No2024-08–July WorkingPaper ThePercolationofKnowledge acrossSpace PierreCotterlaz&ArthurGuillouzouic Highlights Weshowthatthenegativeeffectofgeographicaldistanceonknowledgeflowsstemsfromhowfirmsgain sourcesofknowledgethroughtheirexistingnetwork. Inaggregate,thedistanceelasticityofpatentcitationsflowshasremainedconstantsincethe1980s,despite theriseoftheinternet. Atthemicrolevel,firmsdisproportionatelyciteexistingknowledgesources,andpatentscitedbytheir sources. Weintroduceamodelfeaturingthelatterphenomenon,andgeneratinganegativedistanceelasticityin aggregate. Whilethedistanceeffectshouldhavedecreasedwithconstantcountrycomposition,theriseofEastAsianeconomies,associatedtolargedistanceelasticities,compensatedlowerfrictionsinothercountries. CEPIIWorkingPaperThePercolationofKnowledgeacrossSpace Abstract Thispapershowsthatthenegativeeffectofgeographicaldistanceonknowledgeflowsstemsfromhowfirmsgainsourcesofknowledgethroughtheirexistingnetwork.Westartbydocumentingtwostylizedfacts.First,inaggregate,thedistanceelasticityofpatentcitationsflowshasremainedconstantsincethe1980s,despitetheriseoftheinternet.Second,atthemicrolevel,firmsdisproportionatelyciteexistingknowledgesources,andpatentscitedbytheirsources.Weintroduceaframeworkfeaturingthelatterphenomenon,andgeneratinganegativedistanceelasticityinaggregate.ThemodelpredictsPareto-distributedinnovatorsizes,andcitationdistancesincreasingwithinnovatorsize.Thesepredictionsholdwellempirically.Weinvestigatechangesoftheunderlyingparametersandgeographicalcompositioneffectsovertheperiod.Whilethedistanceeffectshouldhavedecreasedwithconstantcountrycomposition,theriseofEastAsianeconomies,associatedtolargedistanceelasticities,compensatedlowerfrictionsinothercountries. Keywords KnowledgeDiffusion,InnovationNetworks,SpatialFrictions,PatentCitation. JEL L14,O33,R12. WorkingPaper ©CEPII,PARIS,2024 Centred’étudesprospectives etd’informationsinternationales20,avenuedeSégur TSA10726 75334ParisCedex07 contact@cepii.fr www.cepii.fr–@CEPII_Paris Presscontact:presse@cepii.fr CEPIIWorkingPaper Contributingtoresearchininternationaleconomics CEPII(Centred’ÉtudesProspectivesetd’InformationsInternationales)isaFrenchinstitutededicatedtoproducingindependent,policy-orientedeconomicresearchhelpfultounderstandtheinternationaleconomicenvironmentandchallengesintheareasoftradepolicy,competitiveness,macroeconomics,internationalfinanceandgrowth. EDITORIALDIRECTOR: ANTOINEBOUËT VISUALDESIGNANDPRODUCTION:LAUREBOIVIN ISSN2970-491X July2024 Tosubscribeto TheCEPIINewsletter: www.cepii.fr/KeepInformed Allrightsreserved.Opinionsexpressedinthispublicationarethoseoftheauthor(s)alone. 1Introduction Whileinnovationandknowledgespilloversareakeyengineofeconomicgrowth(AghionandJaravel2015),thesespilloversseemtohaveweakenedintherecentpast,andideasseemeverhardertofind(AkcigitandAtes2021;Bloometal.2020).Althoughthesetrendsmaypartlybeduetofirmsactivelytryingtoavoidknowledgediffusioninordertoprotecttheircompet-itiveadvantage,themainmacro-levelimpedimenttoknowledgedisseminationremainsthemeregeographicaldistancebetweeninnovators.Thus,itiscrucialtounderstandtheforcesthatunderlietheimperfectdisseminationofknowledge,bothbetweeninnovatorsandacrossspace,andhowthesehavebeenaffectedbythetremendousprogressofinformationsearchtechnologiesachievedinthepastfourdecades. Thispaperstudiesthenetworkformationunderpinningthenegativeimpactofgeograph-icaldistanceonknowledgeflows.Wefirstdocumenttwocriticalstylizedfacts.Atthemacrolevel,theelasticityofpatentcitationflowswithrespecttodistancehasremainedremarkablystablesincethe1980s,despitetheriseoftheinternet.Atthemicrolevel,weobserveapatternofknowledgepercolationwithinnetworks,withinnovatorsdisproportionatelycitingsourcesoftheirownsources.Wethendevelopamodelthatbridgestheseobservations,incorporatinganetworkformationprocessamonginnovators,andgeneratinganaggregateeffectofdistance.Themodeldeliverstwopredictions:thesizeofinnovatorsshouldbePareto-distributed,andasystematicrelationshouldlinkfirms’sizeandthedistanceoftheircitations.Meetingthesetwopredictionsinturngeneratesanegativedistanceelasticityofknowledgeflowswithrespecttogeographicaldistance,ourinitialstylizedfact.Thedatafitthesepredictionswell,andchangesinthepredictedparameterscorrelatewithchangesinthedistanceelasticity.Toinvestigatedriversofthedistanceeffectdynamics,wedecomposeitalongtwodimensions:technologiesandcountries.Wefindthat,whilechangesinthetechnologicalcompositionofinnovationovertheperiodhadnoimpactontheoveralldistanceelasticity,thegeographicalcompositionofinnovatorsmatteredconsiderably.Theeffectofdistanceshouldhavebecomesmaller,buttheriseofChinaandKorea,withpronounceddistanceeffects,eventuallyoffsetthesegains.Thissuggestsdifferentpatternsofknowledgeacquisitioninthesecountries,whichwehypothesizetobepartlyduetolanguagebarriers. Westartbydocumentingthat,overalmost40years(1980–2017),theelasticityofcitationflowswithrespecttodistancehasremainedremarkablystable.Asisstandardforthestudyoftradeflows,weadoptagravityequationframework(Head