EUROPEANCENTRALBANK

EUROSYSTEM

RalphDeHaas,MikhailMamonov,AlexanderPopov,IlirianaShala

WorkingPaperSeries

Violentconflictandcross-borderlending

No3073

Disclaimer:ThispapershouldnotbereportedasrepresentingtheviewsoftheEuropeanCentralBank(ECB).TheviewsexpressedarethoseoftheauthorsanddonotnecessarilyreflectthoseoftheECB.

ECBWorkingPaperSeriesNo30731

Abstract

Howdoviolentconflictsshapecross-borderlending?Usingdataonsyndicatedloansby14,021creditorstofirmsin179countries(1989—2020),wedocumentadualeffect:foreignbanksreduceoveralllendingrelativetodomesticbanksbutsignificantlyincreasefinancingtomilitaryanddual-usesectorsduringconflicts.Thisreallocationisstrongeramonglenderslessspecializedintheconflictcountry,morespecializedinmilitarylending,anddomiciledinpoliticallynon-alignednations.Effectsaregeographicallycontainedandtemporallylimited,dissipatingpost-conflict.Ourfindingsrevealhowglobalbanksstrategicallyredirectcredittowardmilitarysectorsduringarmedconflicts,despitereducingoverallcountryexposure.

JELclassification:D74,F34,F40,G15,G21

Keywords:Cross-borderlending,syndicatedloans,violentconflict,geopoliticalrisk,bankspecialization

ECBWorkingPaperSeriesNo30732

Non-technicalsummary

Overthelongspanofhumanhistory,peacehasbeentheanomaly,withconflictastheconstant.Whileeconomistshavethoroughlyexaminedthedirectandindirecteconomiccon-sequencesofwar,includingtherolethatsovereignborrowingplaysinmaintainingmilitaryeffort,therelationshipbetweenprivatefinanceandarmedconflictremainsunderexplored.Weexaminethisrelationshipthroughthelensofcross-borderlendingduringviolentconflicts.

Ouranalysisisguidedbytwoopposinghypotheses.Ontheonehand,theextantliter-atureshowsthatcross-borderlenderstendto“runfortheexit”whenfacedwithnegativeshockstothelocaleconomy,suchassystemicbankingcrises.Thisliteraturethussuggeststhatcross-borderlendingshoulddeclinewhencountriesexperienceviolentconflict.Con-versely,armedconflictmaygeneratecountervailingforcesduetorisingdemandforcredit,particularlyindefence-relatedsectors.Foreignbanks,lessdirectlyimpactedbylocalhostil-ities,maybebetterpositionedtoaccommodatethisdemandcomparedtodomesticbanksfacingimmediateconflict-relatedconstraints,andcanthusemergeaspivotalfinanciersofmilitaryproductioninconflictzones.

Tostudythisquestion,weleanonthreesourcesofdata.First,weobtaincomprehensivesyndicatedloandatacovering1.3millionloansby14,021lendersto97,169firmsacross179countriesduring1989—2020.Next,weextractinformationfromtheUppsalaConflictDataProgram,whichprovidesdetailedandcompleteinformationonarmedconflicts,includingbattlefielddeathcounts.Finally,weemployanAI-basedapproachtoidentifymilitary-relatedborrowersbydistinguishingprimarymilitarysectors(exclusivelyproducingdefencegoods)anddual-usesectors(civiliangoodswithmilitaryapplications).

Ourheadlineresultisthatrelativetodomesticbanks,foreignbanksreducesignificantlylendingtonon-militaryfirmsduringconflicts,whilesimultaneouslyincreasingsignificantlymilitarysectorlending.Themilitarylendingincreasestemsprimarilyfromheightenedde-mandbydefencefirmsthatforeignbanksarebetterpositionedtomeet,whilethebroaderreductioninforeignbankexposuretoconflictcountriesappearstobeaclassicsupply-driven

ECBWorkingPaperSeriesNo30733

“flighthome”effect.

