|
PAGE 1
1
PAGE 2
2
PAGE 3
FirstofallIwanttothankmychaircommitteeDr.AlfonsoFlores-Lagunesforhisvaluablehelpanddenitelyforhispatiencebeingmyadvisor.IwouldalsoliketothankmymembercommitteeDr.CarmenCarrion-FloresnotonlyforthethesisbutalsoforthegreatcoursethatwehadinEconometrics.FinallyIwanttothankallFoodandResourceEconomicsDepartment(FRED)familythathelpedmeinmystudiesandmakingmylifebetterinGainesville. 3
PAGE 4
page ACKNOWLEDGMENTS .................................... 3 LISTOFTABLES ....................................... 6 LISTOFFIGURES ....................................... 7 ABSTRACT ........................................... 9 CHAPTER 1INTRODUCTION .................................... 10 2REVIEWOFTHEEKCLITERATURE ......................... 15 2.1Introduction ..................................... 15 2.2TheoreticalEKCDerivation ............................ 17 2.3TheoreticalBackground ............................... 18 2.3.1ArgumentsSupportingEKC ........................ 18 2.3.2CounterargumentsoftheEKC ....................... 20 2.4AlternativeRecentHypotheses ........................... 21 2.4.1Introduction ................................. 21 2.4.2RacetotheBottom ............................. 22 2.4.3NewToxicScenario ............................ 23 2.4.4RevisedEKC ................................ 23 2.5TheDevelopmentofEmpiricalStudies ....................... 23 2.6ResultsSummary .................................. 32 2.7EconometricDrawbacks .............................. 35 3DATA ........................................... 38 3.1DataSelectionDrawbacks ............................. 38 3.2DataSource ..................................... 39 3.3Advantages ..................................... 40 3.4DescriptiveStatistics ................................ 41 3.5TheNatureofEmissionsandRecentActionsAimedatTheirReduction ..... 44 3.5.1NitrogenOxide(NOx) 44 3.5.2SulfurDioxide(SO2) 45 3.5.3AcidRain .................................. 46 3.5.4CleanAirAct ................................ 46 3.5.4.1Acidrainactions ....................... 47 3.5.4.2Cars,trucks,buses,andnon-roadequipment ......... 49 3.5.4.3Permitsandenforcement .................... 52 4
PAGE 5
........................................ 53 4.1ConceptualFramework ............................... 53 4.2PanelDataMethods ................................. 54 4.2.1FixedEffectsModel(FE) .......................... 55 4.2.2RandomEffectsModel(RE) ........................ 56 4.2.3HausmanTest ................................ 59 4.3QuantileRegressionMethods ............................ 61 4.3.1General ................................... 61 4.3.2QuantileRegressionandOptimization ................... 62 4.3.3FixedEffectsforQuantileRegressionandtheApplication ......... 65 5APPLICATIONANDRESULTS ............................. 67 5.1Introduction ..................................... 67 5.2PanelDataMethodsfortheConditionalMean ................... 68 5.3QuantileRegressionMethodsforPanelData .................... 73 5.4Subsamples ..................................... 79 5.4.1Subsample1929-1984 ............................ 79 5.4.2Subsample1985-1994 ............................ 86 5.4.3ResultsDiscussion ............................. 92 5.4.3.1Quantilemethodsvsmethodsforconditionalmean ....... 92 5.4.3.2NOxvsSO2asaresponsevariable .............. 95 6SUMMARYDISCUSSION ................................ 97 6.1Discussion ...................................... 97 6.2FinalRecommendations .............................. 99 REFERENCES ......................................... 100 BIOGRAPHICALSKETCH .................................. 105 5
PAGE 6
Table page 2-1SummaryoftheSulfurSO2EKCstudies ........................ 33 2-2SummaryofEKCstudiesusingdifferentpollutants ................... 34 3-1DescriptiveStatistics ................................... 42 4-1Meanandmedianperformance .............................. 61 5-1FixedandrandomeffectsmodelcoefcientsforNOx 69 5-2FixedandrandomeffectsmodelsummarystatisticsforNOx 69 5-3FixedandrandomeffectsmodelcoefcientsforSO2 71 5-4FixedandrandomeffectsmodelsummarystatisticsforSO2 71 5-5HighquantilesforSO2 72 5-6QuantileregressioncoefcientsforNOx 75 5-7QuantileregressioncoefcientsforSO2 76 5-8FixedandrandomeffectsmodelcoefcientsforNOxusingsubsample1929-1984 ... 80 5-9FixedandrandomeffectsmodelsummarystatisticsforNOxusingsubsample1929-1984 ............................................ 80 5-10QuantileregressioncoefcientsforNOxusingsubsample1929-1984 ......... 81 5-11FixedanfrandomeffectsmodelcoefcientsforSO2usingsubsample1929-1984 ... 82 5-12FixedandrandomeffectsmodelsummarystatisticsforSO2usingsubsample1929-1984 ............................................ 82 5-13QuantileregressioncoefcientsforSO2usingsubsample1929-1984 .......... 84 5-14FixedandrandomeffectsmodelcoefcientsforNOxusingsubsample1985-1994 ... 86 5-15FixedandRandomEffectsModelSummaryStatisticsforNOxusingsubsample1985-1994 ............................................ 87 5-16QuantileregressioncoefcientsforNOxusingsubsample1985-1994 ......... 88 5-17FixedandrandomeffectsmodelcoefcientsforSO2usingsubsample1985-1994 ... 89 5-18FixedandrandomeffectsmodelsummarystatisticsforSO2usingsubsample1985-1994 ............................................ 89 5-19QuantileregressioncoefcientsforSO2usingsubsample1985-1994 .......... 91 6
PAGE 7
Figure page 1-1DifferentIncome-Pollutionrelationshipscenarios .................... 11 2-1EnvironmetalKuznetsCurve:DifferentScenarios .................... 23 3-1Incomeovertime ..................................... 42 3-2NOxconcentrationsovertime .............................. 43 3-3SO2concentationsovertime ............................... 43 5-1EKCxedandrandomeffectsforNOxusingfullsample.A)Fixedeffects.B)Ran-domeffects. ........................................ 70 5-2EKCxedandrandomeffectswithscatterplotforNOxusingfullsample.A)Fixedeffects.B)Randomeffects. ................................ 70 5-3EKCxedandrandomeffectsforSO2usingfullsample.A)Fixedeffects.B)Ran-domeffects. ........................................ 72 5-4EKCxedandrandomeffectswithscatterplotforSO2usingfullsample.A)Fixedeffects.B)Randomeffects. ................................ 72 5-5QuantileregressionforNOxusingfullsample ...................... 74 5-6QuantileregressionforNOxwithscatterplotusingfullsample ............. 74 5-7QuantileregressionforSO2usingfullsample ...................... 77 5-8QuantileregressionforSO2withscatterplotusingfullsample ............. 77 5-9FixedandrandomeffectsforNOxusingsubsample1929-1984.A)Fixedeffects.B)Randomeffects. ...................................... 80 5-10QuantileregressionforNOxusingsubsample1929-1984 ................ 82 5-11FixedandrandomeffectsforSO2usingsubsample1929-1984.A)Fixedeffects.B)Randomeffects. ...................................... 83 5-12QuantileregressionforSO2usingsample1929-1984 .................. 83 5-13FixedandrandomeffectswithscatterplotforNOxusingsample1985-1994.A)Fixedeffects.B)Randomeffects. ................................ 87 5-14QuantileregressionforNOxusingsubsample1985-1994 ................ 89 5-15FixedandrandomeffectswithscatterplotforSO2usingsample1985-1994.A)Fixedeffects.B)Randomeffects. ................................ 90 7
PAGE 8
.................. 90 5-17QuantileregressionsforSO2usingsample1985-1994 .................. 90 5-18CurvesfromquantileregressionandxedeffectsmodelforconditionalmeanforNOx 5-19CurvesfrommedianregressionandxedeffectsmodelforconditionalmeanforNOx 93 5-20CurvesfromquantileregressionandxedeffectsmodelforconditionalmeanforSO2 94 5-21CurvesfrommedianregressionandxedeffectsmodelforconditionalmeanforSO2 95 8
PAGE 9
TheenvironmentalKuznetscurve(EKC)estimatestheincome-environmentaldegradationrelationship,typicallyemployingmeasuresofper-capitaincomeandtheconcentrationofpollutantsintheair,suchasNOx(nitrogenoxide)andSO2(sulfurdioxide).TheliteraturehasconcentratedonestimationoftheEKCatthemeanemployinglongitudinaldataoncountriesorU.S.states.ThetypicalndingisaninvertedU-shapedrelationship,implyingthatpollutionisincreasinginincomeuptoaturningpointbeyondwhichpollutiondecreases.Estimationatthemean,however,likelymasksheterogeneitiesthatcanbepresentathigherand/orlowerquantilesoftheemissions'distribution.ThisstudyappliesmethodsforquantileregressionestimationofpanelxedeffectsmodelstotheestimationoftheEKConU.S.state-leveldataonNOxandSO2pollutantsovertheperiod1929-1994.OurresultsindicatethatmethodsthatfocusonthemeanprovidetoooptimisticestimatesaboutemissionsreductionofNOx,asquantilemethodsrevealthattherateofreductionisusuallysmaller;whiletheoppositeholdsforSO2.Thedifferencesariseduetotherobustnessofquantileregressiontooutlyingobservations.Theseresultshaveimplicationsforpoliciesadvocatingeconomicdevelopmentasameansforimprovingtheenvironment. 9
PAGE 10
