ContentslistsavailableatScienceDirect
Resource
and
Energy
Economics
j o ur na l h o me pa g e:w w w . e l s e v i e r . c o m / l o c a t e / r e e
The
measurement
of
environmental
economic
inefficiency
with
pollution-generating
technologies
Juan
Aparicio
a,
Magdalena
Kapelko
b,
José
L.
Zofío
c,d,∗aCenterofOperationsResearch(CIO).UniversidadMiguelHernández,Elche,Spain bDepartmentofLogistics,WroclawUniversityofEconomicsandBusiness,Wrocław,Poland cDepartmentofEconomics.UniversidadAutónomadeMadrid,Madrid,Spain
dErasmusResearchInstituteofManagement,ErasmusUniversity,Rotterdam,TheNetherlands
a
r
t
i
c
l
e
i
n
f
o
Articlehistory: Received1July2019
Receivedinrevisedform9April2020 Accepted8June2020
Availableonline20June2020
JELclassification: D20 D24 N52 Q50 Keywords:
Environmentaleconomicinefficiency Pollution-generatingtechnologies Technicalandallocativeefficiency measurement
Dataenvelopmentanalysis USagriculture
a
b
s
t
r
a
c
t
Thisstudyintroducesthemeasurementofenvironmentalinefficiencyfromaneconomic
perspective.Wedevelopourproposalusingthelatestby-productionmodelsthatconsider
twoseparateandparalleltechnologies:astandardtechnologygeneratinggoodoutputs,
andapollutingtechnologyfortheby-productionofbadoutputs.Whileresearchinto
environmentalinefficiencyincorporatingundesirableorbadoutputsfromatechnological
perspectiveiswellestablished,nosignificantattemptshavebeenmadetoextendittothe
economicsphere.Basedonthedefinitionofnetprofits,wedevelopaneconomicinefficiency
measurethataccountsforsuboptimalbehaviorintheformofforegoneprivaterevenue
andenvironmentalcostexcess.Weshowthateconomicinefficiencycanbeconsistently
decomposedaccordingtotechnicalandallocativecriteria,consideringthetwoseparate
technologiesandmarketprices,respectively.Weillustratetheempiricalimplementation
ofourapproachusingadatasetonagricultureatthelevelofUSstates.
©2020TheAuthors.PublishedbyElsevierB.V.ThisisanopenaccessarticleundertheCC
BYlicense(http://creativecommons.org/licenses/by/4.0/).
1. Introduction
Measuringtheenvironmentalinefficiencyofproductionunitsisanincreasinglyimportanttopicofrecenteconomic research.Environmentalinefficiencyassessmentintegratesmarketed(desirable,intended,orgood)outputswithnegative environmentalexternalitiesintoproductionmodeling(theproductionofso-calledundesirable,unintended,detrimental,or badoutputs).Suchanalysisisimportantfromtheperspectiveofsustainableproductionbecauseitprovidesvaluableinsights forfirmsandindustrystakeholdersonhowtoadoptenvironmentallyfriendlystrategies,andforpolicymakerstoimprove thedesignofpollutant-abatementinstruments,accountingforenvironmentalchallenges.
However,theexistingenvironmentalefficiencymodelslacktheeconomicinefficiencydimensionoftheanalysis;that is,acomprehensivemeasurethatalsoconsiderstheforegoneprofits,intheformoflowerrevenuesand/orhighercosts, thatarenotonlyrelatedtoatechnologicalinefficientbehavior,butalsotoallocativeinefficiency.Thisimpliesthatfirms shouldnotonlypursuebeingtechnicallyefficientbyexploitingthepotentialoftheproductionfrontier,buttheyshouldalso
∗ Correspondingauthorat:DepartmentofEconomics.UniversidadAutónomadeMadrid,Madrid,Spain. E-mailaddresses:jose.zofio@uam.es,jzofio@rsm.nl(J.L.Zofío).
https://doi.org/10.1016/j.reseneeco.2020.101185
0928-7655/©2020TheAuthors.PublishedbyElsevierB.V.ThisisanopenaccessarticleundertheCCBYlicense(http://creativecommons.org/licenses/by/ 4.0/).
usetheoptimalamountsof(goodandbad)outputsandinputsconsistentwithprofitmaximization(extensibletorevenue maximizationand/orcostminimization);i.e.,theyshouldbeallocativeefficientbysupplyinganddemandingtheoptimal bundles(ormixes)ofoutputandinputs.
Followingtheliteratureoneconomicefficiencymeasurementanditsdecompositionintotechnicalandallocative com-ponentsinitiatedbyFarrell(1957),webringthistheoreticalframeworktothefieldofenvironmentaleconomics.Inthis framework,economicefficiencyanalysisnotonlyconsiderstheproductionofgoodoutputs(associatedtoprivaterevenue), butmustalsoaccountfortheeconomiccostofproducingbadoutputs,whichisproxiedthroughtheso-calleddamagecost functions;andsincebadoutputsareexternalities,theyarealsorelatedintheliteraturetosocialcostfunctions,e.g.,see BretschgerandPattakou(2019).Fromtheperspectiveofthefirm,thepossibilityofsimultaneouslyincreasingmarketgoods whilereducingenvironmentallydamagingoutputsiseconomicallyappealing.Ononehand,thereisaclear(private) incen-tivetoincreaserevenue,butalsocurrentsustainabilityandcorporatesocialresponsibilityconcernsareincreasinglybrought intofirms’decisionmaking.Thismeansthatinefficientfirms,fallingshortfromthebestpracticefrontier,canimproveits environmentalefficiencyatnoprivatecost,andthereforeitispossibletoreducetheexternalitiescausedbybadoutputs bymatchingbestpracticestandardsofefficiency.Nowadaysfirmsroutinelyadvertiseenvironmentalachievementsintheir annualreports;e.g.,fortheairlineindustrythereductionofthecarbonfootprintisbecomingincreasinglyimportant.
Fromtheperspectiveofenvironmentaleconomics,theaboveframeworkcanberelatedtothenotionofnetprofits(ornet revenuesifweconsiderproductiveinputsasgiven,aswedoforsimplicityinourmodel).Thismeansthattheobjective func-tionofthefirmcorrespondstothemaximumofitsmarketrevenuesnetoftheenvironmentalcoststhatitcausesthroughthe inevitableby-productionofbadoutputs(i.e.,thematerialsbalanceprincipleassociatedtothefirstlawofthermodynamics). Ofcourse,thereductionofbadoutputsentailsabatementcostsifthefirmisefficient,therebyproducingatthetechnological frontier(andthisisreflectedbythesubstitutabilitycharacteristicsofthetechnologybetweenthegoodandthebadoutputs, seeMurtyetal.,2012).Butifthefirmistechnicallyinefficient,bothgoodandbadoutputscanbefreelyincreasedand reduced,respectively.Consequently,inefficientfirmscanincreaseprivaterevenuewhilereducingenvironmentaldamage. Insum,theenvironmentaleconomicefficiencymodelthatweproposedintheveinofFarrell(1957)extendstheexisting technologicalmodelsforenvironmentalefficiencymeasurementtoaccountfortheseeconomicdimensionsbypostulating anobjectivefunctionthataimsatmaximizingprivaterevenuenetoftheenvironmentaldamage.
Thedeterminationofeconomicefficiencyisimportantfromamanagerialstandpointfocusedonmarket-oriented perfor-mance,butalsoforotherstakeholdersanddecisionsmakerssuchas,forexample,politicians,localandstategovernments, orregulators(e.g.,environmentalprotectionagencies).Forexample,managersareinterestedinincreasingperformance notonlyinphysicaltermsbytakingadvantageofthebesttechnologyavailable,butalsobyrealizingtheeconomicgains associatedwithallocativeefficiencyimprovements;thatis,thechoiceofoptimaloutputandinputmixes,leadingtoeither maximumprofit,revenueorminimumcost.Theinformationonallocativeefficiencythatoureconomicmodelyieldsis alsorelevantfortheaboveeconomicagentsasimprovingallocativeinefficiencyisarguablycheaperandeasierforfirmsto achievethanimprovingtheirtechnicalinefficiency.Inthissense,beingawareofthelevelofthisinefficiencyandrelated potentialforrevenueincreasesorcostsavingsenablesfirmsto“reapalow-hangingfruit”.Forregulatorsminimizingthe environmentalcostofproductionfromanallocativeperspectiveisalsocritical,asthecostofcarbondioxideemissions (e.g.,relatedtorespiratoryillnesses)couldbelowerthanthoseassociatedtotheuseofpesticides(e.g.,relatedtocancer treatments).Howthepollutinginputscausingbothdamagesshouldberegulatedintermsoftheireconomiccostscannow beaddressedthankstoournewframework.