Furtheranalysisdocumentstheimportantrolethatcountryandbankheterogeneityplayintheeffectwedocument.First,whenaccountingforgeopoliticalalignment,wefindthatwhilebanksfromallorientationsincreasemilitarylendingduringconflicts,banksprefer-entiallydirectmilitaryfinancingtowardnon-alignedorpoliticallydistantcountrieswhileavoidinggeopoliticallysimilarnations.Thissuggeststhatprofitmotivesdominatewhenlendingto“out-group”countries,whilepoliticalconstraintsorregulatoryscrutinylimitmilitaryfinancingwithingeopoliticalblocs.Second,thebroadretreatfromnon-militarylendingtoconflictzonesissystematicallyoffsetbyincreasedmilitarylendingfromtwodis-tinctbanktypes:bankswithoutpriorcountryspecializationaggressivelyredirectcapitaltowardmilitarysectorsinunfamiliarconflictzones,whilebankswithestablishedmilitarysectorexpertisealsosharplyincreasedefencelendingduringconflicts.Thisindicatesthatcross-borderlenderswithmilitaryexpertisebutlimitedcountry-specificrelationshipsarerelativelywellpositionedtocapitalizeonconflict-inducedcreditdemandwithouthavingtocompromiseestablishedcountry-specificclientrelationships.Finally,bothstate-ownedandprivatebanks,aswellasbankandnon-banklenders,displaysimilarreallocationpatterns,indicatingthatpoliticalownershipdoesnotdriveourresults.

Inall,ouranalysisprovidesafirstglimpseintohowglobalbanksstrategicallyredirectcredittowardmilitarysectorsduringarmedconflicts,despitereducingoverallcountryexpo-sure.

ECBWorkingPaperSeriesNo30734

1Introduction

AlthoughtheworldhasenjoyedarelativelypeacefulperiodsincethecarnageofWorldWarII(

Pinker

,

2011

),peacehasbeentheexceptionratherthantherulethroughoutmuchofhumanhistory.Russia’swaronUkraine,escalatingtensionsintheMiddleEast,andprotractedcivilwarsinMyanmar,SudanandYemenserveassoberingremindersofthisrealityandhavethrustgeopoliticalconflictbacktothefore.

Whileeconomistshavethoroughlyexaminedthedirectandindirecteconomicconse-quencesofwar(

BarroandLee

,

1994

;

DavisandWeinstein

,

2002

;

AbadieandGardeazabal

,

2003

;

Acemoglu,JohnsonandRobinson

,

2005

;

Poast

,

2005

;

Tooze

,

2006

;

GlickandTaylor

,

2010

)andhowstatesleveragesovereigndebttosupporttheirmilitaryendeavors(

Kremerand

Jayachandran

,

2006

;

ReinhartandRogoff

,

2009

;

Zielinski

,

2016

),therelationshipbetweenprivatefinanceandarmedconflictremainsunderexplored.Weexaminethisrelationshipthroughthelensofcross-borderlendingduringviolentconflicts.

Twoopposinghypothesesguideourempiricalanalysis.Ontheonehand,existingliter-atureshowsthatcross-borderlenderstendto“runfortheexit”whenfacedwithnegativeshockstothelocaleconomy,suchassystemicbankingcrises.Thisholdsespeciallyintheab-senceofstrongrelationshipsbetweencreditorsandborrowers(

GiannettiandLaeven

,

2012

;

DeHaasandVanHoren

,

2013

).Historicalandcontemporaryevidencealsoindicatesthatbanksaretypicallywaryofwar’sdestabilizingeconomiceffects(

Kirshner

,

2007

),particularlywhenaconflictseriouslydamagescorporateassetsanddiminishesfirms’abilitytopledgecollateral(

Shpak,Earle,GehlbachandPanga

,

2023

).Thisliteraturethussuggeststhatcross-borderlendingshoulddeclinewhencountriesexperienceviolentconflict.

Conversely,armedconflictmaygeneratecountervailingforcesthatincreasedemandforcross-bordercredit,particularlyindefense-relatedsectors.Foreignbanks,lessdirectlyim-pactedbylocalhostilities,maybebetterpositionedtoaccommodatethisdemandcomparedtodomesticbanksfacingimmediateconflict-relatedconstraints(suchasphysicaldamagetobankinginfrastructureandliquiditypressuresfromdepositwithdrawals).Cross-border

ECBWorkingPaperSeriesNo30735

lenderscanthenemergeaspivotalfinanciersofmilitaryproductioninconflictzones.Anec-dotalevidencesupportsthisidea.AnotoriouscaseinvolvestheItalianBancaNazionaledelLavoro,whichuseditsUSbranchtogrant$3billioninunauthorizedcreditstoIraq(1988-1989),withabout$600millionfundingmilitarytechnology(

CIA

,

1989

).Morebroadly,estimatesindicatethatduring2020—2022alone,financialinstitutionsprovided$1trilliontotheglobaldefenseindustry(

Longo,MeggiolaroandFelipe

,

2024

),withEurope’s15largestbankslending€88billiontoarmscompaniessellingtoconflictzones(

Oudes,Slijperand

Uiterwaal

,

2022

).