Caneconomicdevelopmentleadtoimprovedenvironmentalquality?TheEnvironmentalKuznetsCurve(EKC)hypothesistriestoanswerthisquestionintermsoftheincome-pollutionrelationship.AccordingtoBeckerman(1992)( Beckerman 1992 ),Awaytoattainadecentenvironmentinmostcountriesistobecomerich.ThishypothesisisgraphicallypresentedbytheEKC(seeFigure 1 A),wheretherelationshipbetweenpollutionandincomefollowsaninvertedUshape.Attheearlystagesofacountry'sdevelopmentboththeaverageincomeandthepollutionincrease.However,afteracertainlevelofincomeisattained,thepercapitapollutionstartstodecrease. TheoriginalKuznetscurveexpressedSimonKuznets'stheorythatwhenacountry'sgeneralincomebeginstoincrease,incomeinequalitybeginstodecrease( KuznetsandSimon 1955 ).TheEKCexpressesthesametheorybutreplacesinequalitywithpollution.Despitethefactthatthedebateregardingtherelationshipofeconomicgrowthandenvironmentalqualityhasexistedsincethelatesixties,theEKCrstappearedandwaspopularizedintheearlyninetieswithbothGrossmanandKrueger's(1991)( GrossmanandKrueger 1991 )studyoftheenvironmentalimpactsoftheNorthAmericanFreeTradeAgreement(NAFTA)andShakandBandyopadhyay'sstudy( ShakandBandyopadhyay 1992 ),whichwasastudyuponwhichthe1992WorldBankDevelopmentReportwasbased.However,theEKClabelrstcametolightinPanayotou's(1993)study( Panayotoy 1993 ).Beforethesestudiesthebroadbeliefwasthateconomicgrowthandenvironmentalimprovementweremutuallyexclusive( Meadows,Meadows,Randers,andBehrens 1972 ).Thus,theappearanceoftheEKCdramaticallychangedtheinitialviewpointoftheenvironmentalimpactsofeconomicgrowthandhadimportantinuenceinpolicydecisions. TheEKChypothesisismostlybasedontheassumptionthatpeoplehavetheluxurytocaremoreabouttheenvironmentalqualityoncetheyattainagivenstandardofliving.Forpolicymakers,theEKCisimportantsinceitenablesthemtoestimatethelevelofincomeat 10
PAGE 11
DifferentIncome-Pollutionrelationshipscenarios whichenvironmentalqualityimprovementsappearor,alternatively,thelevelofenvironmentaldegradationthatwouldoccurduringeconomicdevelopment.Incaseswhereenvironmentaldegradationexceedsacertainlevel,theenvironmentalorpublichealthconsequencescouldbeirreversible.Insuchsituations,EKCstudiescombinedwithanenvironmentalstudyforthepollutant'simpactscancontributetoenvironmentalpolicydecisionsthatavoidsuchincidents. However,questionshavebeenraisedaboutthevalidityoftheEKCanditsabilitytoproducevalidestimatesforeverycountryandpollutant.TheEKCisconsideredbysomeasanoverlyoptimisticpointofviewofeconomicdevelopmentanditisthusnotembracedbytheentirescienticcommunity.ManyenvironmentalistsandscientistsarguethattheEKCdoesnotfollowaninvertedUcurveorthatthelevelofincomeatwhichthemaximumemissionslevelisattainedhasnotbeenexperiencedbydevelopedcountriesyet.ThereareatleasttwoothertheoriesabouttheEKCshape.TherstsupportsthattheEKCfollowsanNshape(seeFigure 1 B)andthesecondsupportsthatemissionsisamonotonicallyincreasingfunctionofincome. Giventheoftencontroversialnatureofempiricalstudies,ithasbeenhardtocorroborateorrefutethecompetingtheories.ThereareempiricalstudiessupportingtheinvertedUshape,as 11
PAGE 12
Ontheotherhand,EKChypothesissupportersarguethatthefactthatsomepollutantsdonotfollowaninvertedUcurvemighthappenbecausewehavenotyetreachedthelevelofincomethatwillcauseenvironmentimprovementforeverypollutant,andthustherelationshipwilleventuallybeaninvertedUasincomeincreases.However,therearestudies,e.g.( Cole,Rayner,andBates 1997 ; SuriandChapman 1998 )thatuseasaproxyindicatorofenvironmentalimpactstheenergyconsumptionpercapitatotesttherelationshipbetweenincomeandgeneralenvironmentalquality.Althoughtheirresultsshowamonotonicallyincreaseofenergywithincome,thisfactalonecannotdenitelysupportaninvertedUshapebutitcannotruleoutthepossibilityeither.Accordingtomostoftheempiricalresults,thepollutantsthatappeartohaveaninvertedUshapeincludesulfurdioxide(SO2),nitrogenoxide(NOx),Dichloro-Diphenyl-Trichloroethane(DDT)andleadamongotherpollutants.WhilethepollutantsthatoftenfailtoshowaninvertedUshapearecarbondioxide(CO2)andpollutantsrelatedtolandresourceorenergyusage. OpponentsoftheEKChypothesishaveadifferentconceptionoftherelationshipbetweeneconomicandenvironmentalimprovementandpointtotheeconometricweaknessesoftheEKCapplications.Theirmajorconceptualargumentisthatsincethereisachangeintheallocationofproducionactivitieswhileacountrydevelops,changesinenvironmentaldegradationmightbeduetotheseallocationchanges.Activitiespreviouslytakingplaceindevelopedcountrieshavenowbeentransferredtootherlessdevelopedcountries.Asaresultpollutionindevelopedcountriesfallwhileglobalpollutionincreaseswithincome.Thisprocesscannotcontinueforeversincelessdevelopedcountrieswillnotalwaysbeabletondotherlessdevelopedcountriestowhichtheycanexporttheirpollution. 12
PAGE 13
SeldenandSong 1994 ).WereviewthemajorconceptualargumentsforandagainsttheEKCinthenextchapter. TheeconometriccritiqueoftheEKCempiricalestimationsisbasedonproblemswithheteroskedasticity,simultaneity,omittedbias,andcointegration( Stern 2004 ).Inaddition,itisarguethattherelationshipbetweenincomeandpollutionisspuriousandthereisnocausalitythatcanbeinferred.However,theEKCproponentsdonotclaimtoestimatecausality,butrathertherelationshipbetweenincomeandemissions.Additionally,EKCopponentsarguethatincludingexplanatoryvariablesthatareindirectlyrelatedtoeconomicdevelopment,suchasoutputallocation,technology,education,orenvironmentalregulation,underminetheincome-pollutionrelationship[see,e.g.( Holtz-EakinandSelden 1995 )]. AcommoncharacteristicbetweentheproponentsandthedetractorsoftheEKCisthatbothexaminetherelationshipofincomeandthemeanpollutantlevel.Sofar,thereisnoworkthatexaminestherelationshipbetweenincomeandpollutionquantiles.Inmanycases,pollutionquantilesareofhigherinterestthanthemeanvalue.Therefore,ananalysisofwethertheinvertedUshapeatthemeanalsoappearsatthemedianorotherquantilesprovidesabetterunderstandingoftheincome-pollutionrelationship.Ahighemissionlevel,whichhappensinhighquantiles,mighthaveirreversibleenvironmentalimpactsorcausehealthproblemsthatahighmeanmightnotreveal.Findingtherelationshipbetweenseveralpollutionquantilesandincomewillhelpusexploretheincome-pollutionrelationshipinmoredetail,thuscontributingtomoreefcientenvironmentalpolicy.IftheconvexityofEKCweresensitivetoincomelevelsforexampleifhighincomecontributestohighconvexityinlowquantilesbutnotinhighquantilestheneconomicdevelopmentmightnotbeasbenecialasthemeanrelationshipasserts.Ontheotherhand,iftherelationshipisatterinlowquantilesandthehighquantilesshowahighconvexcurve,thenhighincomelevelsmightleadtoenvironmentalimprovementsevenifatthemean 13
PAGE 14
Thechoiceofdatahasalsoreceivedcriticismintheliterature.Inprinciple,tocapturetheentireEKCweneeddatathatbeginsfromareasonablyearlystageofdevelopmentandcoverawiderangeofincome.Additionally,thedataneedstohaveenoughobservations.Previousstudieshaveusedcross-country,paneldata.However,thesepaneldatasetsareconsideredofquestionablequality( ListandGallet 1999 ).ForthisinvestigationweuseUSpaneldataatthestatelevelcoveringtheyears1929-1994fortheNOxandSO2emissions.Thedatacontains48states.Inadditiontothequalityadvantagethatappearstoexistinthisparticulardataset( Millimet,List,andStengos 2003 ),therearetwoadditionaladvantages.First,itcontainsemissionsfortheentirestateandnotsolelythepollutionofurbanareas.Thisfeatureavoidsunderestimatingpollution,sinceitisempiricalynotedthatalongwithdevelopmentoccursdecentralization,whichreducestheemissionsinurbanareasandallocatesthemtotherestoftheregion.Second,wedonothavetodealwithexchangerateissuessinceweusedataonlyfromonecountry. Thisthesisisorganizedasfollows:Thenextchaptercontainsaliteraturereviewofanumberofstudies,theircontributiontotheliteratureandtheirweaknesses.WecovertheoreticalstudiesthathavebeendevelopedtoexplainthenatureoftheEKCaswellasthoseopposingtheEKC.WealsobrieyreviewpreviousliteraturethatnotesparticulareconometricweaknessesofEKCestimationandpresentasummaryoftheresults.Inthethirdchapterwediscusssomeaspectsaboutdatachoiceaswellassomeofourdataandcharacteristicsofemissions.Thefourthchapterdealswiththeconceptualframeworkandmethods,presentingthecontributionofthequantileregressiontotheEKChypothesis.Thefthchapterreportstheresultsoftheeconometricanalysisandprovidesadiscussiononthem.Finally,thisthesisconcludeswithabriefsummaryandnalrecommendationsforfutureresearch. 14
PAGE 15
BeforetheEKChypothesis,thegeneralbeliefabouteconomicgrowthwasquitedifferent.Therewasaconcernabouttheenvironment'sabilitytosustainthegrowththatmanycountiesexperiencedinthesixtiesandseventies.TheinitialstandpointofeconomicgrowthwassubmittedbytheClubofRome'sLimittoGrowthreport( Meadows,Meadows,Randers,andBehrens 1972 ).Accordingtothisreporttheconsumptionofrawmaterials,energyandnaturalresourceswasgrowingalmostatthesamerateaseconomicgrowth( DeBruynandHeintz 1999 ).Pollutionandconsumptionofrawmaterials,energy,andnaturalresourceswereincreasingsimultaneously.However,thisreportwascriticizedevenbeforethedevelopmentoftheEKChypothesis.Malenbaum( Malenbaum 1978 )showedempiricallythattherateofconsumptionofsomemetalswasdecreasingwiththeriseofincomeinsomedevelopedcountries.Thisempiricalresultcontradictedthemonotonicrelationshipbetweenincomeandresourceusage.Malenbaum'sstudywasfollowedbyotherstudiesthatshowedsimilarresultsforothermaterials( Williams,Larson,andRoss 1987 ; Janike,Monch,Ranneberg,andSimonis 1989 ; Tilton 1990 ). 15
PAGE 16