Consideringonlyatechnologicalperspective,theliteratureonmodelingproductiontechnologiesthataccountforbad outputshasdevelopedfollowingtwoapproachesmainly:oneinvolvingparametricmethods(suchasstochasticfrontier analysis,SFA;Aigneretal.,1977),andonebasedonnonparametricmethods(suchasdataenvelopmentanalysis,DEA;
Charnesetal.,1978;Bankeretal.,1984).Commontobothmethods,manydifferentapproacheshavebeenproposedto assessenvironmentalefficiencyofproductionunits.Lauwers(2009)classifiedtheseapproachesintothreegroups.Thefirst groupconcernsenvironmentallyadjustedproductionefficiencymodels,inwhichundesirableoutputsareincorporatedinto theproductiontechnology.Ingeneral,twomainbranchesofstudieswithinthisgroupcanbedistinguished:(i)treatingbad outputsasstrong(free)disposableinputs(Haynesetal.,1993;HailuandVeeman,2001)or(ii)treatingbadoutputsasweekly disposableoutputsandassumingthenull-jointnessofbothbadandgoodoutputs(Färeetal.,1986,1989).Thesecondgroup ofstudiesconsistsoffrontiereco-efficiencymodels(KorhonenandLuptacik,2004;KuosmanenandKortelainen,2005), whichdonotfollowaxiomaticproductionefficiencyframeworks,butrelateaggregateecologicaloutcomeswitheconomic outcomesonly.Inotherwords,eco-efficiencyismeasuredeitherthroughminimizationofenvironmentaloutcomesgiven economicoutcomes(e.g.,valueadded)orthealternativemaximizationofeconomicoutcomesgiventheenvironmental outcomes.Thethirdgroupofstudiesisbasedontheintroductionofthematerialsbalanceprincipleintoproductionmodels (LauwersandVanHuylenbroeck,2003;Coellietal.,2007;Førsund,2009).Thematerialsbalanceprinciplestatesthatflows intoandoutoftheenvironmentareequal,linkingtherawmaterialsusedintheproductionsystemtooutputs,bothintended andresidualones.
Dakpoetal.’s(2016)recentsurveyofenvironmentalefficiencystudiesextendedtheLauwers(2009)classificationinto thefourth,mostrecent,categoryofby-productionmodels,whicharebasedontheideaofdefiningtwosubtechnologiesin parallel:onethatgeneratesgoodoutputsandasecondthatgeneratesbadoutputs.ThisapproachwasintroducedbyMurty etal.(2012)and,asaconsistentandrelativelynewapproach,itsempiricalapplicationsareflourishing(e.g.,Dakpoetal., 2017;Arjomandietal.,2018;Rayetal.,2018),asareitsextensions(e.g.,Serraetal.,2014;Lozano,2015;Dakpo,2016;
Førsund,2018).Werelyonthenovelby-productionapproachtointroduceoureconomicenvironmentalefficiencymodel. Nevertheless,itcouldbeeasilyparticularizedforpreviousapproaches.1Wealsoconsiderrecentqualificationsoftheoriginal
by-productionmodelbyDakpo(2016)andFørsund(2018).2
Asanticipated,regardlessthemodelingapproach—parametricornon-parametric—underthefourlistedcategories,a commonfeatureofallpreviousstudiesisthattheyareonlycapableofmeasuringtechnicalefficiencybyfocusingonthe technologicalsideoftheproductionprocess,therebyneglectingthemeasurementofenvironmentalefficiencyfroman economicperspective.Thisallowsustosummarizewhatourmodeldoes;i.e.,enhancingtheexistingapproachesthroughthe introductionofameasureofenvironmentaleconomicinefficiencythat,groundedonthetheoreticalframeworkproposedby
Farrell(1957),considersbothgoodandbadoutputs,andenablesitsdecompositionintotechnicalandallocativecomponents. Tofillinthegapintheliteraturewepostulateacomprehensiveframeworkthatisconsistentwiththeeconomicbehaviorof organizationsintheirattempttomaximizerevenue,butalsoaccountsfortheenvironmentalinefficiencythatresultsfrom thefailuretominimizetheeconomiccostsassociatedtoenvironmentaldamage.Aspreviouslyremarked,thisresultsinthe definitionofaneteconomicfunctionthatmaximizesthedifferencebetweenprivate(market)revenuelessenvironmental (social)cost,usingpricesofgoodandbadoutputs.3 Inthisregard,ourframeworkiscapableofbalancingprivategains
(revenue)andenvironmentaldamage(cost)intoameasureofeconomicinefficiencythatcanbedecomposedaccordingto technicalandallocativecriteria.
From an appliedperspective we rely onDEA techniques becausemost existing empirical applicationsfollow this approach:theyareflexible,donotimposerestrictiveassumptionsontheparametricspecificationofthetechnology,noron thedistributionofenvironmentalinefficiency.4Nevertheless,thedrawbacksofDEAshouldbealsohighlightedandthese
includeitsdeterministicnatureandthesensitivitytooutliers(foracomprehensiveexpositionofstrengthsandweaknesses ofDEAsee,forexample,Stolp(1990),Berg(2010)).Awareofthesecaveats,whichcanbeeventuallyaddressedthrough,e.g., bootstrappingandotherresamplingtechniques(seethemethodsintroducedbySimarandZelenyuk(2006)employedinthe empiricalsection),wedefinetheDEAprogramsthatallowtheempiricalimplementationofournovelapproach.5Ourpointof
departureistheby-productionmodelintroducedbyMurtyetal.(2012),asitrepresentsthemostrecentextensionof previ-ousapproachesandcanarguablybeseenasageneralizationthat,byconsideringtwoindependenttechnologiesfordesirable andundesirableoutputs,avoidssomeoftheirinconsistencies(namely,themultiplicityofoptimalcombinationsof desir-ableandundesirableoutputsforagivenlevelofinputs,anderroneouslysignedmarginalratesoftransformation−shadow prices−betweenoutputsandinputs).
Wedemonstratethepracticalusefulnessofournewlydevelopedmethodologythroughanapplicationtostate-leveldata oftheUnitedStatesagriculturalsector.Agricultureinvolvestheproductionofnotonlygoodoutputssuchasprimaryfood commodities,butalsoofbadoutputsrelatedwith,forexample,theneedforfuel,theusageofpesticides,fertilizersand otheragriculturechemicals,orthemanagementofmanure(Skinneretal.,1997;Reinhardetal.,1999).Examplesofbad outputsassociatedtothesepollutinginputsinagriculturearegreenhousegasemissions,pesticideandnitrogenleaching andrunoff,risktohumanhealthandfishfromexposuretopesticidesandfertilizers,etc.(seeBalletal.,2001;Kellogetal., 2002;Dakpoetal.,2017).Intheempiricalapplicationwearecapableofconsideringtwoofthesebadoutputs:CO2emissions
andpesticideexposures.
Theremainderofthispaperisstructuredasfollows.Thenextsectionreviewstheby-productionmodelsoftechnical inefficiencyandintroducestheirmathematicalunderpinnings.Thesubsequentsectiondevelopsourextensionallowingthe measurementofeconomic(profit)inefficiency.Wethendiscussourempiricalapplication,brieflycommentingthedataset andpresentingtheresults.Conclusionsaredrawninthefinalsection.
2. Theby-productionmodels
Pittman(1983)andFäreetal.(1986)initiatedtheasymmetricmodelingofoutputswhenmeasuringefficiencydepending ontheirnature,increasingthosethataremarket-orientedwhilereducingthosethataredetrimentaltotheenvironment.A keyquestionishowtoaxiomaticallymodeltheproductiontechnologywhencalculatingtechnicalefficiencythroughdistance functions.Mostparticularly,ascommentedintheintroductiontothispaper,shouldtheaxiomsunderlyingtheproduction
1Detailsonthecharacteristicsoftheby-productionapproacharepresentedinthenextsection.
2Althoughweareawareofothermethodologicaldevelopmentsthatrelyontheby-productionmodel,suchasSerraetal.(2014)orLozano(2015),we
havenotconsideredthemsincetheirgeneralideaistomixtheby-productionapproachwithotherefficiencyframeworks,andnotthemodificationofthe modelperse.Hence,ifapplied,theirresultswouldnotbecomparabletothoseoftheoriginalby-productionmodel.
3Themodelcanbeeasilyenhancedtoincludetheminimizationofinputscost,butinsteadwekeepthedefinitionof“environmentalprofitinefficiency”
asatrade-offbetweenprivaterevenueandenvironmentalcost.
4SeeZofioandPrieto(2001)foranexpositionofearlymodelswithinthenon-parametricapproachbasedontheoutput,input,andhyperbolicdistance
functions,whichweresubsequentlyimplementedinaparametricframeworkbyCuestaetal.(2009).Duetal.(2016)relyonthelatterapproachtoestimate carbonabatementscoststhroughshadowprices.
5Brännlundetal.(1995)measuredprofitinefficiencyunderaquotasystemandtheproductionofundesirableoutputsbyDEAmodels.However,they
didnotusepricesforweightingthenegativeexternalitiesanddonotdecomposeprofitinefficiencyintoitsdrivers,somethingthatwewilldointhispaper. Additionally,wenotethatPhamandZelenyuk(2018)definerevenueinefficiencyinthebankingindustryaccountingfornonperformingloans(NPLs), whicharemodeledasundesirableoutputsundertheapproachofweakdisposability.However,themodelisinternaltothefirm(thatis,privaterevenue), asitdoesnotincludeenvironmentalindicators,whiletheydonotimplementitempirically.
technologyreflecttheirstrongorweakdisposability,andeventually,bemodeledasoutputsorasiftheywereinputs?Among theexistingapproachesfordealingwithundesirableoutputsandefficiency,theby-productionmodelintroducedbyMurty andRussell(2002)andMurtyetal.(2012)iscurrentlyconsideredapreferredoption.