Ouraimistomovebeyondhistoricalandanecdotalevidencebysystematicallyanalyzinghowforeigncreditflowsrespondtoviolentconflicts.Todoso,weleveragecomprehensivesyndicatedloandatafromDealScan,covering1.3millionloansby14,021lendersto97,169firmsacross179countriesduring1989—2020.Cross-bordercreditisakeycomponentofglobalcapitalflows,andalmostthree-quartersofallcross-bordercredittobothdevelopedandemergingcountriescomesintheformofsyndicatedloans(

DoerrandSchaz

,

2021

).

WemergethisinformationwithdatafromtheUppsalaConflictDataProgram(UCDP),whichprovidesdetailedandcompleteinformationonarmedconflicts,includingbattlefielddeathcounts.Inthethreedecadesstudied,civilwarsandotherintrastatehostilitiescomprisethemajorityofviolentconflicts.Figure

1

showsthatinourcombinedDealScan—UCDPdataset,25countriesexperiencedatleastoneyearwithmorethan500battlefielddeaths,and16countriessawatleastoneyearexceeding1,000battlefielddeaths.

Next,wesystematicallyidentifymilitary-relatedborrowersbydistinguishingprimarymilitarysectors(exclusivelyproducingdefensegoods)anddual-usesectors(civiliangoodswithmilitaryapplications).DrawingontheUKStrategicExportControlList,wemaprelevantkeywordsto4-digitSICcodesandemployanAI-basedapproachtoassessmilitaryrelevance.Weretain10primarymilitaryplus79dual-usesectorswithatleast50%averagemilitary-useprobability,togetherrepresenting17%ofoursyndicatedlendingsample.

Drawingonthisnewdataset,westartwithanaggregate-levelanalysisthatcompares

ECBWorkingPaperSeriesNo30736

Figure1.ConflictCountriesbyAnnualBattlefieldDeaths

Note:Thisfigureshowscountrieswhereannualbattle-fieldrelateddeathsexceeded250,500,or1,000atleastonceduring1989-2020andwhereatleastonefirmreceivedasyndicatedloanduringthisperiod.ThenatureandtimingofeachconflictisdescribedinAppendixTable

A.I

.Datasources:UppsalaConflictDataProgramandDealScan.

lendingpatternsbyforeignversusdomesticbankgroupstomilitaryversusnon-militarysectorsduringconflicts.Weestablishtwomainresults.First,foreignbanksreduceoveralllendingtoconflictcountriesrelativetodomesticbanks—consistentwitha“flighthome”effect.Second,incrisistimes,cross-borderlendersincreasetheirlendingtomilitary-relatedsectors.Inshort,weshowthatviolentconflictstriggerbothacontractioninoverallcreditprovisionandareallocationofcross-borderlendingfromnon-militarytodefense-relatedsectors.Thisdualpatternalsoemergesclearlyineventstudiesaroundtheonsetofconflict.

Next,weconductaloan-levelanalysistoexamineindividualbank-firmlendingdecisionsduringconflicts.Usingcomprehensivefixedeffects,weconfirmtheaggregatepatterns:for-eignbanksreducelendingtonon-militaryfirmsby27%duringconflictsrelativetodomesticbanks,whilesimultaneouslyincreasingmilitarysectorlendingby24%morethandomesticbanks.Thesebaselinefindingsproverobusttovaryingconflictintensitythresholds,alter-nativesectorclassifications,differenteconometricspecifications,alternativetreatmentsof

ECBWorkingPaperSeriesNo30737

missingdata,andrestrictionstospecificcountryorlendersubsamples.

Wethenturntosupplyanddemanddecompositionteststoisolatetheunderlyingmech-anisms.Bysaturatingourspecificationswitheitherbank-yearorcountry-sector-yearfixedeffects,weseparatesupplyanddemandforces.Theresultsrevealthatthemilitarylend-ingincreasestemsprimarilyfromheighteneddemandbydefensefirmsthatforeignbanksarebetterpositionedtomeet,ratherthanproactivesupply-sidetargetingbylenders.Mean-while,thebroaderreductioninforeignbankexposuretoconflictcountriespersistsevenaftercontrollingfordemandfactors,consistentwithaclassicsupply-driven“flighthome”effect.