GrossmanandKrueger 1991 )oftheenvironmentalimpactsofNAFTA.ThisstudyalongwiththeShakandBandyopadhyaystudy( ShakandBandyopadhyay 1992 ),thelatterofwhichprovidedthefoundationfortheWorld'sBankDevelopmentReport1992,resultedinthepopularizationoftheEKC.Followingthosestudies,manyarticles[e.g.( GrossmanandKrueger 1995 ; ListandGallet 1999 ; Millimet,List,andStengos 2003 )]werepublished,supportingempiricallytheEKCtheoryforsomepollutants(mainlyairpollutants).Despitethefactthattherehasneverbeenanempiricalproofthatthisrelationshipholdsforeverypollutant,manyauthors[e.g.( Beckerman 1992 ; Lomborg 2001 )]arguethattheeconomicgrowthwillnallyleadtoenvironmentalqualityimprovement. Ontheotherhand,thereareopponentsoftheEKCtheoryprovidingtheoreticalargumentsagainstit[e.g.( Ansuategi,Barbier,andPerrings 1998 ; Arrow,Bolin,Cosnstanza,Dasgupta,Folke,Holling,Jansson,Levin,Maler,Perrings,andPimentel 1995 ; CopelandandTaylor 2004 ; Ekins 1997 ; Pearson 1994 ; Stern 1998 ; Stern,Common,andBarbier 1996 )],aswellasempiricalresultsthatdonotprovideevidenceofaninvertedUcurveorprovideaturningpointataveryhighincomelevelforsomepollutants[e.g.( SternandCommon 2001 ; TorrasandBoyce 1998 )].AnotherpointregardingtheEKChypothesisthatreceivesharshcriticismisthatstudiesappeartohaveawidevarietyofresultseveniftheyareabletoshowaninvertedUshape.InitsdefenseEKCproponentspostulatethatthedifferencescanbeascribedtothefollowingfactors:(i)somestudiesuseemissionswhileothersuseurbanconcentrationsasindicatorsofenvironmentalpressure,(ii)differentestimationmethodsused,(iii)differentdataused,(iv)differentmethodsemployedtotransferthenationalpercapitaincomedatatocomparablemonetaryunits,(v)theinclusionofdifferentexplanatoryvariablesotherthanincomeinordertoexplainsomepollutant'svariation( DeBruynandHeintz 1999 ). Despiteallthesedisagreements,thegeneralimpressionfromtheexistingliteratureisthattheempiricalstudiesareabletoprovideaninvertedUshapeforsomeairpollutantssuchassus-pendedaerosolparticulates,SO2,NOxandcarbonmonoxide(CO).Bycontrast,pollutantsthat 16
PAGE 17
Lopez'sarticle( Lopez 1994 )TheEnvironmentasafactorofproduction:Theeffectsofeconomicgrowthandtradeliberalizationtheoreticallyexaminestheincome-pollutionrelationshipandwhetherthepollutionimpactsareinternalizedornot.Additionally,heattemptstondtheimpactsoftradeliberalizationsontheenvironment.Lopezdividestheenvironmentalfactorsintotwocategories:onethatprovidesstockfeedbacktotheproducers(suchasshandforeststockandagriculturalsoilquality)andonethatdoesnot(suchasairquality).Heshowsthatincasewheretheresourcesprovidestockfeedback,iftheproducerinternalizestheseeffects(eitherbygovernmentpolicyinducementorprivateinternalization),thecombinationofeconomicgrowthandtradeliberalizationwilldecreasetheenvironmentaldegradation.Inthecasewhereresourcesdonotprovidestockfeedbackoneoftwocasesoccur:intherstcasethepreferencesarehomotheticandinthesecondtheyarenon-homothetic.Homotheticpreferencesimplyaunitincomeelasticityinenvironmentalqualityandsotheeconomicgrowthleadstoenvironmentaldegradation,buthearguesthatenvironmentaldegradationcannotcontinueforever.Thustheenvironmentalqualitywillbecomealuxurygoodandsothehomotheticpreferencesdonotprovideagoodapproximation.Inthesecondcaseheshowsthatnon-homotheticityundersomeconditionsissufcienttoprovidetheinvertedUcurveforthepollution-incomerelationship. Lopez'sarticle( Lopez 1994 )isbasedontheassumptionthatthepollutioniscausedbyproductionactivitiesandnotbyconsumption,whichisalsotheassumptionatworkinSeldenandSong'sarticle( SeldenandSong 1995 )NeoclassicalGrowth:TheJcurveforabatementandtheinvertedUcurveforpollution.SeldenandSongmodifyFoster'sgrowthmodeland 17
PAGE 18
JohnandPecchenino 1994 )theoreticallyexplorestheeffectsofoverlappinggenerations.Itderivesadynamicmodelwheretheyfocusmoreonexternalitiesthatarederivedbyconsumptionratherthanproduction.ThemodelderivesaninvertedUcurveforsomecasesbutitdoesnotexcludeamonotonicrelationshipbetweenincomeandpollutant.McConnell( McConnell 1997 )concentratesontheimpactofincomeelasticityonenvironmentalqualitydemand.Theoutcomeisthathigh-incomeelasticityisnotsufcienttoderiveanEKCbuteitherlow-incomeelasticity(greaterorequaltozero)isnecessaryforpollutionreduction.However,McConnellclaimsthathigh-incomeelasticityforenvironmentalqualitywouldhelpenvironmentalimprovement.Ontheotherhand,Lieb( Lieb 2001 )generalizesStokey'smodelandshowsthathigherincomeelasticitydoesnotprovidenecessarilyamorepositiveormorenegativeslopeinthepollution-incomerelationship.Additionally,Liebshowsthatassumingthatpollutioncomesfromconsumption,anecessaryconditionforenvironmentalimprovementisconsumptionsatiation.FinallyAndreoniandLevinson( AndreoniandLevisnon 2001 )argueaninvertedUcurvecanbeobtainedevenwithouttheassumptionsmadeinpreviouslyliterature( Stern 2004 ). DeBruynandHeintz 1999 ; Stern 2004 )].Sinceinmostempiricalapplicationstheestimationoftheincome-emissionsrelationshipisobtainedfromareduced-formequationthatcontainsonlyincomeasanexplanatoryvariable,theempiricalapplicationsdonotallowmakinganyinferenceaboutincomecausation.Hence,theargumentsabouttheEKChypothesisarebasedprimarilyontheory.InthenexttwosectionswediscusstheargumentsandthecounterargumentsoftheEKChypothesis,respectively. 18
PAGE 19
SeldenandSong 1994 ).Hence,itismorelikelythattheexpensesneededtoreducepollutionwilllikelycomefromhigh-incomecountries.Iftheelasticityisgreaterthanunity,thena1%increaseinincomewillincreasetheexpendituresonabetterenvironmentbymorethan1%.Thus,incomeincreasesatalowerratethantheexpensesforenvironmentimprovements.Theexpensescouldbeobtainedeitherthroughdonationstoenvironmentalorganizationsorbyconsumingfewerproductsthatharmtheenvironment( DeBruynandHeintz 1999 ). 2. 3. ShakandBandyopadhyay 1992 )examinedtherelationshipbetweenSO2concentrationsandcivilrightsmovementsandfoundthattheemissionsarehigherinlessdemocraticcountries( DeBruynandHeintz 1999 ).Onthecontrary,TorrasandBoyce( TorrasandBoyce 1998 )foundtheoppositeresultforlow-incomecountries( DeBruynandHeintz 1999 ). 4. Panayotou 1997 )provideevidencethatimposingenvironmentalfriendlyregulationswouldhelpbothlow-incomeanddevelopedcountriesinenvironmentalprotection.Ifenvironmentalfriendlyregulationsareconnectedwithincreasedincome,thenitisreasonabletobelievethathigherincomeleadstoenvironmentalimprovement.AccordingtoDasguptaetal( Dasgputa,Laplante,Wang,andWheeler 2002 ),therearethreereasonsthatcanexplainwhyhigh-incomecountriesimposestricterregulationsforenvironmentalprotection.First,aftermajorinvestmentshavebeenmadeinsectorslikehealthandeducationtheenvironmentbecomesahighprioritytogovernment.Second,adevelopedcountryhasthemeanstomonitorpollutionactivitiesandimposeregulations.Finally,higherincomeandeducationallowlocalcommunitiestoenforcehigherenvironmentalstandards,inadditiontofederalregulations. 19
PAGE 20
Stern 2004 ).Overall,economicdevelopmenthasapositiveimpactontechnology,whichinturnhasanimpactontheenvironmentbyusinglesspollutantinputsperunitofoutput.Anotherwayinwhichtechnologycanbenettheenvironmentisthereductionofemissionsperunitofoutput.TechnologyappearstobeanimportantfactorbehindtheinvertedUshapeoftheEKC. 6. Stern 2004 ).Thisarticlereferstomanypreviousstudiesanditcouldbeconsideredasaliteraturereviewitself.OtherarticlesthathavebeencontributingtothecriticismoftheEKCare( Stern,Common,andBarbier 1996 ; Stern 1998 ; SternandCommon 2001 )amongothers.Besidesthosearticles,thereareothersthatrevealpossibleweaknessesoftheEKCeitherbyprovidinganalternativemethodofestimationorbyusingpollutantsthatdonotconrmtheEKChypothesis.WewilldealwithsomeofthesearticlesintheEconometricDrawbacksandResultsSummarysectionsofthischapter.Forthemoment,welistsomeofthemaincounterargumentshere. 20
PAGE 21
FloresandCarson 1995 ; Komen,Gerkin,andFolmer 1996 ; KristomandRiera 1996 ))haveestimatedelasticitysmallerthanunit.Additionally,astudybyMcConnell( McConnell 1997 )showsthatelasticitygreaterthanunitydoesnotnecessarilyprovideevidenceofanEKC. 2. DeBruynandHeintz 1999 ).Furthermore,thestudiesthathaveinvestigatedthisinuencedohaveinconsistentoutcomes.Forexample,theShakandBandyopaohyay'sstudy( ShakandBandyopadhyay 1992 )whichimplicitlysupportstheideathatopenpoliticalsystemsarebenecialfortheenvironmentfoundthattheSO2concentrationsarehigherinmoredemocraticcountries. 3. 2.4.1Introduction Dasgputa,Laplante,Wang,andWheeler 2002 ),wediscussbrieyinthissectiontheracetothebottomandthenewtoxics(bothpessimistic)andtherevisedEKC(optimistic). 21
PAGE 22
2-1 alongwiththerestofthehypothesesthatwediscussinthissection. Dasguptaetal( Dasgputa,Laplante,Wang,andWheeler 2002 ),though,arguethatglobal-izationinfacthasapositiveimpactontheenvironment.Globalizationhasresultedintransferringconsumerandinvestorinuencetoincludecompaniesindevelopingcountries,pressuringthosecompaniestoadjustthestandardsofproductstheyproduceandthetypesofresourcesthattheyuse( Stern 2004 ).IftheRTBhypotheseswasaccurate,thenwiththeincreasedcapitalmobilityandliberalizationthathaveoccurredinpreviousdecades,thereshouldhavealreadybeenarelax-ationofenvironmentalregulations.Yetinsteadwehaveseentheopposite:ageneraltrendtowardstricterregulations( Dasgputa,Laplante,Wang,andWheeler 2002 ). 22
PAGE 23
EnvironmetalKuznetsCurve:DifferentScenarios Dasgputa,Laplante,Wang,andWheeler 2002 ). 23
PAGE 24