Theby-productionapproachpositsthatcomplexproductionsystemsaremadeupofseveralindependentprocesses (Frisch,1965).Inthismodel,thetechnologycanbeseparatedintosetsofsub-technologies;onefortheproductionofgood outputsandoneforthegenerationofbadoutputs.The“global”technologyimpliesinteractionsbetweenseveralseparate sub-technologies.Førsund(2018)andMurtyandRussell(2018)recentlyclassifiedtheby-productionapproachamongthe multi-equationmodelingapproachesandarguedthatanimportantadvantageofthisapproachisthatitrepresents pollution-generatingtechnologiesbyaccountingfortheMaterialBalancePrinciple,therebysatisfyingthelawsofthermodynamics. Additionally,asMurtyetal.(2012)remarked,theby-productionmodelavoidstwoinconsistenciesofpreviousapproaches. Inparticular,severaltechnicalefficiencycombinationsofgoodandbadoutputs,withvaryinglevelsofbadoutput,couldbe possiblewhenholding(pollutingandnon-polluting)inputquantitiesfixed.However,intheabsenceofabatementactivities implementedbythefirm,thistypeofcombinationiscontrarytothephenomenonofby-production,sinceby-production impliesthat,atfixedlevelsofinputs,thereisonlyonelevelofpollutionatthefrontieroftheproductionpossibilityset. Moreover,itispossibletoobserveanegativetrade-offbetweentheinputsassociatedwithpollution,likefuel,andtheir associatedbadoutput,suchasCO2,whichrepresentsaclearinconsistency(morefuelbutlessCO2).Thesearethereasonswhy
theby-productionapproachisutilizedinthecurrentstudytointroducetheconceptofenvironmentaleconomicinefficiency takingmarketpricesintoaccount.
Inordertoreviewthestandardby-productionapproach,letusformallydefinex∈Rn
+asavectorofinputs,y∈Rm+ as
avectorofgoodoutputs,z ∈Rm
+ asavectorofbadoutputs(e.g.,pollutants),andletusassumethatpDMUshavebeen
observed.Murtyetal.(2012)presentedtheirmodelbysplittingtheinputvectorintotwogroups:non-pollutinginputs, x1 ∈Rn+1andpollution-generatinginputs,x2 ∈Rn+2,withn1+n2=n.6,7Thefirstsetcouldcompriseland,labor,andsoon,
whilethesecondset,inthecontextofourempiricalapplicationonagriculture,consistsofinputslikefuel,fertilizers,and pesticides,whichproducecertainpollutantsasby-products,suchasCO2emissionsandpesticideexposures.Inthisway,
the‘global’technology,denotedbyT ,istheintersectionoftwosub-technologies,T1andT2.WhereasT1isthestandard
productiontechnologywithonlygoodoutputs,T2representstheproductionofbadoutputs.InthemodelbyMurtyetal. (2012),bothtechnologiesarelinkedthroughthelevelofthepollutinginputs.Inmoredetail,Murtyetal.(2012)definein generaltermsthetechnologyas:
T=T1∩T2, (1) where T1=
(x1,x2,y,z)≥0:f (x1,x2,y)≤0 , (2) T2= (x1,x2,y,z)≥0:z≥g (x2) (3) andf andgarecontinuouslydifferentiablefunctions.ThesetT1 isastandardtechnologyset,reflectingthewaysinwhichtheinputscanbetransformedintotheintendedoutputs.Thestandardfree-disposabilitypropertiesmaybeimposed byassumingthat ∂f (x1,x2,y)
∂x1 ≤0,
∂f (x1,x2,y)
∂x2 ≤0and
∂f (x1,x2,y)
∂y ≥0.NotealsothatT1 imposesnoconstraintonz;thatis,it
isimplicitlyassumedthattheby-productdoesnotaffecttheproductionofbadoutputs.Ontheotherhand,T2 reflects
aresidual-generationmechanism.Itisworthmentioningthat,intheformulationofT2,pollutionisreallytreatedasan
output.Inparticular,Murtyetal.(2012)assumethat ∂g(x2)
∂x2 >0.Thisexpressionand theformulationofT2 capturethe factthatpollutionisanoutputoftheproductionprocessforwhichdisposalisnotfree.Thispropertywascalled“costly disposability”ofresiduals.InwordsofMurtyetal.(2012):“Costlydisposabilityimpliesthepossibilityofinefficienciesin thegenerationofpollution(e.g.,ifagivenlevelofcoalgeneratessomeminimallevelofsmoketheninefficiencyintheuse ofcoalmayimplythatthislevelofcoalcanalsogenerateagreateramountofsmoke)”.
Inthenon-parametricframeworkofDEA,thetwosub-technologiesmaybeexpressedmathematicallyundervariable returnstoscale(VRS)as:
T1=
(x1,x2,y,z)≥0: p d=1 dx1d≤x1, p d=1 dx2d≤x2, p d=1 dyd≥y, p d=1 d=1,d≥0 , (4) T2= (x1,x2,y,z)≥0: p d=1 dx2d≥x2, p d=1 dzd≤z, p d=1 d=1,d≥0 . (5)6 AyresandKneese(1969)proposedthesetwosamegroupswhenintroducingthematerialsbalanceprincipletoeconomists.
7 AsMurtyetal.(2012),weassumethatDecisionMakingUnitsapplyuniformabatementfactorsand,consequently,thesefactorsarenotexplicitly
T1in(4)istherepresentationofthegeneralsetT1in(2)underseveralpostulatesasconvexityandminimalextrapolation
(seeBankeretal.,1984).ThesamecanbesaidforT2in(5)withrespecttothegeneralexpressionofT2in(3).Notethatthe
sub-technologiesaredefined,inthisframework,withtwodifferentintensityvariables:and.Otherwise,wewillhavea confusionbetweenthesevariablesintheabovetwoproductionpossibilitysets.
Regardingthemeasurementoftechnicalefficiency,Murtyetal.(2012)showedthatsomeconventionalapproaches,like thehyperbolicanddirectionaldistancefunctiondefinedonT=T1∩T2,areinadequateinthecontextofby-production.
Weusetheterm“output-oriented”inthiscontextbecausethesedistancefunctionsmeasureefficiencywithrespectto both goodandbadoutputs simultaneously.In thisway,theweaknessis duetothefactthat thetwo aforementioned measuresusethesamecoefficient(decisionvariable)fordeterminingefficiencybothinT1 forthegoodoutputsandT2
forthebadoutputs.Thisimpliesthatitispossibletoreachtheefficiencyfrontierforsomeofthesub-technologies,but theobservationcanfallshortofachievingthefrontieroftheotherone.Forconsistency,efficiencyintheby-production approachrequiresmodelsthatprojecttheassessedobservationsontoboththeefficientfrontierofT1 andtheefficient
frontierofT2.