Inanextstep,weanalyzewhetherbanksalignlendingpracticeswiththeirhomecountry’sgeopoliticalinterestsduringconflicts.Weclassifycountriesusingformalgeopoliticalblocs(BRICSvs.NATO/G7)andUNGeneralAssemblyvotingpatternstodistinguishWestern,Eastern,andneutralalignments.Whilebanksfromallorientationsincreasemilitarylendingduringconflicts,thedestinationpatternsrevealstrategicconsiderations:bankspreferentiallydirectmilitaryfinancingtowardnon-alignedorpoliticallydistantcountrieswhileavoidinggeopoliticallysimilarnations.Westernbanks,forinstance,significantlyincreasemilitarylendingtoEasternandneutralconflictcountriesbutreduceittoWesternconflictzones.Thissuggeststhatprofitmotivesdominatewhenlendingto“out-group”countries,whilepoliticalconstraintsorregulatoryscrutinylimitmilitaryfinancingwithingeopoliticalblocs.

Tofurtherexaminemechanisms,weanalyzehowbankspecializationshapescross-borderlendingduringviolentconflicts,usingmeasuresofrelativespecialization(

Paravisini,Rap-

poportandSchnabl

,

2023

).Theresultsrevealageneralizedflight-homeeffect,withallbanksreducingnon-militarylendingtoconflictzones.However,thisbroadretreatissystematicallyoffsetbyincreasedmilitarylendingfromtwodistinctbanktypes:bankswithoutpriorcoun-tryspecializationaggressivelyredirectcapitaltowardmilitarysectorsinunfamiliarconflictzones,whilebankswithestablishedmilitarysectorexpertisealsosharplyincreasedefenselendingduringconflicts.Thisindicatesthatourdocumentedreallocationisspecificallydrivenbycross-borderlenderswithmilitaryexpertisebutlimitedcountry-specificrelation-

ECBWorkingPaperSeriesNo30738

ships.Suchbanksappearrelativelywellpositionedtocapitalizeonconflict-inducedcreditdemandwithouthavingtocompromiseestablishedcountry-specificclientrelationships.

Severalextensionshelptodelineatethescopeandboundariesofourmainresults.Wefindnoevidenceofspillovereffectstoneighboringnon-conflictcountries,indicatingthatforeignbankstargetprimaryconflictcountriesratherthanbroaderregions.Moreover,therelativeincreaseinmilitarylendingdissipatescompletelywithinthreeyearspost-conflict,suggestingbanksreacttoimmediateconflictdynamicsratherthanlong-termstrategic(re-)positioning.Finally,bothstate-ownedandprivatebanks,aswellasbankandnon-banklenders,displaysimilarreallocationpatterns,indicatingthatpoliticalownershipdoesnotdriveourresults.

Relatedliterature.Thispapercontributestothreestrandsoftheliterature.First,weextendresearchoninternationalprivatecapitalflows.Previousworkhasanalyzedhowin-vestorsallocatecapitalabroad(

LaneandMilesi-Ferretti

,

2007

;

CoeurdacierandRey

,

2013

;

BrunoandShin

,

2015

;

Maggiori,NeimanandSchreger

,

2020

;

Coppola,Maggiori,Neiman

andSchreger

,

2021

),howthisallocationaffectsrecipienteconomies(

Calvo,Leidermanand

Reinhart

,

1993

),andhowprivatecapitalflowsco-moveaspartofaglobalcycle(

Rey

,

2015

).Severalpapersexaminehowcross-bordercreditflows,especiallyintheformofsyndicatedlending,cantransmitfinancialandreal-economicshocksacrossborders.

1

Ouranalysisex-tendsthisliteraturebyrevealingacontrastinglendingdynamicduringviolentconflicts:whilecross-borderlenderssignificantlyreduceoveralllending,theysimultaneouslyredirectcapitaltowardsectorspositionedtobenefitfromlocalinstability,particularlythemilitary-industrialcomplex.Thispatternconfirmsexistingevidenceaboutdistanceconstraintsoncross-borderlending,whilerevealinganewmechanismthroughwhichforeignbanksreallo-catecredittowardmilitarysectorsduringarmedconflicts.

Second,wecontributetoanemergingliteraturethatexamineshowfinancialmarketsinteractwithmilitaryconflict.Previousresearchhasprimarilyfocusedonsovereignbor-

1E.g.,

CetorelliandGoldberg

(

2011

);

GiannettiandLaeven

(

2012

);

PopovandUdell

(

2012

);

DeHaas

andVanHoren

(

2013

);

Cerutti,HaleandMinoiu

(

2015

);

Hale,KapanandMinoiu

(

2020

);

DoerrandSchaz

(

2021

).