TheEKCdrewattentionaftertheGrossmanandKruegerstudy( GrossmanandKrueger 1995 )fortheenvironmentalimpactsofNAFTA.However,sincewearenotexamininganytradeeffectonpollutionimpacts,weconsideranarticlefrom1995thattheypublishedintheQuarterlyJournalofEconomicsentitled,titleEconomicgrowthandenvironment.InthisarticleGrossmanandKruegertrytocovermanydimensionsofenvironmentalquality.Thustheyusethefollowingfourpollutionindicators:urbanairpollution,thestateoftheoxygeninriverbasins,fecalcontaminationofriverbasinsandcontaminationofriverbasinsbyheavymetals.Foreachindicatortheyuseseveralpollutants.Forexample,instudyingairpollutiontheyestimateEKCforSO2andsuspendedparticleswhichtheydividedintotwocategories,heavyparticles,andsmoke,duetothedifferenthealthproblemsthattheycause.Theyusedifferentemissionsfortherestoftheindicators,thoughtheavailabledatadoesnotallowthemtoexaminemanyaspectsforotherpollutantindicators.ThesourceofthedataistheGlobalEnvironmentalMonitoringSystem(GEMS),whichcontainsairpollutionemissionsonlyforurbanareasfordifferentyearsandcities.Fortheremainingindicatorstheemissionscontainobservationsfordifferentyears.ThemethodologythattheauthorsuseisstraightforwardandsimilartotheoneusedfortheoriginalstudyoftheNAFTAimpacts.Theyestimatethereduced-formfortheincomeemissionsrelationshipequation.Thereduced-formequationisspecicallygivenbyequation( 2 )below.Accordingtotheauthorsanalternativemethodthatcouldbeusedisamodelwithstructuralequationsrelatingtoendogenousofincomevariablessuchasenvironmentalregulations,technologyetc.Theauthorssupportthechoiceofthismodelbasedontwoadvantages.Firstthereduced-formgivesthenetincomeeffectonemissionsconcentration,whilewiththestructuralmodelsomeonewouldberequiredtosolvebackinordertondthenet 24
PAGE 25
whereYitisameasureoftheemissionconcentrationinstationiandyeart,GitistheGDPpercapitainyeartinthestationi,GitistheaverageGDPpercapitaoverthepriorthreeyears,Xitisavectorofothercovariates,anditisanerrorterm.The'saretheparameterstobeestimated.IntheXitcovariatestheyincludealineartimetrendinordertoseparatetheeffectsthatareduetoglobaleffects,suchasglobaladvancesintechnologyandlocalincomeincrease.Othercovariatesthatwereincluded,inordertoreducetheresiduals,werevariablesthatwererelatedtothesiteofthemonitoringstations'locationandthemethodthatwasusedforthemonitoring.TheaverageGDPpercapitaforthepastthreeyearswasaddedbecausepastincomeissupposedtohaveagreatinuenceonenvironmentalbehaviorandsincethelagofGDPishighlycorrelatedwithincome,theauthorspreferredtousetheaverageofthepastthreeyears.AlmostallthepollutantsconrmtheEKChypothesis.TheairpollutionandspeciallytheSO2thatweareinterestedinhaveapeakatfourthousanddollarsbutattheveryhigh-incomeshowsthatitstartstoriseagain,thoughwecannotrelyonregionofincomesincetherewerefewobservationsatthislevelofincome.Similarlymostofthepollutantsshowapeakbetweenfourtoeightthousanddollarsbutsomeofthemshowariseagainatthehigherlevelsofincome. AnotherarticlethatconsiderablycontributedtotheempiricalEKCliteratureisaPanyotou'sarticle( Panayotou 1997 )withthetitle``DemystifyingtheenvironmentalKuznetscurve:turningablackboxintoapolicytool.ThisarticleattemptstoextendtheEKCestimationandidentifytheeffectsofenvironmentalpolicyandeconomicgrowthrate.Whileweexpectanunambiguouslypositiveeffectofenvironmentalpolicyonenvironmentalquality,theeffectofeconomicgrowthdependsonthemagnitudesofdemandandsupply,scaleandcompositionsfactors.Thesefactorscanhavecontradictoryresultsonenvironmentalquality.Therefore,in 25
PAGE 26
2 )belowandthethirdisgivenbytheequationillustratedin( 2 ).ThelatterwasusedtondexplicitlytheeffectofincomeonenvironmentaldegradationsoPanayotoudecomposedtheincomeintothefactorsthatwerepreviouslydescribed(scale,compositions,demandandsupply).Xit=0+YYit+YYY2it+YYYY3it+DDit+DDD2it+DDDD3it+GGit+GYGitYit+PPit+YPYitPit+tt+it 26
PAGE 27
KnackandKeefer 1995 ).Allmodelswereestimatedbyxedandrandomeffects.Totestwhetherthexedortherandomeffectsmodelwasthemostappropriate,theHausmantest( Hausman 1978 )wasemployed,whichultimatelyrejectedthehypothesisoftherandomeffectsinfavorofthexedeffectsmodel.Toavoidanheteroskedasticityproblem,generalizedleastsquares(GLS)wasemployedtoestimatethemodel.Additionallytheauthortriedtocheckthemodelformulticollinearityusingpartialcorrelationcoefcients.Nomulticollinearitywasdetectedfortheexplanatoryvariablesbuttherewasadetectionofsomelower-andhigher-ordertermsforsomevariables.Therstmodelwiththecubicincomeandpopulationdensityasex-planatoryvariablesconrmedtheinvertedUcurvebuttheR2was0.148,eventhoughtheincomecoefcientswhereallsignicantforat1%level.ThesecondmodelalsoconrmedtheinvertedUcurve,hadconsiderablehigherR2(R2=0:238)anddemonstratedthatthepopulationdensitywassignicant.Thecoefcientsforthepolicyandgrowthratewerehighlysignicantandhadtheexpectedsigns,whiletheinteractiontermsandthelineartimetrendwereinsignicantandthustheyweredropped.SimilartoGrossmanandKrueger( GrossmanandKrueger 1995 )thecurvestartstoturnupagainafter$15000(1985prices)butsincetherearefewobservationswiththathighincomeitcannotbeconsideredasasignicantoutcome.ThepeakoftheSO2concen-trationoccursat$5000(1985prices),whichiscomparablewithmanypreviousarticles[e.g.( GrossmanandKrueger 1995 ; Shak 1994 )].Theresultsusingthethirdmodelcharacterizedfromequation( 2 )showthattheemissionshaveanegativerelationwithGDPbutafter$13000itbecomespositive.However,likethepreviousmodelsitcannotbeconsideredsignicantforthesamereasonsandadditionallythecubiccomponentofincomeisinsignicant.TheincreaseoftheR2wassubstantial(itroseto0.502),whilethegrowthrateandthepolicycoefcientskepttheirsignicanceandsign.Discoveringthathighgrowthratesresultinhigherlevelsofenvironmentaldegradationasincomerisesisalsoanimportantnding.Thepolicyinteraction 27
PAGE 28
Anotherapproachtoestimatetheincome-pollutionrelationshipthatdiffersfromtheinitialstudieshasbeenattemptedbyS.M.deBruyn,J.C.J.M.vanBerghandOpschoorintheirarticleEconomicgrowthandemissions:reconsideringtheempiricalbasisofenvironmentalKuznetscurves( DeBruyn,denBergh,andOpschoor 1998 ).TheircritiqueabouttheinitialapproachessuchasthoseofShakandBandyopadhyay(1992),SeldenandSong(1994)andGrossmanandKrueger(1995)isthatincludingalineartimetrendinanequationwithacubictransformationofincomeandusingpaneldatafortheestimationwillnotnecessarilyprovidethecorrectEKCforeveryindividualcountry.Theirargumentsarebasedontheassumptionthatifthelineartimetrendissignicantandnegativeandiftheindividualcountry'stimetrendisdifferentthantheonethatthedataprovides,thentheestimatedEKCwillprovidewronginformationforthecountry'sEKCandthuspointpolicydecisionsinthewrongdirection.InthecasethatthetimetrendisinsignicanttheneveniftheresultsprovideaninvertedUshapeforthepanel,theindividualcountry'sEKCmightevendifferfromaninvertedUshapebecausethetimeeffectmightnotbelinearortheEKCmightnotbeshiftinginauniformdirection.ThususingpaneldatatoprovideanEKCforeverycountrymightprovidemisleadingresults.TotestempiricallytheirmodeltheyestimatedtheEKCforthreepollutants(CO2,NOx,SO2)infourcountries(Netherlands,UK,USAandWesternGermany).FindingmanysimilaritiesbetweenresourcesdemandandEKCtheoreticalbackground,theauthorsdevelopedamodelinspiredbyaresourceeconomicmodelfordemandofresourcesbasedontheintensitytousewhichfollowsaninvertedUcurveandisthussimilartotheEKChypothesis.Themodelbasedontheintensitytouseisrepresentedbythefollowingequation( 2 ): whereemissionspercapitaEitforcountryjandyeartareequaltotheproductofthelevelofGDP,YjtandtheemissionintensityofuseUjt(=Ejt=Yjt).Thuschangesovertimecanbeexplainedbythechangesinintensitytouseandchangesinincome.Becauseintensityofuseis 28
PAGE 29
2 )wehaveln(Ejt=Ejt1)=ln(Yjt=Yjt1)+ln(Ujt=Ujt1)andtheintensitytouseisnotconsideredasexogenous,theauthorsusedequation( 2 )fortheirmodel Similartothemodelsthatmostofthepreviousstudiesused,equation( 2 )isareduced-formequation.Theintensitytouseisexpressedby1j;2j;3jwhere1jexpressestheconstanttechnologicalandstructuralchanges,2jshowshowtheintensitytouseiscorrelatedwithincome(sinceYjt1isthelagofGDPpercapita),3jshowshowtheintensitytouseiscorrelatedwithenergyprices,sincePjtisanindexofpriceenergyandthecoefcient0jrepresentstheinuenceofeconomicgrowthonemissions.Themodelwasestimatedthreetimesforeachcase,onewithoutrestrictionsandtwobyimposing1j=0and2j=0respectively,andtheregressionwithbettertwasselected.Theresultsshowedagoodtforalmostallcases(excepttheNetherlandsforSO2emissions)sincetheR2wasbetween0.35and0.7,whiletheDurbin-Watson(DW)andLjung-BoxQ-testsat4and8lagswerewithinnormalcriticallevels.Theresultsforthecoefcientsshowedthat0jispositiveineverycase(exceptNetherlandsforSO2emissionswherethetwasprettylow).Thuseconomicgrowthhasapositivecorrelationwithemissionsgrowth.While0jdifferedsubstantiallyamongthecases,mosttimesitwasclosetounity,whichmeansthata1%growthrate,inducesa1%increaseinemissions.Ontheotherhand,theintensitytousehadanoverallnegativeeffectinallcases,counteractingthepositiveeffectofthegrowthrate.Incaseswheretheconstanteffectwasused,WesternGermanywastheonlycountrywithpositivecoefcientforNOxemission,whiletheincomecoefcientwasnegativeinallcaseswhereitwasused.Thepricecoefcientwassignicantat5%levelonlyinonecase. ThefactthatincomemighthaveadifferenteffectindifferentcountriesisatopicthatListandGalletexamineintheirarticleTheEnvironmentalKuznetsCurve:doesonesizetall?( ListandGallet 1999 ).Theauthorswanttodetermineiftheemissionsfollowthefamiliar 29
PAGE 30