TheabovementioneddrawbacksofstandardapproachesmotivatedMurtyetal.(2012)toproposeadifferentmeasurefor dealingwithgoodandbadoutputsunderby-production.ForDMU0,thismeasureisgood-output-specificand
bad-output-specific,andisbasedontheindexpreviouslydefinedbyFäreetal.(1985):
min 1 2
⎡
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎣
1 m m j=1 jstandard efficiency + m1 m k=1 k
environmental efficiency
⎤
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎦
s.t. p d=1 dxid≤xi0, i=1,...,n p d=1 dyjd≥yj0/j, j=1,...,m p d=1 d=1, p d=1 dxid≥xi0, i=n1+1,...,n p d=1 dzkd≤kzk0, k=1,...,m p d=1 d=1, j≤1, j=1,...,m k≤1, k=1,...,m d≥0,d≥0, d=1,...,p (6)Model(6)projectstheassessedDMU0(x10,x20,y0,z0) ontotheefficientfrontierofthetechnologyTbyincreasinggood
outputsandreducingbadoutputs.Thesechangesarevariable-specific,usingadifferentdecisionvariableforeachdimension: j,j=1,...,m,andk,k=1,...,m.Theconstraintsofmodel(6)coincidewiththerestrictionsthatdefinetheDEAproduction
possibilitysetsT1 andT2 in(4)and(5).Additionally,theoptimalvalueof(6)coincideswiththemeanofthestandard
separable.Inthiscase,thismeansthattheoptimalvaluecanbedeterminedasthemeanofamodelthatminimizes 1 m m
j=1 jonT1andamodelthatminimizesm1
m
k=1 konT2: min 1 m m j=1 j s.t. p d=1 dxid≤xi0, i=1,...,n p d=1 dyjd≥yj0/j, j=1,...,m p d=1 d=1, j≤1, j=1,...,m d≥0, d=1,...,p min 1 m m k=1 k s.t. p d=1 dxid≥xi0, i=n1+1,...,n2 p d=1 dzkd≤kzk0, k=1,...,m p d=1 d=1, k≤1, k=1,...,m d≥0, d=1,...,p (7)ItisworthmentioningthattherecentpaperbyFørsund(2018)arguedthatnon-pollutioncausinginputsshouldalsobe includedintechnologyT2giventhatsubstitutionbetweenthetwogroupsofinputscanhelpmitigatethepollution.Dakpo etal.(2017)indicatedthatsomeadditionalconstraintsmustbeaddedtotheby-productionapproachofMurtyetal.(2012)in ordertoguaranteethattheprojectionpointsfortheinputdimensionsarethesameinT1andT2.Inparticular,thecondition
thatshouldbeincorporatedtomodel(6)wouldbe:
p
d=1 dxid= p d=1dxid,
∀
i.Hereafter,weuseTMtodenotetheproductionpossibilitysetdefinedastheintersectionofT1andT2in(1)and(2),respectively,asawayofhighlightingthatthedefinition
ofthistechnologycorrespondstotheoriginalproposalofMurtyetal.(2012).Inthesameway,weuseTDtodenotethe
productionpossibilitysetdefinedfromtheoriginalby-productionapproachbutincorporatingtheconstraints
p
d=1 dxid= p d=1dxid,
∀
i,aspointedoutbyDakpoetal.(2017).Finally,wewillutilizeTMF todenotetheproductionpossibilitysetdefinedbyMurtyetal.(2012)butincorporatingnon-pollutinginputsintechnologyT2.Likewise,TDFdenotestheproduction
possibilitysetàlaDakpoetal.(2017)butagainconsideringnon-pollutinginputsinthedefinitionoftechnologyT2.
Tointroduceoureconomicinefficiencymodel,weextendthestate-of-the-artofby-productionapproach(Murtyetal., 2012;Dakpoetal.,2017andFørsund,2018)byincorporatinginformationonmarketprices.Todothat,weresorttoduality theoryfollowingChambersetal.(1998),and,morerecently,Aparicioetal.(2015),Aparicioetal.(2016a),andAparicioetal. (2016b).Inparticular,werecallrelevantdualityresultsconcerningthedirectionaldistancefunction8.Consequently,we
startoutbydefiningthistypeofmeasurefromanoutput-orientedperspectiveinthecontextofby-production.Underthe viewpointintroducedbyMurtyetal.(2012),weneedameasurethatallowsustoprojecttheassessedobservationsonto
8 Althoughthedirectionaldistancefunctioniswell-knownduetoitsflexibilityandbecauseitencompassestheShepharddistancefunctions,itpresents
somedrawbacks.Thismeasureneglectsslacksand,therefore,itdoesnottakeintoaccountallsourcesoftechnicalinefficiency(see,e.g.,Ray,2004). Additionally,theuseofthedirectionaldistancefunctioncouldresultininfeasibilitiesundercertainconditions,whichareanalysedindetailinBriecand Kerstens(2009).
theefficientfrontiersofT1 andT2simultaneously.Inthisway,the“by-production”directionaloutput-orienteddistance
functionfortheMurtyetal.(2012)approachwithdirectionalvectorg= (0,y0,z0) isdefinedasfollows: → B
x0,y0,z0;TM = max ıT1ˇT1+ıT2ˇT2 s.t. p j=1 j0xij≤xi0, i=1,...,n1 (8.1) p j=1 j0xij≤xi0, i=n1+1,...,n2 (8.2) − p j=1 j0yrj+ˇT1yr0≤−yr0, r=1,...,m (8.3) p j=1 j0=1, (8.4) − p j=1 j0xij≤−xi0, i=n1+1,...,n2 (8.5) p j=1 j0zkj+ˇT2zk0≤zk0, k=1,...,m (8.6) p j=1 j0=1, (8.7) ˇT1,ˇT2,j0,j0≥0 (8.8) (8)Model(8)projectstheassessedDMU0(x10,x20,y0,z0) ontotheefficientfrontierofthetechnologyTbyincreasinggood
outputsandreducingbadoutputs.Inthismodel,thechangesinoutputsarenotvariable-specific,i.e.,itdoesnotutilize adifferentdecisionvariableforeachdimension.Instead,itusesthesameexpansionfactorforthegoodoutputs,ˇT1,and thesamereductionfactorforthebadoutputs,ˇT2.Moreover,theconstraintsofmodel(8)coincidewiththerestrictions thatdefinetheDEAproductionpossibilitysetsT1andT2in(4)and(5).Additionally,theexogenouscoefficientsıT1 ≥0and
ıT2 ≥0,ıT1+ıT2=1,whichappearintheobjectivefunction,areweightsthatarefixedexogenouslybythecorresponding decisionmaker(manager,politician,regulator,etc.)toreflecttherelativeimportanceofthestandard(traditional)wayof producingversusthenewandcleanparadigmforgeneratinggoodsandservices.Additionally,itslineardualis:
→ B
x0,y0,z0;TM = min n1 i=1v
1 i0xi0+ n2 i=n1+1v
1 i0xi0− m r=1 u1r0yr0+˛10+ − n2 i=n1+1v
2 i0xi0+ m k=1 u2k0zk0+˛20 s.t. n1 i=1v
1 i0xij+ n2 i=n1+1v
1 i0xij− m r=1 u1r0yrj+˛01≥0, j=1,...,p (9.1) m r=1 u1r0yr0≥ıT1, (9.2) − n2 i=n1+1v
2 i0xij+ m k=1 u2 k0zkj+˛20≥0, j=1,...,p (9.3) m k=1 u2 k0zk0≥ı T2, (9.4)v
1 i0,v
2i0,u1r0,u2k0≥0, (9.5) ˛1 0,˛20free (9.6) (9)Finally,tocompletethisopeningsection,werecallthefirstadditivemeasureanddecompositionofeconomicinefficiency proposedintheliterature.WerefertotheNerlovianprofitinefficiencymeasure,whichcanbedecomposedintotechnical
inefficiency(thedirectionaldistancefunction)andaresidualterminterpretedasallocativeinefficiency(Chambersetal., 1998).9
In the standard production context, considering private revenue and cost only, and given a vector of input and output prices (ω,q) ∈Rn++m and technology T , the profit function ˘ is defined as ˘T(ω,q)=
max x,y
m r=1 qryr− n i=1 ωixi: (x,y)∈T.ProfitinefficiencyàlaNerloveforDMU0isdefinedasoptimalprofit(thatis,the
valueof theprofitfunctionatmarketprices)minusobservedprofit,both normalizedbythevalueofa reference
vec-torg= (gx,gy)∈Rn+m + : ˘T(ω,q)−
m r=1 qryr0− n i=1 ωixi0 m r=1 qrgyr+ n i=1 ωigix
.Additionally,Chambersetal.(1998)showedthatprofitinefficiency
maybedecomposedintotechnicalinefficiencyandallocativeinefficiency,wheretechnicalinefficiencycorrespondstothe directionaldistancefunction→DT(x0,y0;gx,gy)=max
ˇ:(x0−ˇgx,y0+ˇgy)∈T : ˘T(ω,q)− m r=1 qryr0− n i=1 ωixi0 m r=1 qrgry+ n i=1 ωigix =→DT x0,y0;gx,gy +AIN T x0,y0;ω,q;gx,gy (10)Inmodel(10),theleft-handsidecorrespondstoameasureofprofitinefficiency,definedasthenormalizeddifference betweenmaximumprofitandactualprofitatobservedmarketprices.Thismaybedecomposedintotechnicalinefficiency, i.e.,thevalueofthedirectionaldistancefunction→DT(x0,y0;gx,gy),andpriceorallocativeinefficiencyAITN(x0,y0;ω,q;gx,gy).
3. Measuringeconomicinefficiencywithby-productionmodelsinDEA 3.1. EconomicinefficiencymodelconsideringMurtyetal.’s(2012)technology
Wefirstintroducesomenotationanddefinitions.Givenafixedlevelofinputx0= (x10,...,xn0)∈Rn+andafixedlevel
ofbadoutputz0= (z10,...,zm0)∈Rm+,letusalsodefineasr (x0,z0,q,T ) themaximumfeasiblerevenuegiventheoutput
pricevectorq= (q1,...,qm)∈Rm++: r (x0,z0,q,T )=sup y
m r=1 qryr: (x0,y,z0)∈T =sup y m r=1 qryr: (x0,y,z0)∈ [T1∩T2] . (11)Eq.(11)representsagenericformulationforexpressinghowtodeterminethemaximumrevenue
m
r=1
qryr forgood
outputsthatcanbeobtainedgivenatechnologyT=T1∩Tandfixedquantitiesofinputsx0andbadoutputsz0.
UnderMurtyetal.’s(2012)approach, thisoptimizationproblemcanbealwayssolvedindependentlyonT1 andT2.