ECBWorkingPaperSeriesNo30739

rowing.

Horn,ReinhartandTrebesch

(

2024

)documenthowwarstriggerdramaticfinancialchangesasgovernment-to-governmentlendingincreaseswhileoverallprivatecapitalflowsshrink.

DiGiuseppe

(

2015

)findthatsovereigncreditenablesstatestosimultaneouslyfi-nancemilitaryandcivilianspending,circumventingbudgetaryconstraints,while

Federle,

RohnerandSchularick

(

2025

)demonstratehowfinancialaccesscandrivemilitarysuccess:countriesexperiencingcommoditywindfallssignificantlyimprovetheirchancesofvictoryastheycanrampupmilitaryexpenditures.Ourcontributionistodocumenthowprivatecross-bordercreditflowscanbeachannelformilitaryfinancingduringviolentconflicts,whendomesticcreditmarketsareoftenconstrained(

Mamonov,OngenaandPestova

,

2024

).

Third,weshedlightonthefinancialrepercussionsofgeopoliticalfragmentation.Re-centpapersdocumenthowglobaltrade,investment,andsupplychainshavefragmentedalonggeopoliticallinessincetheonsetofRussia’sinvasionofUkraineandescalatingUS-Chinatradetensions.Thesestudiesrevealaprogressivefragmentationofeconomiclinkagesbetweenrivalgeopoliticalblocs,partiallymitigatedbyemerging“connector”countries.

2

Anemergingliteraturehasbeguntoexplorehowgeopoliticaltensionsaffectfinancialmarkets.

NiepmannandShen

(

2025

)showthatinternationallyactiveUSbanksrespondtogeopoliticalriskbyreducingcross-borderanddomesticlending,whilemaintainingcreditsupplythroughlocalaffiliatesinriskycountries,adynamicthatgeneratessignificanteco-nomicspillovers.

Danisewicz,Park,SchaeckandZheng

(

2025

)findthatcross-borderlendersstrategicallyrespondedtoRussia’scounter-sanctionsontheEU’sagriculturalsectorbyin-creasinglendingandofferingmorefavorableloantermstothissector.Theyalsoshowthatbankswithsector-specificexpertisecanmitigateeconomicdisruptionsinsanctionedindus-tries.

Efing,GoldbachandNitsch

(

2023

)findthatwhileGermanbanksreducedcross-borderlendingtosanctionedcountries,theirforeignaffiliatesinlessregulatedjurisdictionsexpandedcredit.Similarly,

Besedeˇs,GoldbachandNitsch

(

2017

),alsoforGermany,showthatdirect

2See

AlfaroandChor

(

2023

);

Chupilkin,JavorcikandPlekhanov

(

2023

);

Aiyar,MalacrinoandPresbitero

(

2024

);

Chupilkin,Javorcik,PeevaandPlekhanov

(

2024

);

Gopinath,Gourinchas,PresbiteroandTopalova

(

2025

).

ECBWorkingPaperSeriesNo307310

financialtransactionswithsanctionedcountriesdeclinebutarepartiallyreplacedbyflowsthroughintermediarynations.Thus,bothstudiesdemonstratehowregulatorygapsallowprivatecapitalflowstoadapttoandpartiallycircumventgeopoliticalrestrictions.

Unlikethisexistingworkonsanctionsandregulatoryarbitrage,weexamineviolentcon-flictsasadistinctshockthattriggersasymmetriccreditreallocation:foreignbanksreducecivilianlendingwhilesimultaneouslyincreasingmilitary-sectorfinancing,particularlyinpo-liticallynon-alignedcountries.Ourresultscomplement

Kempf,Luo,Sch¨aferandTsoutsoura

(

2023

),whofindthatideologicalalignmentshapescross-bordercapitalallocationduringpeacetime.Byanalyzingdifferentiallendingpatternsfrompoliticallyalignedversusnon-alignedcountriestoconflictzones,weidentifyapreviouslyunexaminedchannelthroughwhichgeopoliticaldistanceenablesratherthanconstrainsfinancialflows—specificallyto-wardmilitarysectorsduringarmedconflicts.