wherePjitisthepercapitaemissionofpollutantjforstateiandyeart.Xisavectorofincometransformations(K=4isthecubictransformation),jiisthevectorofpotentiallyheterogeneouscoefcientsofthelineartimetrendTandjitistheerrorterm.Whileallthepreviousstudiesallowedonlyfordifferentintercepts,ListandGallet'sstudyallowsfordifferentcoefcientsacrossthestates,therebyavoidingtheheterogeneitybias.Moreover,byestimatingthemodelasasystemofequationsofthetwoemissions,theygainefciency.Additionally,equation( 2 )allowsfortime-specictimetrends,whichreducesunexplainedvariations.Tondifthecoefcientsshouldbeconsideredasrandomorxedeffects,theyemployedtheHausman( Hausman 1978 )test,whichrejectedtherandomeffectshypothesisinfavorofxedeffects.Thusthemodelwasestimatedasasystemusingtheseeminglyunrelatedregressions(SUR)method.Theresultsconrmedtheirhypothesissincehomogeneitytestsrejectedthetraditionalspecicationsinfavorofthemoregeneralmodelthatallowsthecoefcientstovary.TheEKCwasstillconrmedfromthemoregeneralmodel,buttheturningpointswerequitedifferent.Comparedwiththeresultsfromthetraditionalspecications,ingeneraltheSO2appearstohaveaturningpointatloweralevelofincome,whiletheturningpointforNOxappearstooccuratahigherlevelofincome.Theauthorsexploredpossiblereasonsforthisheterogeneitybytestingthedifferencesinvariablessuchaseducationlevel,populationdensityandnumberofheatingdegreedays.Theydiscoveredthatstateswhichhavelowerturningpointshavehigherpopulationdensitiesandwarmerclimates,thoughtheseresultscanonlybeconsideredaspreliminary. AnotherstudythatarguesthatthetraditionalmodelsemployedtoestimatetheEKCaremisspeciedwaspublishedbyMillimet,List&StengoswithtitleTheenvironmentalKuznets
PAGE 31
Millimet,List,andStengos 2003 ).InordertoestimatetheEKC,theauthorsemployedtwocommonlyusedparametricmodelsandasemiparametricpartiallylinearregression(PLR).Fortheparametricmodelstheyemployedthetraditionalcubictransformationofincomeandfortheotherapproachtheyusedasplineequationofthecubictransformation,inspiredbySchmalenseeetal( Schmalensee,Stoker,andJudson 1998 ).Thesemiparametricmodelhasnospecicfunctionalformofincome.Therefore,thesemiparametricmodelislessrestrictive.Thedisadvantageofthesemiparametricornonparametrictechniquesisthatsincetheymakefewerassumptionsabouttheobjectbeingestimated,theytendtobeslowertoconvergeonobjectsbeingstudiedthancorrectlyspeciedparametricestimators.Inaddition,unliketheirparametriccounterparts,theconvergencerateistypicallyinverselyrelatedtothenumberofvariablesinvolved.Thusnonparametricmethodsaresuitedinsituationswhereweknowlittleaboutthefunctionalform,thenumberofvariablesissmallandthedatasetisreasonablylarge;hencethissituationisidealforasemiparametricestimation.BasedonthosefactstheyappliedthePLRmodelfortheEKCestimation.Totesttheirhypothesis,theyusedaspecicationtestthatwasproposedbyZheng( Zheng 1996 )aswellasbyLiandWang( LiandWang 1998 ).Thetestshowedthatthenullhypothesisoftheparametricmodelwasrejectedinfavorofthesemiparametricmodel.Theresultsfromthisapplicationshowedthatparametricmodelsaremorepessimisticinthesensethattheturningpointhappensforahigherincomecomparedtotheoneestimatedbythesemiparametricmodel.ThegraphsshowedthatthedataseemstobetbetterbythesemiparametricmodelaccordingtotheEKC,especiallyfortheSO2.Though,usingatestforserialcorrelationcomestoaconclusionthatthePLRsuffersfromserialcorrelationwhichmakestheconclusionofthespecicationtestquestionablesincetheidenticallyindependentdistribution(i.i.d.)hypothesisisrejected. Finally,otherstudiesthatemploysemiparametricornonparametricmethodstoestimatetheEKCwasconductedbyFlores( Flores 2007 )andZapata,PaudelandMoss( Zapata,Paudel,andMoss 2008 ).Flores'studyconsidersnonparametricestimationofboththeEKCanditsturning 31
PAGE 32
Stern 2004 ).Thisexplanationisasuppositionthatappearsinmanyexistingstudiesbutthereisnoclearevidenceforit.Inthenextpartofthischapterwewillpresentresultsfromtheliteratureanddiscusssomeofthepracticalreasonsthatmighthaveleadtothem. Abroadlyexaminedpollutant,whichwillbeexaminedaswellasinthisstudy,isSO2.Table 2-1 ,whichwasprovidedbyStern( Stern 2004 ),summarizesseveralstudiesofSO2.ThelowestturningpointisprovidedbythePanayotou( Panayotoy 1993 )studyat$3,137,whilethestudyofSternandCommon( SternandCommon 2001 )projectsthehigherturningat$101,166.Table 2-2 ,whichwasprovidedbydeBruynandHeintz( DeBruynandHeintz 1999 )showstheresultsfromseveralstudiesforseveralemissionslikeNOxthatweexamineinthepresentstudy.Aswecansee,EKCisconrmedinbothstudiesforNOx,whilethathappensforthemajorityoftheotherpollutantsandstudies.Aswealreadymentionedthereisagreatvariationintheresultsasdemonstratedinbothtables.Somereasonsareasfollows. 1. Stern,Common,andBarbier 1996 ),arguingabouttheheteroskedasticityoferrors,usedgeneralizedleastsquares(GLS)inordertoestimatetheEKC.Studieshaveusedifferentmethodsarenotstrictlycomparablesinceitisnaturalforthemtoyielddifferentresults. 32
PAGE 33
SummaryoftheSulfurSO2EKCstudies Panayotou(1993)$3,137EmissionsNo-Ownestimates1987-8855developedanddevel-opingcountriesShak(1994)$4,379ConcentrationsYesTimetrend,loca-tionaldummiesGEMS1972-8847citiesin31countriesTorrasandBoyce(1998)$4,641ConcentrationsYesIncomeinequality,literacy,politicalandcivilrights.urban-ization.locationaldummiesGEMS1977-91Unknownnumberofcitiesin42countriesGrossmanandKrueger(1991)$4,772-5,695ConcentrationsNoLocationaldummies,populationdensity,trendGEMS1977,'82,'88Upto52citiesinupto32countriesPanayotou(1997)$5,695ConcentrationsNoPopulationdensity,policyvariablesGEMS1982-84Citiesin30developedanddevelopingcountriesCole,RaynerandBates(1997)$8,232EmissionsYesCountrydummy,technologylevelOECD1970-9211OECDcountriesSeldenandSong(1994)$10,391-10,620EmissionsYesPopulationdensityWRIprimarilyOECDsource1979-8722OECDand8develop-ingcountriesKaufmann,Davidsdottir,Garnham,andPauly(1998)$14,730ConcentrationsYesGDP/Area,steelexports/GDPUN1974-8913developedand10developingcountriesListandGaller(1999)$22,675EmissionsN/A-USEPA1929-94USStatesSternandCommon(2001)$101,166EmissionsYesTimeandcountryeffectsASL1960-9073developedanddevel-opingcountries Source:Stern(2004)
PAGE 34
SummaryofEKCstudiesusingdifferentpollutants AuthorsMethods(Ef-fects)SO2(peak)(through)type)Part(peak)typeNOxemis.(peak)CO2emis.(peak)FaecalColiform(peak)1/Dissolvedoxygen(peak)DeforestationExch.RatesAdditionalvariables GrossmanandKrueger,1995GLS(re)N(4100)(13000)conc.EKC(6200)conc.EKC(8000)EKC(2700)PPPLaggedincomeShakandBandy-opadhyay,1992OLS(fe)EKC(3700)conc.EKC(3300)conc.MIN(1200)(11400)MIatPPPVarietyofothervariablesPanayotou,1993OLS(pcs)EKC(3000)emis.EKC(4500)emis.EKC(5500)EKC(1200)MERSeldenandSong,1994GLS(re,fe)EKC(10300)emis.EKC(10300)EKC(11200)PPPPopulationdensityTorrasandBoyce,1998OLS(pcs)N(3400)(14000)conc.atatN(5100)(19900)PPPInequalityVariablesHoltz-EakinandSelden,1995OLS(fe)EKC(35400)PPP Source:deBruynandHeintz(1999) Notes: N=N-shapedcurve,U=U-shapedcurve,EKC=invertedU-shapedcurve,MI=monotonicallyincreasingcurve,at=allparame-tersexceptinterceptinsignicant.PeaksroundedatUS$100. GLS=generalizedleastsquares,OLS=ordinaryleastsquares,re=randomeffects,fe=xedeffects,pcs=pooledcrosssection Conc=concentrations,emis=emissions,PPP=purchasingpowerparity,MER=marketexchangerate Particlesdifferwithrespecttohowthesearebeingmeasured Dissolvedoxygenisanindicatorforenvironmentalquality,notdegradation.Andforthesereasonswetaketheinverseofdissolvedoxygen.HenceanEKCinfactreectsaU-.shapedcurveandthemonotonicallydecreasingpatternfoundbyShakandBandy-opadhyayreectscontinuousdeterioration. Turningpointsformodelswithpopulationdensity,forSO2usingrandomeffects,forparticlesandNOxusingxedeffects.
PAGE 35
GrossmanandKrueger 1995 ),theyusedasamplethatwasprovidedbyGEMS,whileShakandBandyopadhyay( ShakandBandyopadhyay 1992 )usedadatabasefromtheCanadianCenterforInlandWaters.TheresultwasthatthelatterstudyfoundanNshapeforfaecalcoliformwhiletheformerstudyfoundaninvertedUshape.Anotherdatasetthatwasbroadlyusedintheempiricalstudies,andisalsothedatasetthatthepresentstudyuses,isthepaneldatafromtheU.S.EnvironmentalProtectionAgency's(EPA)NationalAirPollutantEmissionTrends,whichcontainsobservationsofthe48statesoftheU.S. 3. 4. SeldenandSong 1994 ),laggedincome( GrossmanandKrueger 1995 ),tradevariables( ShakandBandyopadhyay 1992 ; TorrasandBoyce 1998 ),variablesrepresentingthestructureoftheeconomy( Lucas,Wheeler,andHettige 1992 ; SuriandChapman 1998 ; Kaufmann,Davidsdottir,Garnham,andPauly 1998 )andvariablesthatcapturetheeffectsofpoliticalandcivilrights( ShakandBandyopadhyay 1992 ; TorrasandBoyce 1998 ).Includingsuchvariableshascapturedpartoftheincome-relatedeffectthatotherstudieshaveoverlookedandhashenceledtoachangeintheturningpointsandtheincome-pollutionrelationship( DeBruynandHeintz 1999 ).Ontheotherhand,excludingthesevariablesfromthemodelhastheadvantageofestimatingthegrosseffectofincometothepollution.Thesevariableschangeatleasttosomeproportionduetoincome'sincreasethusincludingtheminthemodelwillresultinmissingthateffectasincome'seffect. Stern 2004 )]whilesomeofthemtrytoeliminatetheproblemsthattheEKCapplicationsface[e.g.( Stern,Common,andBarbier 1996 ; Millimet,List,andStengos 2003 )].TheEKCeconometriccriticismsarebasedonheteroskedasticity,omittedvariablebiasandcointegration.Inthispartofthechapterwediscussbrieytheseeconometricproblems,whichhavebeenalsodiscussedbyStern( Stern 2004 ). 35
PAGE 36