Therefore,asforT1,maximumfeasiblerevenuegiventheoutputpricevectorq= (q1,...,qm)∈Rm++maybedeterminedby:
r
x0,z0,q,TM =sup y m r=1 qryr: (x0,y,z0)∈TM =sup y m r=1 qryr: (x0,y,z0)∈T1 . (12)Again,(12)representsageneralformulationforstatinghowtodeterminethemaximumrevenue
m
r=1
qryrforgoodoutputs
thatcanbeobtainedgiventhesub-technologyT1andfixedquantitiesofinputsx0andbadoutputsz0.
9 SeealsoKoopandDiewert(1982)andZieschang(1983)forearlierdecompositionsofeconomic(cost)efficiencyintotechnicalandallocative
com-ponents.TheseauthorsimplementFarrell’s(1957)decompositionbasedontheradialinputmeasurewithinaparametric(Cobb-Douglas)deterministic approach.
Next,weexplicitlyshowhowthevalueofr
x0,z0,q,TMcanbecalculatedinDEAundertheby-productionframework (seeRay,2004): r
x0,z0,q,TM = max ,y s r=1 qryr s.t. p j=1 jxij≤xi0, i=1,...,n1 (13.1) p j=1 jxij≤xi0, i=n1+1,...,n2 (13.2) − p j=1 jyrj+yr≤0, r=1,...,m (13.3) p j=1 j=1, (13.4) j≥0, j=1,...,p (13.5) yr≥0, r=1,...,m (13.6) (13)Model(13)istheDEAimplementationofthegeneralexpressionin(12)fordeterminingmaximumrevenue.Theobjective functionisthesame,whiletheconstraintscoincidewiththosethatdefineT1in(4).
Thedualprogramof(13)is(14):10
min c,d, = n1
i=1 ci0xi0+ n2 i=n1+1 ci0xi0+ s.t. n1 i=1 ci0xij+ n2 i=n1+1 ci0xij− m r=1 dr0yjr+ 0≥0, j=1,...,p (14.1) dr0≥qr, r=1,...,m (14.2) ci0≥0 i=1,...,n (14.3) (14)Being(14)thedualproblemof(13),thedecisionvariablesci0,i=1,...,n,anddr0,r=1,...,m,canbeinterpretedas
shadowpriceswhilethedecisionvariable 0maybeinterpretedasshadowprofit(seeAparicioetal.,2015).
Ifrevenuemaximizationisassumed,asisthecasehere,thefirmfacesexogenouslydeterminedmarketoutputprices. Followingthisline,wemaysupposethattheobjectiveoftheDMUistochoosetheoutputscombinationthatyieldthe maximumrevenueattheapplicableprices.Inthissense,revenueinefficiencymeasureshowcloseistheobservedrevenue oftheDMUunderevaluationtothemaximumfeasiblerevenue.Toevaluateeconomiclossduetorevenueinefficiency, inthecontextofthedirectionaloutputdistancefunctions,FäreandPrimont(2006)provedthatanormalizedmeasureof
revenueinefficiency,inparticulartheratio
r(x0,q,T )− m
r=1 qryr0 m r=1 qrgr,maybedecomposedintotechnicalinefficiency,→Do
x0,y0;g
,plusaresidualterminterpretedasallocativeinefficiencyintheFarrelltradition,wherer (x0,q,T ) and →
Do
x0,y0;g
denotethe‘standard’revenuefunctionanddirectionaloutputdistancefunction,respectively,andgisthecorrespondingreference directionalvector.
Likewise,wecanintroducecostefficiencyfollowingthesamerationale,andbasedonthecostfunction.However,in ourcontextweareinterestedinenvironmentalcostfunctionsratherthanprivatecosts,representingameasureofthe (monetary)minimaldamagecausedbytheproductionofundesirableoutputs.Theenvironmentalcostfunctionrepresents a“monetizedmetric”oftheecologicalfootprintofthebadoutputs;see,forexample,Pearceetal.(1996)whorelatethe damagepertonofCO2withthesocialcostofcarbon(SCC).Correspondingly,anobservationiseconomicallyinefficientin
environmentaltermsif,giventheamountofundesirableoutputsproduced,itcauseslargerdamagethanthatrepresented
10Actually,thedualprogramofmodel(13)hasanadditionalsetofnon-negativityconstraintsforthedecisionvariablesd
r0,r=1,...,m.However,this
bytheminimumenvironmentalcostfunction(eitherasaresultoftechnicalorallocativeinefficiencies).Letusassumethat itispossibletoobserveorestimatepricesfortheundesirableoutputs:w= (w1,...,wm)∈Rm++.UnderMurtyetal.’s(2012)
approach,theeco-damagefunctionwillbenon-parametricallydetermineddirectlyfromT2asfollows.
D
x0,y0,w,TM = min ,z m k=1 wkzr s.t. p j=1 jxij≥xi0, i=n1+1,...,n2 (15.1) − p j=1 jzkj+zk≥0, k=1,...,m (15.2) p j=1 j=1, (15.3) j≥0, j=1,...,p (15.4) zk≥0, k=1,...,m (15.5) (15)Model(15)minimizestheenvironmentalcost,
m
k=1
wkzr,associatedwiththeproduction/emissionofz(badoutputs)
givenafixedquantityofpollution-generatinginputs.Theconstraintsin(15)coincidewiththerestrictionsthatdefinethe sub-technologyT2in(5).
Thedualprogramof(15)is(16): max e,f, n2
i=n1+1 ei0xi0−0 s.t. n2 i=n1+1 ei0xij− m k=1 fk0zkj−0≤0, j=1,...,p (16.1) fk0≤wk, r=1,...,m (16.2) ei0,fk0≥0 (16.3) (16)Being(16)thelineardualofmodel(15),itiswell-knownthattheoptimalvaluesofbothmodelsarerelated.
Wenowderive,byduality,anormalizedmeasureofeconomicinefficiencyandshowhowitcanbedecomposedinto (desirable)revenueinefficiencyandeco-damageinefficiency.Inordertodothat,wefirstprovethefollowingtechnical proposition. Proposition1. LetıT1,ıT2 >0. Then, inf t,h
⎧
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎩
rx0,z0,t,TM − m r=1 tryr0+ m k=1 hkzr0−DT2 x0,y0,h,TM :min⎧
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎩
m r=1 tryr0 ıT1 , m k=1 hkzk0 ıT2⎫
⎪
⎪
⎪
⎪
⎪
⎬
⎪
⎪
⎪
⎪
⎪
⎭
≥1⎫
⎪
⎪
⎪
⎪
⎪
⎬
⎪
⎪
⎪
⎪
⎪
⎭
≥→Bx0,y0,z0;TM .Proof. Let x0 ∈Rn+, y0 ∈Rm+, z0 ∈R+m and let t ∈Rm+, h∈Rm+ such that min
⎧
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎩
m r=1 tryr0 ıT1 , m k=1 hkzk0 ıT2⎫
⎪
⎪
⎪
⎪
⎬
⎪
⎪
⎪
⎪
⎭
≥1. Letc0∗,d∗0, ∗0
be an optimal solution of (14) and let e∗0,f0∗,∗0 be an optimal solution of (16) when x0 ∈Rn+,v
10,u10,˛10,
v
20,u20,˛20=
c0∗,t, 0∗,e∗0,h,∗0 is a feasible solution of (9). Constraints (9.5) and (9.6) are trivially satisfied. Regarding (9.1), n1 i=1 c∗i0xij+ n2 i=n1+1 c∗i0xij− m r=1 tryrj+ 0∗≥ by(14.2) n1 i=1 c∗i0xij+ n2 i=n1+1 c∗i0xij− m r=1 d∗r0yrj+ 0∗
≥ by(14.1) 0. As for (9.2), m r=1 tryr0 ıT1 ≥1 since m r=1 tryr0 ıT1 ≥min
⎧
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎩
m r=1 tryr0 ıT1 , m k=1 hkzk0 ıT2⎫
⎪
⎪
⎪
⎪
⎬
⎪
⎪
⎪
⎪
⎭
≥1. Therefore, m r=1tryr0≥ıT1. In the same way,
it is possible to prove that (9.3) and (9.4) are also satisfied. In particular, constraint (9.3) holds by (16.1) and (16.2). Consequently,
c∗0,t, ∗0,e∗0,h,∗0 is a feasible solution of (9). Regarding the objective function of (9) evaluated at this point, →Bx0,y0,z0;TM ≤ n1 i=1 c∗i0xi0+ n2 i=n1+1 c∗i0xi0− m r=1 tryr0 + ∗0− n2 i=n1+1 e∗i0xi0+ m k=1 hkzk0+ ∗0 = rx0,z0,t,TM − m r=1 tryr0+ m k=1 hkzr0−D x0,y0,h,TM, since models (13) and (14) have the same optimal value and models (15) and (16) also have the same optimal value. In this way, →B