2Data

2.1Datasources

OuranalysisrequiresustomergethedatafromtheUppsalaConflictDataProgram(UCDP)withtheLoanPricingCorporation’s(LPC)DealScandata.TheUCDPprovidescompre-hensiveandharmonizedinformationonarmedconflictsandorganizedviolenceovernearlyfourdecades.Wefocusonstate-basedarmedconflicts,whichcausemostbattle-relatedfa-talities(

Melander,PetterssonandThemn′er

,

2016

).Theseareconflictsbetweentwoparties,ofwhichatleastoneisastategovernment,resultinginatleast25fatalitieswithinayear.Weaggregatebattledeathsatthecountry-yearlevel.

FromDealScan,wecollectcomprehensivedataonsyndicatedlendingtocorporationsgloballyovera31-yearperiod(1989-2020).Thisextensivedatasetallowsustoobservetheuniverseofsyndicatedloantransactions,capturinglendingrelationshipsbetweenfinancialinstitutionsandborrowersacrossdifferentcountries,sectors,andtimeperiods.Wesplit

ECBWorkingPaperSeriesNo307311

eachloanintosyndicatemembersharestocreateourunitofobservation:asyndicatedloanportionbyanindividualbanktoanindividualborrowerinagivenyear.SinceDealScanprovidesloansharedistributionsforonly26%ofloans,weimputemissingsharesusingeachbank’shistoricalaveragesharefromloanswithknownallocationsandthenre-weighthesesharessothattheyaddupto100%.

3

WeconvertamountstoUSdollarsanddateeachobservationtotheloan’soriginationyear.DealScanprovidesthecountriesofbothlendersandborrowers(wemanuallydoublecheckbankheadquarterslocations).Inlinewiththeliterature(

Mian

,

2006

;

GiannettiandLaeven

,

2012

),weclassifyaloanasforeignwhenthebankoritsparentcompanyisincorporatedinadifferentcountrythantheborrower.Forexample,aloantoaNigeriancompanyfromCitibankisconsideredforeignregardlessofwhetheritcomesfromCitibankUSoralocalCitibankaffiliateinNigeria.

2.2Identifyingmilitaryanddual-usesectors

Wecategorizemilitary-relatedsectorsintotwotypes:primarymilitaryanddual-use.Pri-marymilitarysectorsexclusivelyproducegoodsandtechnologiesformilitaryanddefensepurposes,suchasmissilesandtanks.Incontrast,dual-usesectorsproducegoods,tech-nologies,orservicesintendedforcivilianusebutwithaclearcapabilitytoperformmilitaryfunctions.Forexample,commercialaircraftenginetechnologycanbereadilyadaptedforfighterjets,withthesameadvancedmaterials,propulsionsystems,andengineeringprinci-plesservingbothciviliantransportationandmilitaryaviationneeds.

Becausetheexportofdual-useproductscanposenationalsecurityconcerns,manycoun-triescreatelistsofitemsthatrequireexportauthorization.WeusetheUK’sStrategicExportControlListfromtheDepartmentforBusinessandTrade(specificallythe“MilitaryList”and“Dual-UseList”)toidentifymilitaryanddual-useindustries.Fromthefirstlist,wecollectkeytermssuchas‘weapon’,‘gun’,‘a(chǎn)rtillery’,‘tank’,‘bomb’,‘torpedo’,‘missile’and‘explosives’.Wethenidentifyall4-digitSICcodesontheNAICS/SICwebsitethat

3Ourresultsarerobusttoalternativeimputationmethods,suchassplittingtheoverallloanequallyacrossallbanksinthesyndicate(

DeHaasandVanHoren

,

2013

).SeeAppendixTable

L.I

.

ECBWorkingPaperSeriesNo307312

mentionthesegoods.Thisyields10primarymilitarySICcodes.

Forthedual-uselist,weapplyasimilarbutslightlyadjustedapproach.WeextractkeywordsfromtheUK’sdual-usecategorytitles(suchas‘nuclear’and‘a(chǎn)ircraft’)andsearchforthesetermsontheNAICS/SICwebsite.Thisgenerates115potentialdual-useSICcodes.Wethenevaluateeachsector’slikelihoodofmilitaryassociationbyaskingChatGPT-4otoassesstheprobabilityofmilitaryproductioninvolvementforall125codes(10primarymilitaryplus115dual-usesectors).Weperform50iterations,randomlyreorderingthe125SICcodeseachtime,andcalculatetheaverageprobabilityforeachsector(

Appendix

B

describesthisapproachinmoredetail).TheSpearmanrankcorrelationinFigu