasweakenedsince the 2000s.From2000to2016,havingonemorerobotisedandprocess- innovativefirmincreasedtheworked-hourlyproductivitylevelby 0.064percentagepoints.
Bycontrast,thecomplementaritybetweenprocessinnovation anddigitisationhasalsocontributedtotheadvanceinproductiv- ity.From2000to2016,havingonemoreprocess-innovativeand digitisedfirmincreasedworked-hourlyproductivitylevelby0.046 percentage points.However,theintroduction ofrobotisationto thetwo-complementarityeffectbetweenprocessinnovationand digitisationdidnotimprovethecontributiontotheproductivity advance.Thethree-complementarityeffect(robotisation,process innovationanddigitisation)onproductivitywas0.046percentage pointsinthe2000–2016interval(thesamepercentageasinthe 1991–2016interval).
Thecomparisonbetweentheindividualandcomplementarity effects in theexplanationofproductivitysuggestsweakresults intermsofthecombinationoftechnologicalfactors.Considering two-complementaritiesand the2000–2016interval,theoverall individualeffectofrobotisationandprocessinnovationonproduc- tivityperhourworkedwas0.167percentagepoints,clearlyabove
thecomplementarityeffect(0.064).Likewise,thejointindividual effectofprocessinnovationanddigitisationwas0.166percentage points;again,wellabovethecomplementarityeffect(0.046).Addi- tionally,consideringthree-complementarities,theresultsobtained arestillmoredifferentiated.Theoverallindividualeffectofrobo- tisation, process innovation, and digitisation on worked-hourly productivitywas0.216percentagepointsinthe2000–2016inter- val,andthecomplementarityeffectwasmuchlower(0.046).
The complementarity effects from estimating the long-term trend of employment level are presented in Table 5 (two- complementarities,Eq.10),Table6(three-complementarities,Eq.
12),andTable7(fourcomplementarities,Eq.13).Asfortheindivid- ualeffectsofemploymentestimation,thefirstcolumn(Model1) analysestheeffectsofLCWandcapitalperworkeronemployment.
Inthesecondcolumn(Model2),theeffectsofcombinedtechnology (robotisation,innovation–productandprocess–anddigitisation) andhumancapitalareincorporated.Inthethirdcolumn(Model3), sizeandindustrydummiesarealsointegrated.
Thecomplementarity effects of theuseof industrial robots, innovation,digitisation,andhumancapitalsignificantlydecrease theemploymentlevel(positivecontributiontothedowntrendin employment).Theseindividualeffectsarerobust(increasesinthe changeof adjustedR2 andmodelspvalue=0.000)totheinclu- sionofLCW,capitalperworker,andsizeandindustrydummies asexplanatoryvariables.In theexplanationof theemployment reduction,thecoefficientsofLCWandofcapitalperworkerevolve downwardswhen thetechnologicalandhumancapital comple- mentaritiesareincorporatedintotheanalysis(fromModel1to Model 2). Technological and human capital complementarities seemtodisplacelabourandcapitalreturns.Similarly,theintroduc- tionofthesizeandindustrydummiesreflectsfirmheterogeneity.
LargefirmsinefficientandexportingindustriesandSMEsinR&D- intensiveindustriestendtobelesslabour-intensivebythemselves, reducingtheeffectsoflabourcostandtechnologicalandhuman capitalcomplementaritiesonemployment.
Regardingtwo-complementarityeffects (Model3inTable8), theresultsobtainedconfirmthesignificantandnegativeeffects ofrobotisationandhumancapital,processinnovationandhuman capital, productinnovation and humancapital, and digitisation and human capital onemployment. From 1991to 2016, more robotisedandhumancapital-intensive(employees withtertiary education)firmsreducedemploymentlevels(increasednegative employment trend) by 0.129 percentage points, an additional process-innovativeand human capital-intensivefirm decreased employment by 0.084 percentage points, and an additional product-innovativeand humancapital-intensive firm decreased employmentby0.127percentagepoints.
These results have accelerated from 2000 to 2016. Having one more robotised and human capital-intensive firm reduced employmentlevelsby0.151percentagepoints,andanadditional process-innovativeand human capital-intensivefirm decreased employment by 0.089percentage points. By contrast, thecon- tribution of product innovation and human capital decreased (0.118 percentage points). Additionally, and in the 2000–2016 interval, two-complementarity effects between digitisation and human capital also reduced employment by 0.054 percentage points.There wasalso a greater labour-displacing and a lower capital-displacingeffect. An additional real euro of labour cost peremployeedecreasedemploymentby0.202percentagepoints between2000and2016,comparedto0.193percentagepointsin 1991–2016.Anadditionalrealeuroofcapitalperworkerdecreased employmentby0.051percentagepointsinthe2000–2016interval, comparedto0.069percentagepointsinthe1991–2016interval.
The results related to the three-complementarity and four- complementarity effects (Model 3 in Tables 6 and 7) and are in line with those obtained in the two-complementarity anal-