Stern,Common,andBarbier 1996 ).ThustheresidualisnegativelycorrelatedwiththetotalGDP.Evidenceofheteroskedasticityap-pearedinstudy( Schmalensee,Stoker,andJudson 1998 ).Inordertocorrectthisproblem,Stern( Stern 2002 )estimatedadecompositionusingfeasibleGLS.Inthisapplication,thegoodnessoftwassignicantlyimproved. 2. SternandCommon 2001 ).Theauthorsbasedtheirargumentsonthefollowingthreefacts:(a)differencesbetweenthexedandrandomeffectsmodelsusingtheHausmantest;(b)differentestimatesindifferentsubsamples;(c)existenceofserialcorrelation( Stern 2004 ).TheHausmantestconrmsthatthereisasignicantdifferencebetweenrandomandxedeffects.Thisindicatesthatthetermsofincomearecorrelatedwiththexedeffects.Consideringthatthexedeffectsexpressthemeaneffectoftheomittedvariables,itisverylikelythatincometermswillbecorrelatedwithomittedvariables,yieldingbiasedestimators.Theauthorsusedtwosubsamplesinordertoseeiftheyproducedifferentresults.Onesamplecontainednon-OECDcountriesandtheotherOECDcountries.Theestimatesfromthenon-OECDsampleprovidedaveryhighturningpoint,whiletheOECDsampleprovidedaturningpointwithintherangeofthesample.ThefactthatthedatacannotbepooledtogetherisconrmedfromaChowF-test,whichrejectsthishypothesis.Additionally,thearticleprovidesevidencethatthemodelsuffersfromserialcorrelation,indicatingmisspecicationofomittedvariablesormissingdynamics( Stern 2004 ).Anarticlethattriedtodealwithmisspecicationproblemis( Millimet,List,andStengos 2003 ),tobethatdiscussedpreviouslyinthesectionThedevelopmentofempiricalstudies. 3. Stern 2004 ).ThoughtherearesomestudiesthathavetriedtoidentifyifthereiscointegrationinEKCmodels,theresultsfromthesestudiesdonotprovideaclearcutpictureofthecointegration.Forexample,PermanandStern( PermanandStern 2003 )testforunitrootandcointegrationasdoSternandCommon's( SternandCommon 2001 )dataandmodelsrespectively.Thetestsforunitrootdemonstratethattheserieshavestochastictrends.Ontheotherhand,thetestsforcointegrationshowthataroundhalfoftheindividualcountriesregressionscointegrate,thoughsomeofthemhaveawrongsign.Usingsubsamples,theresultscontinuetogiveavaguepicture.Somerejectandsomefailtorejectthenoncointegrationhypothesis.However,incaseswherecointegrationisfound,besidestheshapeofEKCdifferingacrosscountries,thehypothesisofacommonintegratedvectorwasstronglyrejected.Inordertocorrectforthelackofcointegrationtheyusedtherstdifference.TheresultsshowedthattheEKCpredictedlargeturningpointsbutstilltheparameterswerestatisticallydifferentbetweenthegroupsofcountries.Additionally,DayandGrafton( DayandGrafton 2003 )usingCanadiantimeseriesdatafailedtorejectthehypothesisofnocointegrationinalmosteverycase.AnotherstudythatfailstogiveaclearansweraboutthepresenceofcointegrationisdeBruyn's( DeBruyn 2000 ).Usingthedata 36
PAGE 37
Thepicturethatwegetfromtheliteraturedoesnotallowustoarriveatanalconclusionabouttheappropriatenessofthemethodsusedthusfarandtheirabilitytoprovidetrustworthyresults.IntheEKCempiricalstudies,likealltheempiricalapplications,thereareproblemsthatwehavetoconsiderbeforemakinganalconclusionabouttheEKChypothesisaswellasthepositionoftheturningpoints(iftheyexist).DespitethenumerousstudiesthathaveexploredtheEKC,wearenotinapositiontosayiftheinvertedUcurveexists,atleastforeverypollutant.Thedataqualityandpollutantindicatorthathavebeenusedtoderivetheincome-pollutionrelationshipseemtobethemainreasonforthedifferentresults.Inthenextchapter,wediscussthemethodstobeusedinourstudyandthereasonsthatledustothisapproach. 37
PAGE 38
InEKCapplications,animportantissueisdataselection.WestartthischapterbydescribingthedrawbacksthatappearindataselectionofEKCstudiesandtheadvantagesofpaneldata,astheoverwhelmingmajorityoftheexistentEKCliteratureusespaneldata.Thissectionisfollowedbyadiscussionofthesourcesanddetailsofthedatathatisusedinthepresentstudy.Wecontinuewiththedescriptivestatisticssectionwherewepresentedsomediagrams,andwenishwithsomebasicinformationaboutthepollutantsthatareusedinthisstudy. 1. 2-1 .Becauseinmostcasestheprovinceorstateleveldatasetisnotavailable,andthetimeseriesareusuallynotlargeenoughforvaluableestimates1,researchersareconstrainedtousingcrosscountryorpaneldatainordertocollectenoughobservations. 2. DeBruyn,denBergh,andOpschoor 1998 )whichusestimeseriesforEKCestimation,including20to30observationsforeachtimeseriestheyuse. 38
PAGE 39
4. Stern 2004 ).Anotherdrawbackinthechoiceofthepollutantisthatnotallofthemfollowthesamepath.ThereisnospecicpollutantthatcanbeusedtotesttheEKChypothesis,yetusingdifferentpollutantleadstodifferentresults.Whilesomeattemptshavebeenmadetocaptureallenvironmentalimpacts( SuriandChapman 1998 ; Cole,Rayner,andBates 1997 ),theindicatorsusedinthesestudiesarebasedonenergyusewhichdoesnotnecessarilyfollowsaninvertedUcurvewithincome( Galli 1998 ).Thus,tosomeextent,theselectionofpollutantsdependsonthehypothesiswewanttotest.Butthetestresultsfromonepollutantcannotconrmthatotherpollutantswouldhavethesameresults. ListandGallet 1999 )andMillimet,ListandStengos( Millimet,List,andStengos 2003 ).TheEPA'sreportcontainsannualobservationsfrom1929to1994for48statesintheU.S..Forthepollutants,weuseemissionspercapitaofNOxandSO2in 39
PAGE 40
Hsiao 2007 ).Belowwepresenttwooftheadvantagesthatourstudygainsbyusingpaneldata. 1. Itcontainsmoresamplevariabilitysincebothcross-sectiondataandtimeseriesconstitutepaneldata.Thelargenumberofcross-sectionalunitsalongwiththesignicantrangeintimeallowforgreatervariability.Thusinferenceusingpaneldataismoreprecise. 2. Becausedifferentquantitativetechniquescanbeappliedtopaneldata,thistypeofdataofferstheopportunitytocapturemorecomplexeffects.Forexample,wecanusexedeffectstocontrolforsomeoftheimpactoftheomittedvariablesbysimplyaddingdummyvariablesforindividualsandtime. 40
PAGE 41
1. 2. 3. 4. Aswecanseefromgure 3-1 ,apartfromthetheeconomiccrisisoftheearly'30s,incomerisessteadily,reachingitsmaximumlevelinthelastyearofthestudy.Atthesametime,NOxandSO2donotseemtofollowtheoppositepathandreduceovertime.WhileNOxconcentrationwasdecreasingovertime,apartfromthesecondhalfofthe30'swhenitincreasedsubstantially,SO2wasincreasingovertimeuntilthelastdecadesinwhichitseemstohave 41
PAGE 42
DescriptiveStatistics VariableMin1stQuartileMedianMean2ndQuartileMaxStdDev Income1,1625,8498,4369,08912,38022,4604,241.5NOx0.02300.05140.07590.09280.10661.13600.0735SO20.00210.05920.09680.16470.18401.61800.2060 Incomeovertime stabilizedapartfromtheyear1988.FromthediagramswecaninferthatU.S.SO2concentrationcontainsmoreoutliersthanNOx.Fromtable 3-1 ,wecanseethatincomeobservationscontainawiderangefrom$1,162to$22,460in1987dollarswhichhelpsusmakesafeinferencesforthatrangeofincome.Inaddition,wecanseethatthereislargevariationnotonlyofincomebutofemissionstoo,whichwillhelpourmodelprovidepreciseinderence.Missingobservationsregardingincomelevelslowerthan$1162resultinlossofimportantinformationabouttheincome-emissionrelationshipsinceonethirdoftheworld'scountrieshavelowerthan$1100incomepercapita( Millimet,List,andStengos 2003 ) 42
PAGE 43
Figure3-3. 43
PAGE 44
Theprimaryhumanactivitysourcesofnitrogenoxidearemotorvehicles,electricalutilitiesandotherindustrial,commercialandresidentialsourcesthatburnfuels.Nitrogenoxidesformwhenfuelburnsathightemperatures,inthesourcesmentioned.MobilesourcesareresponsibleformorethanhalfofallnitrogenoxideemissionsintheUnitedStates.About40percentofNOxemissionsarefrompowerplants.Therestisemittedfromvarioussourceslikeindustrialandcommercialboilers.Bothon-roadandnonroadmobilesourcesaremajornitrogenoxidepolluters. Nitrogenoxidescantravellongdistances,causingavarietyofhealthandenvironmentalproblemsinlocationsfarfromtheiremissionssource.Theseproblemsincludeozoneandsmog,whicharecreatedintheatmospherefromnitrogenoxides,hydrocarbons,andsunlight.Onsmoggydays,asymptomisdifcultybreathingortroublingseeingobjectsinthedistance.Nitrogenoxideemissionsalsocontributetotheformationofparticulatematterthroughchemicalreactionsintheatmosphere.NOxcanharmtheenvironmentinseveralways,suchasformatingground-levelozoneandacidaerosols,deterioratingwaterquality,andreactingwithotherelementstoformtoxicchemicals.MoredetailedtheproblemstheNOxcancausearethrough: 44
PAGE 45
Moreover,oneofthemostseriousimpactsisacidrain,whichwillbediscussedinaseparatesubsectionsinceitiscausedbyacombinationofSO2andNOx. 45
PAGE 46
TheCleanAirActAmendmentsof1990createanew,balancedstrategyfortheNationtoattacktheproblemofurbansmog.Overall,thenewlawrevealstheCongress'shighexpectations 46
PAGE 47
TheCleanAirActrequiresEPAtosetNationalAmbientAirQualityStandardsforsixcommonairpollutants.ThesecommonlyfoundairpollutantsarefoundallovertheUnitedStates.Theyareparticlepollution(oftenreferredtoasparticulatematter),ground-levelozone,carbonmonoxide,sulfuroxides,nitrogenoxides,andlead.Thesepollutantscanharmourhealthandtheenvironment,andcausepropertydamage.Ofthesixpollutants,particlepollutionandground-levelozonearethemostwidespreadhealththreats.BecausethemaininterestofthisstudyistoconcentrateonthepollutionproblemsthatarecausedfromNOxandSO2emissions,wewillonlyexaminethepartofCleanAirActthatconcernthesepollutants.TheCleanAirActcanbeseparatedinsomeactionsaccordinglytothepurposeoftheaction.Thus,weonlydiscussactionsthataffectstheNOxandSO2emissions. TheinitialphaseofEPA'sAcidRainProgramwentintoeffectin1995.Thelawrequiredthehighestemittingunitsat110powerplantsin21Midwest,Appalachian,andNortheasternstatestoreduceemissionsofSO2.Thesecondphaseoftheprogramwentintoeffectin2000,further 47
PAGE 48