x0,y0,z0;TMis a lower boundoftheset
⎧
⎨
⎩
n1 i=1 ci0∗(t) xi0+ n2 i=n1+1 ci0∗(t) xi0− m r=1 tryr0+ ∗0(t)− n2 i=n1+1 e∗i0(h) xi0 + m k=1 hkzk0+∗0(h) :∀
(t,h)∈S0 , where c∗0(t) ,d∗0(t) , ∗0(t) is any optimal solution of (14) when q=t and e∗0(h) ,f0∗(h) ,∗0(h) is any opti-mal solution of (16) when w=h. Note that⎧
⎨
⎩
n1 i=1 c∗i0(t) xi0+ n2 i=n1+1 c∗i0(t) xi0− m r=1 tryr0+ ∗0(t)− n2 i=n1+1 e∗i0(h) xi0 + m k=1 hkzk0+∗0(h) :∀
(t,h)∈S0 = rx0,z0,t,TM − m r=1 tryr0+ m k=1 hkzr0−D x0,y0,h,TM :∀
(t,h)∈S0 ,withS0=⎧
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎩
(q,w) ∈Rm++m:min⎧
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎩
m r=1 qryr0 ıT1 , m k=1 wkzk0 ıT2⎫
⎪
⎪
⎪
⎪
⎬
⎪
⎪
⎪
⎪
⎭
≥1⎫
⎪
⎪
⎪
⎪
⎬
⎪
⎪
⎪
⎪
⎭
.Now,giventhat theinfimumofa setisthegreatestlowerbound
ofthatset,weseethatinf
t,h
rx0,z0,t,TM − m r=1 tryr0+ m k=1 hkzr0−D x0,y0,h,TM : (t,h)∈S0 ≥→Bx0,y0,z0;TM , whichistheinequalitythatwewereseeking.Let (q,w) ∈Rm+++mbemarketpricesforgoodandbadoutputs,respectively.Then,(˜q, ˜w)= (q,w)
min{ m
r=1 qr yr0 ıT1 , m k=1 wkzk0 ıT2 } ∈S0= {(q,w)∈Rm+m+ :min{ m r=1 qryr0 ıT1 , m k=1 wkzk0 ıT2 }≥1}. Consequently,applyingProposition1,wegetr
x0,z0, ˜q,TM − m r=1 ˜qryr0+ m k=1 ˜ wkzr0−D x0,y0, ˜w,TM ≥ inf t,h rx0,z0,t,TM − m r=1 tryr0+ m k=1 hkzr0−D x0,y0,h,TM : (q,h)∈S0 ≥→Bx0,y0,z0;TM . (17)Thisinequalitywillbeusefulfor statingthedesiredrelationshipbetweeneconomicenvironmentalinefficiencyand
→
B
x0,y0,z0;TM .Finally,giventhatr
x0,z0,t,TMisafunctionhomogeneousofdegree+1intandD
x0,y0,h,TMisafunction homo-geneousofdegree+1inh,then
rx0,z0,q,TM − m r=1 qryr0 +m k=1 wkzr0−D x0,y0,w,TM min
⎧
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎩
m r=1 qryr0 ıT1 , m k=1 wkzk0 ıT2⎫
⎪
⎪
⎪
⎪
⎬
⎪
⎪
⎪
⎪
⎭
≥→Bx0,y0,z0;TM . (18)Notethattheleft-handsideof(18)maybeinterpretedasa(normalized)measureofeconomicenvironmentalinefficiency. Additionally,followingFarrell’stradition,theright-handsidecanbeinterpretedas(environmental)technicalinefficiency andtheresidualtermassociatedwithclosingtheinequalitycouldbeinterpretedasallocativeinefficiency.Moreover,itis possibletodecomposetheleft-handsideof(18)into
r
x0,z0,q,TM − m r=1 qryr0+ m k=1 wkzr0−D x0,y0,w,TM min⎧
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎩
m r=1 qryr0 ıT1 , m k=1 wkzk0 ıT2⎫
⎪
⎪
⎪
⎪
⎪
⎬
⎪
⎪
⎪
⎪
⎪
⎭
OverallInefficiency = = rx0,z0,q,TM − m r=1 qryr0 min
⎧
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎩
m r=1 qryr0 ıT1 , m k=1 wkzk0 ıT2⎫
⎪
⎪
⎪
⎪
⎪
⎬
⎪
⎪
⎪
⎪
⎪
⎭
(Good)RevenueInefficiency + m k=1 wkzr0−D x0,y0,w,TM min
⎧
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎩
m r=1 qryr0 ıT1 , m k=1 wkzk0 ıT2⎫
⎪
⎪
⎪
⎪
⎪
⎬
⎪
⎪
⎪
⎪
⎪
⎭
Eco-DamageInefficiency . (19)
However,notethatthenormalizationtermusedin(18)and(19)−thatis,min
⎧
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎩
m r=1 qryr0 ıT1 , m k=1 wkzk0 ıT2⎫
⎪
⎪
⎪
⎪
⎬
⎪
⎪
⎪
⎪
⎭
−dependsontwodifferentterms,incontrasttowhathappenswithrespecttotheNerlovianprofitinefficiencymeasurein(10).Thismeansthat,
dependingontheobserveddataforeachDMU,overallinefficiencyisnormalizedbyeither
m
r=1 qryr0 ıT1 or m k=1 wkzk0 ıT2 .Inother words,inthesamesample,theDMUscouldusedifferentnormalizationfactorsfortheirmeasureofoverallinefficiency. Somethingthatmakesdifficultthecomparisonofresults.Hence,byanalogywiththestandardapproachbasedonthe directionaldistancefunction,wesuggestresortingtoanendogenousvalueforıT1and,therefore,alsoforıT2=1−ıT1,suchthat m
r=1 qryr0 ıT1 = m k=1 wkzk0ıT2 .Inotherwords,thevalueofthisendogenousıT1makesthetwocomponentsbeequal.Itiseasyto checkthatthisvalueisıT1∗=
m
r=1 qryr0/ m r=1 qryr0+ m k=1 wkzk0 .3.2. EconomicinefficiencymodelconsideringDakpoetal.’s(2017)approach
WenowturntoDakpoetal.’s(2017)approach.Inthiscase,theprojectionpointsinthetwosubtechnologiesfortheinput dimensionsmustcoincide.The“by-production”directionaloutputdistancefunctionunderthisapproachisasfollows:11
→ B
x0,y0,z0;TD =max ıT1ˇT1+ıT2ˇT2 s.t. p j=1 j0xij≤xi0, i=1,...,n1 (20.1) p j=1 j0xij≤xi0, i=n1+1,...,n2 (20.2) − p j=1 j0yrj+ˇT1yr0≤−yr0, r=1,...,m (20.3) p j=1 j0=1, (20.4) − p j=1 j0xij≤−xi0, i=n1+1,...,n2 (20.5) p j=1 j0zkj+ˇT2zk0≤zk0, k=1,...,m (20.6) p j=1 j0=1, (20.7) − p j=1 j0xij+ p j=1 j0xij≤0, i=n1+1,...,n2 (20.8) ˇT1,ˇT2,j0,j0≥0 (20.9) (20)Model(20)islikemodel(8),wheretheDakpoetal.’s(2017)approachhasbeenconsideredthroughconstraint(20.8), whichinterconnectstheprojectionsofthepollution-generatinginputsinT1andT2.
11Constraints(20.2)and(20.5)implythat− p
j=1 j0xij+ p j=1j0xij≥0,foralli=n1+1,...,n2.Thisinequality,togetherwith(20.8),implies p
j=1 j0xij= p j=1j0xijforalli=n1+1,...,n2,whichcoincideswiththeconstraintrelatedtoDakpoetal.’s(2017)approach.Weprefertoinclude(20.8)insteadof p
j=1 j0xij= p j=1Itslineardualis: → B (x0,y0,z0;TD)=min n1
i=1v
1 i0xi0+ n2 i=n1+1v
1 i0xi0− m r=1 u1r0yr0+˛10+ − n2 i=n1+1v
2i0xi0+ m k=1 u2k0zk0+˛20 s.t. n1 i=1v
1 i0xij+ n2 i=n1+1v
1 i0xij− m r=1 u1 r0yrj+˛10+ (21.1) − n2 i=n1+1 i0x ij≥0,j=1,...,p m r=1 u1r0yr0≥ıT1, (21.2) − n2 i=n1+1v
2i0xij+ m k=1 u2k0zkj+˛20+ n2 i=n1+1 i0xij≥0,j=1,...,p (21.3) m k=1 u2 k0zk0≥ıT2, (21.4)v
1 i0,v
2 i0,u 1 r0,u2k0, i0≥0, (21.5) ˛1 0,˛20free (21.6) (21)Models(20)and(21)arerelatedbythetheoryofLinearProgramming.
Inthiscontextwenowdefineanewsupportfunction,representingprofitinDakpoetal.’smodel,as
x0,q,w,TD : x0,q,w,TD =max m r=1 qryr− m k=1 wkzk s.t. p j=1 j0xij≤xi0, i=1,...,n1 (22.1) p j=1 j0xij≤xi0, i=n1+1,...,n2 (22.2) − p j=1 j0yrj+yr≤0, r=1,...,m (22.3) p j=1 j0=1, (22.4) − p j=1 j0xij≤−xi0, i=n1+1,...,n2 (22.5) p j=1 j0zkj−zk≤0, k=1,...,m (22.6) p j=1 j0=1, (22.7) − p j=1 j0xij+ p j=1 j0xij≤0, i=n1+1,...,n2 (22.8) yr,zk,j0,j0≥0, (22.9) (22)whichmaximizesthedifferencebetweenprivaterevenueandeco-damagecostsinourby-productioncontext.Notethat theDakpoetal.’s(2017)approachhasbeenconsideredinmodel(22)throughconstraint(22.8).