EachallowanceisworthonetonofSO2emissionsreleasedfromtheplant'ssmokestack.PlantsmayonlyreleasetheamountofSO2equaltotheallowancestheyhavebeenissued.IfaplantexpectstoreleasemoreSO2thanithasallowances,ithastopurchasemoreallowancesorusetechnologyandothermethodstocontrolemissions.Aplantcanbuyallowancesfromanotherpowerplantthathasmoreallowancesthanitneedstocoveritsemissions. Thereisanallowancesmarketthatoperateslikethestockmarket,inwhichbrokersoranyonewhowantstotakepartinbuyingorsellingallowancescanparticipate.Allowancesaretradednationwide. EPA'sAcidRainProgramhasprovidedbonusallowancestopowerplantsforinstallingcleancoaltechnologythatreducesSO2releases,usingrenewableenergysources(solar,wind,etc.),orencouragingenergyconservationbycustomerssothatlesspowerneedstobeproduced.EPAhasalsoawardedallowancestoindustrialsourcesvoluntarilyenteringtheAcidRainProgram. The1990CleanAirActhasstiffmonetarypenaltiesforplantsthatreleasemorepollutantsthanarecoveredbytheirallowances.AllpowerplantscoveredbytheAcidRainProgramhavetoinstallcontinuousemissionmonitoringsystems,andinstrumentsthatkeeptrackofhowmuchSO2andNOxtheplant'sindividualunitsarereleasing.PowerplantoperatorskeeptrackofthisinformationhourlyandreportitelectronicallytoEPAfourtimeseachyear.EPAusesthisinformationtomakesurethattheplantisnotreleasingquantitiesofpollutantsexceedingtheplant'sallowances.Apowerplant'sprogramformeetingitsSO2andNOxlimitswillappearontheplant'spermit,whichisledwiththestateandEPAandisavailableforpublicreview. Besidestheground-breakingfeaturesintheAcidRainProgram,the1990CleanAirActencouragedotherinnovativeapproachesthatspurtechnology.Theseapproachesallowbusinesses 48
PAGE 49
Thesereductionswouldnotbepossiblewithoutcleaner,verylowsulfurgasolineanddieselfuel.Inadditiontotheirdirectemissionsbenets,cleanerfuelsenablesophisticatedemissioncontroldevicestoeffectivelycontrolpollution.CongressrecognizedtheimportanceofcleanerfuelstoreducingmotorvehicleemissionsandgaveEPAauthoritytoregulatefuelsintheCleanAirAct. TheCleanAirActrequirescertainmetropolitanareaswiththeworstground-levelozonepollutiontousegasolinethathasbeenreformulatedtoreduceairpollution.Otherareas,includ-ingtheDistrictofColumbiaand17states,withground-levelozonelevelsexceedingthepublichealthstandards,havevoluntarilychosentousereformulatedgasoline.Reformulatedgasolinereducesemissionsoftoxicairpollutants,suchasbenzene,aswellaspollutantsthatcontributetosmog. 49
PAGE 50
Since2006,renershavebegunsupplyingdieselfuelwithverylowsulfurlevelsforhighwaydieselvehicles.Aswithgasolinevehicles,efcientnewemissioncontrolsondieselenginesrequirethisUltra-LowSulfurDiesel(ULSD)fueltofunctionproperly.Highwaydieselfuelsulfurlevelsare97percentcleanerthandieselpriorto2006.In2007,renersbeganreducingsulfurindieselfuelusedfornonroaddieselengines,suchasconstructionequipment. TheCleanAirActencouragesdevelopmentandsaleofalternativefuels.Alternativefuelsaretransportationfuelsotherthangasolineanddiesel,includingnaturalgas,propane,methanol,ethanol,electricity,andbiodiesel.Thesefuelscanbecleanerthangasolineordieselandcanreduceemissionsofharmfulpollutants.Renewablealternativefuelsaremadefrombiomassmaterialslikewood,wastepaper,grasses,vegetableoils,andcorn.Theyarebiodegradableandreducecarbondioxideemissions.Inaddition,mostalternativefuelsareproduceddomestically,whichisbetterforoureconomy,energysecurityandhelpsoffsetthecostofimportedoil. TheCleanAirActalsorequiresEPAtoestablishanationalrenewablefuel(RF)program.Thisprogramisdesignedtosignicantlyincreasethevolumeofrenewablefuelthatisblendedintogasolineanddiesel. Dieselenginesaremoredurableandaremorefuelefcientthangasolineengines,butcanpollutesignicantlymore.Heavy-dutytrucksandbusesaccountforaboutone-thirdofnitrogenoxidesemissionsandone-quarterofparticlepollutionemissionsfromtransportationsources.Insomelargecities,thecontributionisevengreater.Similarly,nonroaddieselenginessuchas 50
PAGE 51
EPAhasissuedrulestocutemissionsfromonroadandnonroadvehiclesbymorethan90percentbycombiningstringentemissionsstandardsfordieselenginesandclean,ultra-lowsulfurdieselfuel.UndertheCleanAirAct,EPAisalsoaddressingpollutionfromarangeofnonroadsources,includinglocomotivesandmarinevessels,recreationalvehicles,andlawnandgardenequipment.Together,thesesourcescompriseasignicantportionofemissionsfromthetransportationsector. CongressrequiredconformityintheCleanAirActAmendmentsof1990.Inotherwords,transportationprojectssuchasconstructionofhighwaysandtransitraillinescannotbefederallyfundedorapprovedunlesstheyareconsistentwithstateairqualitygoals.Inaddition,transportationprojectsmustnotcauseorcontributetonewviolationsoftheairqualitystandards,worsenexistingviolations,ordelayattainmentofairqualitystandards. Theconformityprovisionsrequireareasthathavepoorairqualitynow,orhaditinthepast,toexaminethelong-termairqualityimpactsoftheirtransportationsystemandensurethatitiscompatiblewiththearea'scleanairgoals.Indoingso,thoseareasmustassesstheimpactsofgrowthonairpollutionanddecidehowtomanagegrowth. Propermaintenanceofacar'sengineandpollutioncontrolequipmentiscriticaltoreduceexcessiveairpollution.Tohelpensurethatsuchmaintenanceoccurs,theCleanAirActrequirescertainareaswithairpollutionproblemstoruninspectionandmaintenance(I/M)programs.The1990Actalsoestablishedtherequirementthatpassengervehiclesbeequippedwithonboarddiagnostics.Thediagnosticssystemisdesignedtotriggeradashboardcheckenginelightalertingthedriverofapossiblepollutioncontroldevicemalfunction.Tohelpensurethatmotoristsrespondtothecheckenginelightinatimelymanner,theActrequiresthatI/Mprogramsincludeaninspectionoftheonboarddiagnosticsystem. 51
PAGE 52
OperatingpermitsareespeciallyusefulforbusinessescoveredbymorethanonepartoftheCleanAirActandadditionalstateorlocalrequirements,sinceinformationaboutallofasource'sairpollutionisinoneplace.Thepermitprogramsimpliesandclariesbusinesses'obligationsforcleaningupairpollutionandcanreducepaperwork.Forinstance,anelectricpowerplantmaybecoveredbytheacidrain,toxicairpollutant,andsmog(ground-levelozone)sectionsoftheCleanAirAct.Thedetailedinformationrequiredbythoseseparatesectionsisconsolidatedintooneplaceinanoperatingpermit. Businessesseekingpermitshavetopaypermitfees,muchlikecarownerspayingforcarregistrations.Thesefeespayfortheairpollutioncontrolactivitiesrelatedtooperatingpermits. 52
PAGE 53
Inthischapterwewilldiscussthemethodsthatwereemployedtoconductthisstudyaswellastheconceptualframeworkthatledustousethesemethods.Webeginwiththeconceptualframework,thenwemovetoadiscussionofthemethodsthatwereusedtoanalyzepaneldatasetsandnallywedescribethequantileregressionmethodsthatarethemainfocusofourapplication.Firstweillustratequantileregressionmethodsforcross-sectionaldatasetsandthenwecenteronpaneldatamethodsaswellastheprecisealgorithmsusedbythesoftware(Rversion2.7.2)toestimatethecoefcientsandtheirstandarderrors. 53
PAGE 54
Wooldridge 2005 )andGreene's( Greene 2002 ). Usingpaneldataresultsinobservationsnotbeingidenticallyorindependentlydistributed.Forexample,inourcase,wheretheincomeandtheemissionsaremeasuredforeverystateforaseriesofyears,measurementscorrelateovertimeorcorrelatebetweentheyearsinvariousstates.Therefore,observationsarenotindependentlydistributed,resultinginbiasedandinconsistentestimatesincaseswheretimeseriesorcross-sectionmethodsareused.Thiscanbeeasilyshowninequation( 4 ),wherez0iaretheunobservedindividualheterogeneitywhichcontainsallthecharacteristicsthatarenotincludedintheregressorsandportraystheindividual(ortime)attributessuchasability,geographicalcharacteristics,etc.Incaseswhereweusemethodsofcross-sectionalanalysisandpooltheobservations,equation( 4 )becomesequation( 4 ) wherevit=z0i+"itisanerroroftencalledcompositeerror.Weknowfromtheassumptionsoftheordinaryleastsquares(OLS)thattheerrortermshouldbeuncorrelatedwiththeregressors. 54
PAGE 55
4 )willprovidebiasedandinconsistentestimatesincaseswheretheindividualeffectsz0iarecorrelatedwiththeregressorsxit. 4 )andcanbeestimatedbyOLS.Individualicanrepresentindividuals,states,nations,families,rms,cities,etc.acrosstime.Incaseswherethereisalargeamountoftimeobservationsortheobservationsaresupposedtovaryacrosstime,andwewanttocapturethisvariation,theLSDVmodelisdescribedbyequation( 4 )whereadditionaldummyvariablesareincludedforeverytimeperiod(weexcludeoneyearinordertoavoidcollinearity).Thexedeffectsmodelwheretimeandindividualeffectsareincludedisoftencalledatwo-wayxedeffectsmodel.Thenamexedeffectsdoesnotindicatenon-stochasticpropertiesbutinsteadindicatesthattheindividualortimeeffectsremainconstantovertime( Greene 2002 ). WhereyitandXitaretheobservationoftheithindividualinthetimeperiodt. Thetwo-wayxedeffectsmodelisgivenbytheequation: InorderforthexedeffectsmodeltoprovideunbiasedandconsistentestimatorswehavetomakesomeassumptionsasillustratedinWooldridge( Wooldridge 2005 ). Foramodeloftheformofequation( 4 )wehavethefollowing: AssumptionFE.1
PAGE 56
AssumptionFE.4 AssumptionFE.6 4 )canbereformulatedas 56
PAGE 57
wherethereareKregressorsincludingaconstanttermwhichisthemeanoftheunobservedindividualheterogeneity.Theuiistherandomheterogeneitycomponentspecictotheithindividualandisconstantthroughtime.Fromequation( 4 )wecaneasilyseethatui=z0iE[z0i].TobeREmodelconsistent,wehavetomakesomeadditionstothepreviouslydescribedassumptionsfortheFEmodel.Thus,followingWooldrige( Wooldridge 2005 )wereplaceFE.3,FE.4andFE.5withRE.3,RE.4andRE.5respectively,thereforewehave: AssumptionRE.3 E["2itjX]=2"; E[u2itjX]=2u;
PAGE 58
E[uiujjX]=0; 4 ),theerrortermofequation( 4 )canbewrittenasfollows: Thereforeusingequations( 4 )wehave: E[itisjX]=2+2u; E[2itjX]=2u;t6=s E[2itjsjX]=08tandsifi6=j 58
PAGE 59
4 )wecandeducethaterrortermsareseriallycorrelated.Hence,theusageofOLSwouldresultininefcientestimators.Consequently,toestimateourmodelefcientlywehavetousegeneralizedleastsquares(GLS). Usingmatrixalgebra,Greene( Greene 2002 )showsthatthetransformationforGLSofyiandXiis: SwamyandArora 1972 ). 59
PAGE 60