Thelineardualof(22)is:
x0,q,w,TD =min n1 i=1 ci0xi0+ n2 i=n1+1 ci0xi0+ 0+ − n2 i=n1+1 ei0xi0+0 s.t. n1 i=1 ci0x ij+ n2 i=n1+1 ci0x ij− m r=1 dr0y rj+ 0− n2 i=n1+1 ai0x ij≥0, (23.1) j=1,...,p, dr0≥qr, (23.2) − n2 i=n1+1 ei0x ij+ m k=1 fk0z kj+0+ n2 i=n1+1 ai0x ij≥0, (23.3) j=1,...,p, fk0≤wk, (23.4) ci0,d r0,ei0,fk0,ai0≥0 (23.5) 0,0free. (23.6) (23)
ByLinearProgramming,theoptimalvaluesofmodel(22)and(23)arerelated. Next,weshowarelationshipbetween
x0,q,w,TD and→Bx0,y0,z0;TD . Proposition2. LetıT1,ıT2 >0. Then, inf t,h⎧
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎩
x0,t,h,TD − m r=1 tryr0+ m k=1 hkzr0:min⎧
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎩
m r=1 tryr0 ıT1 , m k=1 hkzk0 ıT2⎫
⎪
⎪
⎪
⎪
⎪
⎬
⎪
⎪
⎪
⎪
⎪
⎭
≥1⎫
⎪
⎪
⎪
⎪
⎪
⎬
⎪
⎪
⎪
⎪
⎪
⎭
≥→Bx0,y0,z0;TD .Proof.FollowingthesamestepsthaninProposition1,wegetthedesiredresult. ApplyingProposition2,withmarketprices (q,w),wegetthefollowinginequality.
x0,q,w,TD − m r=1 qryr0− m k=1 wkzk0 min
⎧
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎩
m r=1 qryr0 ıT1 , m k=1 wkzk0 ıT2⎫
⎪
⎪
⎪
⎪
⎬
⎪
⎪
⎪
⎪
⎭
≥→Bx0,y0,z0;TD . (24)Theleft-handsidein(24)maybeinterpretedasameasureofeconomicenvironmentalinefficiency,whichcouldbe decomposedintotechnicalinefficiency(theright-handsidein(24))andaresidualterm,interpretedasallocativeinefficiency.
3.3. EconomicinefficiencymodelconsideringFørsund’s(2018)proposal
Finally,itispossibletoincorporateFørsund’s(2018)proposal,adaptingMurtyetal.(2012)andDakpoetal.(2017).To dothis,itissufficienttoincludethenon-pollutinginputsinthesub-technologyT2.TheresultsofProposition1and2are
validfor→B
x0,y0,z0;TMFand→B
x0,y0,z0;TDF.Hence,wehavemodel(25).
→ B
x0,y0,z0;TMF = max ıT1ˇT1+ıT2ˇT2 s.t. p j=1 j0xij≤xi0, i=1,...,n1 (25.1) p j=1 j0xij≤xi0, i=n1+1,...,n2 (25.2) − p j=1 j0yrj+ˇT1yr0≤−yr0, r=1,...,m (25.3) p j=1 j0=1, (25.4) − p j=1 j0xij≤−xi0, i=1,...,n (25.5) p j=1 j0zkj+ˇT2zk0≤zk0, k=1,...,m (25.6) p j=1 j0=1, (25.7) ˇT1,ˇT2,j0,j0≥0, (25.8) (25)Model(25)islikemodel(8),wheretheFørsund’s(2018)approachhasbeenconsideredbychangingi=n1+1,...,n2by
i=1,...,ninconstraint(25.5).And D
x0,y0,w,TMF = min ,z m k=1 wkzr s.t. p j=1 jxij≥xi0, i=1,...,n (26.1) − p j=1 jzkj+zk≥0, k=1,...,m (26.2) p j=1 j=1, (26.3) j≥0, j=1,...,p (26.4) zk≥0, k=1,...,m (26.5) (26) with rx0,z0,q,TM − m r=1 qryr0+ m k=1 wkzr0−D x0,y0,w,TMF min⎧
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎩
m r=1 qryr0 ıT1 , m k=1 wkzk0 ıT2⎫
⎪
⎪
⎪
⎪
⎬
⎪
⎪
⎪
⎪
⎭
≥→Bx0,y0,z0;TMF . (27)Model(26)allowstodeterminethedamagefunctionwhentheMurtyetal.(2012)approachisadaptedthroughFørsund’s (2018)proposal.Additionally, theleft-handsidein (27)maybeinterpreted asa measureofeconomic environmental inefficiency.Inparticular,itispossibletodecomposeitinto
r
x0,z0,q,TM − m r=1 qryr0+ m k=1 wkzr0−D x0,y0,w,TMF min⎧
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎩
m r=1 qryr0 ıT1 , m k=1 wkzk0 ıT2⎫
⎪
⎪
⎪
⎪
⎪
⎬
⎪
⎪
⎪
⎪
⎪
⎭
OverallInefficiency = = rx0,z0,q,TM − m r=1 qryr0 min
⎧
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎩
m r=1 qryr0 ıT1 , m k=1 wkzk0 ıT2⎫
⎪
⎪
⎪
⎪
⎪
⎬
⎪
⎪
⎪
⎪
⎪
⎭
(Good)RevenueInefficiency
+ m
k=1 wkzr0−D x0,y0,w,TMF min⎧
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎩
m r=1 qryr0 ıT1 , m k=1 wkzk0 ıT2⎫
⎪
⎪
⎪
⎪
⎪
⎬
⎪
⎪
⎪
⎪
⎪
⎭
Eco-DamageInefficiency . (28)
RegardingDakpoetal.’s(2017)model,includingFørsund’s(2018)extension,wehave: B(x0,y0,z0;TDF)=max ıT1ˇT1+ıT2ˇT2 s.t. p
j=1 j0xij≤xi0, i=1,...,n1 (29.1) p j=1 j0xij≤xi0, i=n1+1,...,n2 (29.2) − p j=1 j0yrj+ˇT1yr0≤−yr0, r=1,...,m (29.3) p j=1 j0=1, (29.4) − p j=1 j0xij≤−xi0, i=1,...,n (29.5) p j=1 j0zkj+ˇT2zk0≤zk0, k=1,...,m (29.6) p j=1 j0=1, (29.7) − p j=1 j0xij+ p j=1 j0xij≤0, i=n1+1,...,n2 (29.8) ˇT1,ˇT2, j0,j0≥0 (29.9) (29)And
x0,q,w,TDF =max m r=1 qryr− m k=1 wkzk s.t. p j=1 j0xij≤xi0, i=1,...,n1 (30.1) p j=1 j0xij≤xi0, i=n1+1,...,n2 (30.2) − p j=1 j0yrj+yr≤0, r=1,...,m (30.3) p j=1 j0=1, (30.4) − p j=1 j0xij≤−xi0, i=1,...,n (30.5) p j=1 j0zkj−zk≤0, k=1,...,m (30.6) p j=1 j0=1, (30.7) − p j=1 j0xij+ p j=1 j0xij≤0, i=n1+1,...,n2 (30.8) yr,zk,j0,j0≥0, (30.9) (30)whichresultsinthefollowinginequality:
x0,q,w,TDF − m r=1 qryr0− m k=1 wkzk0 min⎧
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎩
m r=1 qryr0 ıT1 , m k=1 wkzk0 ıT2⎫
⎪
⎪
⎪
⎪
⎬
⎪
⎪
⎪
⎪
⎭
≥→Bx0,y0,z0;TDF . (31)Inequalities(27)and(31)makeitpossibletodefinetechnicalandallocativetermsasdriversofthecorrespondingmeasure ofeconomicenvironmentalinefficiency.IntheempiricalapplicationwesolvethemodelscorrespondingtoMurtyetal.(2012)
andDakpoetal.(2017),enhancedwithFørsund’s(2018)proposal.Thisrepresentsatotaloffourmodels. 4. Empiricalapplication
4.1. Datasetandvariables
Theempiricalillustrationreliesonstate-leveldataintheUnitedStatesthatcomesfrommultipleagencies.Thedataset consistsofaggregatedfirmdataforeachstateas,unfortunately,andduetostatisticalconfidentialityreasons,wedonot haveaccesstotheindividualmicrodata.12ThemainsourceofdataistheU.S.DepartmentofAgriculture(USDA)Economic
12 Weperformamacro-levelanalysisinourempiricalapplication,assumingthattheDMUs(thestates)canbecompared.Amoresuitableanalysiswould
consistinestimatingameta-frontier(O’Donnelletal.,2008;Batteseetal.,2004)usingthedataforallthefirmsinallthestatesandthen,decomposing inefficiencyintowithin-stateinefficiencyandagapbetweenthetechnologyofeachstateandtheglobalfrontier.Thislinecouldbeagoodavenuefor furtherresearch.
Table1
Descriptivestatisticsofinput-outputdata(implicitrealquantities),2004.
Variable Mean SD Coefficientofvariation
Non-pollution-generatinginputs
Capitalservices(million$) 541.068 450.804 0.833 Landserviceflows(million$) 650.678 738.002 1.134 Laborservices(million$) 1,292.827 1,186.855 0.918 Pollution-generatinginputs
Energy(million$) 166.647 144.484 0.867
Pesticides(million$) 164.921 163.858 0.994
Goodoutputs
Livestockandproducts(million$) 2,103.076 1,997.443 0.950
Crops(million$) 2,819.480 3,419.130 1.213
Badoutputs
CO2emissions(tonsofCO2equivalents) 996,394.520 914,820.595 0.918
Pesticideexposures(number) 2,458.708 2,256.445 0.918 Notes:SD=Standarddeviation.
ResearchService(ERS),whichcompiledthedatanecessarytocalculateagriculturalproductivityintheUS,and,inparticular, thepriceindicesandimplicitquantitiesoffarmoutputsandinputsforeachofthe48continentalstatesfor1960−2004.The datasethasbeenvalidatedandusedextensivelyinpreviousresearch(forexample,inBalletal.,1999;ZofioandKnoxLovell, 2001;HuffmanandEvenson,2006;SabasiandShumway,2018).Acriticalreviewofthedatainlightofrecentdevelopments canbefoundinShumwayetal.(2015,2016).Toillustrateourmodels,weconsiderthemostrecentyearavailableinthe dataset(2004)andassumethattheproductionprocessischaracterizedbythefollowingthreenon-pollutinginputs(capital servicesexcludingland,landserviceflows,andlaborservices),twopollutinginputs(energyandpesticides),andtwogood outputs(livestockandcrops).ThepricesforthesevariablesweredirectlyobtainedfromtheERSdataset.Pricesindicesin theERSdatasetareconstructedusingtheTörnqvistformulation.13
Asfortheundesirableoutputproductiongeneratedbyenergyconsumption,weconsidercarbondioxide(CO2)emissions
fromtheagriculturalsectorassociatedwithfuelcombustion,alsofor2004(expressedintonsofCO2equivalents),obtained
fromtheU.S.EnvironmentalProtectionAgency(EPA).14ThepriceofCO
2emissionsisproxiedbythemarketclearingprice
setinthestateofCalifornia(priceofcarbonemissionsexpressedinthousandsofdollarspertonofCO2equivalents),since
ageneralmarketforCO2forthewholeUSdoesnotexist.Inparticular,thisisthepriceofcarbonfortradableallowances
withafuturescontractthatoriginatesfromtheCaliforniangreenhousegasestradingmarketundertheCaliforniaCapand TradeProgram,2019.Weconsidertheaverage2012priceanddeflateitto2004usingtheconsumerpriceindexinabsence ofasuitabledeflator(USBureauofLaborStatistics).15Themeasureofbadoutputrelatedtopesticidesisthenumberof
pesticideexposuresperstatefor2004obtainedfromtheCentersforDiseaseControlandPreventionattheUSDepartment ofHealth&HumanServices.Astheapproximationofthepriceofthisbadoutputweusethecostofhospitalizedtreatment ofpesticide-relatedpoisonings(inthousandsofdollars)asestimatedinPimentel(2005).Becausethiscostisprovidedfor 1995,wefurtherexpressitto2004pricesusingthepriceindexformedicalservicesasobtainedfromtheU.S.Bureauof LaborStatistics.16
Tables1and2summarizethedescriptivestatisticsofinput-outputquantitiesandtheircorrespondingprices,respectively, fortheUSstatesin2004.AppendixA.1.oftheonlinesupplementalmaterialaccompanyingthepaperpresentsthisdatafor eachstate.
4.2. Results
4.2.1. Technical,allocativeandprofitfrontiers
Whensolvingourfourreferenceeconomicmodels–thatis,Murtyetal.(2012)andDakpoetal.(2017),each com-plementedwithFørsund’s(2018)proposal−itisrelevanttodetermine,fromatechnologicalperspective,thenumberof
13ThedetailsonthemethodofconstructionofallvariablesarecontainedinthefollowingwebpageoftheUSDA-ERS:https://www.ers.usda.gov/
data-products/agricultural-productivity-in-the-us/methods/.
14Sincethesedataaregiveninoveralltermsforthewholecountry,wefurtherdisaggregateitbystate,usingforthatpurposethesharethateachstate
hasinfarmproductionexpensesforgasoline,fuels,andoils,asreportedbytheU.S.DepartmentofAgriculture,expressedinthousandsofdollars.
15California’sGHGemissionsprogramisthefourthlargestintheworldaftertheEuropeanUnion’sEmissionsTradingSystem,SouthKoreanEmissions
TradingSchemeandtheEmissionTradingSystemintheChineseprovinceofGuangdong.Fromitsbeginningin2012itcoveredthepowerandindustrial facilities,anditexpandedtonaturalgasandtransportationfuelsin2015,allowingittocoverapproximately85percentofCalifornia’sGHGemissions.
16Theconsiderationofthesemarketpricesforthebadoutputs,whichcanbeconsideredastheopportunitycostsofenvironmentaldamage,enablesthe
analysisofallocativeinefficiencywithrespecttotheenvironmentallyefficientfrontier.Marketpricescanbecomparedtotheshadowpricescorresponding tothemarginalrateoftransformationofbadoutputsforgoodoutputs,whichareobservedattheoptimalprojectionofthefirmonthetechnological frontier.Comparingbothsetsofpricescanhelpregulatorstoadoptspecificpollutingabatementinstrumentssuchastaxes,permits,oremissionstandards. Onthistopic,arecentandcomprehensivestudycomparingshadowpricecalculationsforcarbondioxideemissionsinChina,usingbothparametricand non-parametrictechniques,canbefoundinMaetal.(2019).
Table2
Descriptivestatisticsofpricesforinput-outputdata,2004.
Variable Mean SD Coefficientofvariation
Non-pollution-generatinginputs
Capitalservices(indicesrelativetoAlabama) 1.105 0.028 0.026 Landserviceflows(indicesrelativetoAlabama) 1.134 0.636 0.560 Laborservices(indicesrelativetoAlabama) 1.158 0.325 0.280 Pollution-generatinginputs
Energy(indicesrelativetoAlabama) 1.441 0.166 0.115 Pesticides(indicesrelativetoAlabama) 1.142 0.252 0.220 Goodoutputs
Livestockandproducts 1.245 0.197 0.158
Crops(indicesrelativetoAlabama) 1.067 0.156 0.146 Badoutputs
CO2emissions(thousand$pertonofCO2equivalents) 0.011 0 0
Pesticideexposures(thousand$) 6.222 0 0
Notes:SD=Standarddeviation.Pricesofinputsandgoodoutputsvaryacrossstatesandareexpressedinrelativetermswithrespecttothefirststate, Alabama,considering1996asthebaseyear.Pricesofbadoutputsareuniqueforallstatesandaredeflatedtothe2004referenceyear.
Table3
NumberofefficientUSstates.
Model Technical Allocative Profit
T T1 T2
Murtyetal. 6 19 11 4 4
Dakpoetal. 13 30 16 6 6
Murty&Førsund 10 19 24 4 4
Dakpo&Førsund 19 30 28 15 15
Fig.1. Averagetechnicalinefficienciesbymodels(absolutevalues).
observationsthatareefficient,therebydefiningthefrontieroftheglobalby-productiontechnologyT,consistingofboth theintendedproductionT1 technology,(4)(hereafter,conventionalorstandardtechnology)andthepollution-generating technologyT2,(5)(hereafter,pollutingtechnology).Table3showsthatthenumberofobservationsdefiningtheproduction frontierisgreaterintheconventionaltechnologyT1thaninthepollutingtechnologyT2,exceptinthecaseoftheMurtyetal. (2012)modelincorporatingFørsund’sproposal.Clearly,notalltechnicallyefficientstatesareallocative-efficientandthereby achieveprofitefficiency.California,Delaware,IowaandVermontareefficientinallmodels.Approximately10percentof USstates(fourorsixoutof48)arefullyefficient,exceptinDakpoetal.’sapproachenhancedwithFørsund’sassumption, wherethenumberofprofit-efficientstatesincreasesto31.25percent17.
4.2.2. Technicalinefficiency:resultswithinandbetweenmodels
Departingfromthisgeneralportraitofinefficiencyfrequenciesatthetechnical,allocative,andoverallprofitinefficiency levels,wenowfocusonthetechnologicalside,withFig.1portrayingtheaverageabsolutetechnicalefficiencyvaluesinT1,
T2andtheirglobalby-productionaggregateT,acrossthefourmodels.