Hausman 1978 )providesaspecicationtestbetweentheFEandREmodelinordertotestwhichoneismoreappropriate.TheHausmantestisbasedonthehypothesisthattheestimatesfromFEandREshouldnotdiffersignicantlyifindividualeffectsareuncorrelatedwithregressorswhiletheREmodelismoreefcient.FollowingGreene( Greene 2002 )theHausmantestisachi-squaredtestwhichisbasedonthefollowingWaldcriterion: Hausman 1978 )wehavethefollowingresult: sinceHausman'sessentialresultisthatthecovarianceofanefcientestimatorwithitsdifferencefromaninefcientestimatoriszero,wehavethefollowing: Therefore,usingequations( 4 )and( 4 )wehave: 60
PAGE 61
4.3.1General 4-1 ),ifthesampledistributionisnotnormalorhasthickertails,themedianperformsbetterthanthemean.Ontheotherhand,ifthesampleapproximatesanormaldistribution,themeanappearstoperformbetter. Table4-1. Meanandmedianperformance DistributionDensityMeanMedian Normal1 2exp(z2 2exp(jzj)21.57Cauchy1 Imbens 2005 ) 4 )itestimatesequation( 4 ). whereistheselectedquantile. Themedian'sandmean'spropertiesusuallyplayacrucialroleinthemethodsthatareusedtoestimateregressionrelationships.Therefore,similartothepropertiesofmedianandmean,amodelthatestimatestheconditionalmeaniseasiertocollapsethanamodelthatestimatesthe 61
PAGE 62
Koenker 2005 ). Employingdecisiontheoryasanalternativewayofestimatingthequantileofasampleinsteadofrankingtheobservations,leadstominimizingtheexpectedloss.Ifweconsiderthatthelossisgivenbythepiecewisefunction( 4 ) thenwehavetondthe^xwhichminimizestheexpectedlossforsome2(0;1).Therefore,inordertondthequantilewehavetominimize:
PAGE 63
Followingthisminimizationproblem,KoenkerandBasset( KoenkerandBasset 1978 )providedanestimationmethodoftheconditionalquantileQy(jx)asafunctionofx's.Knowingthatthemeancanbeestimatedbysolvingtheminimizationprobleminequation( 4 ),theestimationoftheconditionalmeanofyonxas(x)=xTcanbeestimatedbyminimizing( 4 ). 63
PAGE 64
4 ),theconditionalquantileQy(jx)asafunctionofx'scanbegivenby( 4 ) Aswedescribedpreviously,oneadvantageofquantileregressionisthatitismorerobustanditdoesnothavetoassumeGaussianerrorsinordertoprovideconsistency,likeleastsquares.Nowthatwehavedescribedhowitispossibletoestimatequantileregression,wehavetodiscussthetheconditionsunderwhichtheestimatorsareconsistent.Supposethataconditionalquantilefunctionisgivenbyequation( 4 ): andthattheconditionaldistributionfunctionsFniofYi,i=1;2;:::;satisfythecondition: Koenker 2005 ),ElBantiandHallin( ElBantliandM 1999 )showedthattheaboveconditionisnecessaryandsufcientforconsistency,providedweassumethatthesequenceofdesignmatricesXsatisesthefollowingconditions
PAGE 65
Ontheotherhand,thereexistpenalizedmethods( Koenker 2004 )thatareusedforlongi-tudinaldatathatimprovetheefciencyoftheFEmodel.However,theymostlysuitedforcaseswheretherearejustfewobservations(usuallylessthan5)ofoneofthetwo-wayeffects,timeorindividuals.Nevertheless,notheoreticalbackgroundhasbeendevelopedyettousetwo-waypenalizedmethods.Therefore,evenifitwasmoreappropriate,itwouldbeinfeasibletouse.IncaseswhereweneededtoimprovetheFEmodel'sefciencywecouldestimateamodelwherequantileregressionsareestimatedsimultaneouslybyimposingidenticalindividualeffectsforeveryquantileandminimizingtheequation( 4 )( Koenker 2004 ) wherewkisaweighttocontroltherelativeinuenceoftheqquantilesf1;:::;qgontheesti-mationoftheithparameter.Nevertheless,usingthismodelweimposeonemoreassumption:individualeffectsdonotdependonthequantile. Therefore,sinceinourcasetherearesufcientdegreesoffreedomtoprovidesignicantresults,themoreappropriatemodeltouseistheFEbyestimatingeachquantileregression 65
PAGE 66
4 )havingeachindividualeffectidependonthequantile Qy(jxi)=i()+t()+x()+ui Therefore,theestimationmethodofthismodelisgoingtobethesameastheonethatwedescribedintheprevioussubsectionforthequantileregressionandissynopsizedinequation( 4 ).Inorderforsoftwaretosolvethisminimizationproblemandobtainestimators,weusethemethoddescribedindetailbyKoenkerandd'Orey( Koenkerandd'Orey 1987,1994 ).ThismethodisamodiedversionofBarrodaleandRobertsalgorithm( BarrodaleandRoberts 1974 )forl1-regressionandisquiteefcientforproblemsthatincludeuptoseveralthousandobservations.Additionally,italsoimplementsaschemeforcomputingcondenceintervalsfortheestimatedparameters,basedonaninversionofaranktestdescribedbyKoenker( Koenker 1994 ).Inadditiontothatmethod,thesoftwareprovidesseveralotheroptionstouseliketheFrisch-Newtoninteriorpointmethodwhichisusedforproblemswithverylargenumberofobservations. The,softwareemployedseveraloptionsfortheestimationofthestandarderrors.Althoughasymptoticestimationofthestandarderrorsispossible,inmanycasesitissomewhatproblem-atic.Weusebootstrapmethodsinordertoestimatethestandarderrors.ForourapplicationweusedthebootstrapoptionthatisbasedontheMarkovchainmarginalbootstrapproposedbyHeandHu( HeandHu 2002 )andKocherginsky,HeandMu( Kocherginsky,He,andMu 2005 ).ThesoftwareweusewasdevelopedinRbyKoenkeranditisdiscussedindetailinQuantileregressioninR:Avignette( Koenker 2005 ) 66
PAGE 67
Followingthemethodusedinmoststudiestoestimatetherelationshipofenvironmentaldegradation,weusedathirddegreepolynomialofincome.Thereasonforthisisbasedonthetheoryregardingtheshapethattheincomeandenvironmentalrelationshiphas.Aswediscussedintheliteraturereviewchapter,accordingtothistheorythepossiblecurvesareamonotonicincreasingcurve,aninvertedUcurveliketheEKChypothesizesandanNcurve.Inordertobeabletocapturethesecurves,weuseathirddegreepolynomialofincome.Henceourmodelisexpressedinequation( 5 ) whereE=capisemissionspercapita,inourapplicationNOxorSO2shorttonespercapita,whichindicatetheenvironmentaldegradation.Incomepercapitaismeasuredin1987dollars.Ourresultsarecomparabletotheexistentliteraturesincethemajorityofstudiesalsouseathirddegreepolynomial.Becausethedatasetweusehasbeenusedbyotherresearchers( Millimet,List,andStengos 2003 ; ListandGallet 1999 ),wereplicatethexedandrandomeffectsmodelinordertohaveallthestatisticalresultsandbeabletomakeadequatecomparisonswiththeresultsfromquantileregression. Inthefollowinganalysisweprovidediagramswiththecurvesthateachmodelprovidesaswellastableswithcoefcientsandtheirsignicancelevels.Thetablesdonotincludethe 67
PAGE 68
ReplicatingthestudiesthatusedFEweprovideidenticalresultswiththeestimatesfromstudiesthatusedthesamedataset( Millimet,List,andStengos 2003 ; ListandGallet 1999 ).Fromtable 5-1 wecanseethatintheFEmodelallcoefcientsarehighlysignicant,whiletheREmodelprovidesaninsignicantsquareincome.Sincetheestimatedequationisareducedform,wehavetorefrainfrommakingcausalconjecturesfromthevaluesoftheestimatedcoefcients.Additionally,sinceitisdifculttorealizewhatashortton(theunitinwhichresponsevariables,NOxandSO2,aremeasured)cancausetoenvironment.Nevertheless,thecoefcients'estimatedvaluesareimportantbecausetheyareusedtoconstructthecurvesoftheincome-pollutionrelationshipandtheturningpointsofthecurveiftheyexist. Fromtable 5-2 wecanseethattheHausmantestindicatesthatthenullhypothesisisrejectedinfavoroftheFEmodel(soxedandrandomeffectsdonotprovidesimilarresults).Thus,incomeislikelycorrelatedwithindividualortimeeffects,sotheREmodelisinconsistent.Table 5-2 alsoprovidessomeadditionalstatisticslikeR2andDurbin-Watson.R2intheFEappearstohaveaveryhighvalue(0.63)atleastincomparisontotheexistentliterature.Thisfactisprobablyduetotheuseofthetwo-wayFEmodelbutautocorrelationcouldalsocontributeto 68
PAGE 69
FixedandrandomeffectsmodelcoefcientsforNOx CoefcientValueStd.errort-valuep-value Randomeffectsmodel FixedandrandomeffectsmodelsummarystatisticsforNOx Sumofsquaredresiduals6.26115.44R2.6342380.2458Durbin-Watson0.2975p-value<2.2e-160.3002p-value<2.2e-16F-statistic88.3068(3052,3)343.711(3164,3) HausmantestofH0:REvs.FE:2=195,df=3p-value=<2.2e-16] Giventhenatureofourstudy,weconcentrateontheshapeoftheestimatedEKCcurve.Fromgure 5-1 wecanseethatbothrandomandxedeffectsconrmtheEKChypothesisusingNOx.Nevertheless,theresultsfortheturningpointoftheFEandREmodeldiffer.TheturningpointofFEisapproximatelyat$8,600,whiletheREmodelshowsaturningpointat$14,000.However,sincetheHausmantestindicatedinconsistencyontheREestimates,theFEresultsaredeemedmoretrustworthy.Figure 5-2 containsthecurvesfromtheestimatedequationsaswellastheobservationsinthesample.Wecanseethatthereisnotalargenumberofoutlierscomparedtothenumberofobservations(3168). 69
PAGE 70
EKCxedandrandomeffectsforNOxusingfullsample.A)Fixedeffects.B)Randomeffects. Figure5-2. EKCxedandrandomeffectswithscatterplotforNOxusingfullsample.A)Fixedeffects.B)Randomeffects. Table 5-3 showsthatallcoefcientsofthexedandtherandomeffectsmodelusingSO2asthedependentvariablearehighlysignicant.TheHausmantestfromtable 5-4 showsthattheREmodelwasrejectedinfavorofFEmodelasinthecaseofNOx.Therefore,thereislikelycorrelationbetweenindividualsortimeeffectsandincomepercapita.TheDurbin-WatsonpointoutthattheREandFEmodelssufferfromautocorrelation.R2isagainhighforFEandisrelativelylowintheREmodel. Figure 5-3 showsthecurvesfromtheFEandREmodelsforSO2.Aswecanseethereisatremendousdifferencenotonlyintheshapeofthecurvesbutalsointhescaleoftheresponse 70
PAGE 71
FixedandrandomeffectsmodelcoefcientsforSO2 CoefcientValueStd.Errort-valuep-value RandomEffectsModel FixedandrandomeffectsmodelsummarystatisticsforSO2 Sumofsquaredresiduals43.5820128.605R-squared.7006.061195Durbin-Watson.0102,p-value<2.2e-16.0898,p-value<2.2e-16F-statistic62.11(115,3052)68.7478(3164,3) HausmantestofH0:REvs.FE:2=154df=3p-value<2.2e-16 5-3 clearlyshowsthatthecurveisshiftedupwards.especiallyintheFEmodel.ThisfactcanbeinferredalsofromthedescriptivestatisticsoftheSO2emissionvariable.Wecanseefromtable 5-5 thatathighquantilesofSO2andincomethatafterthe80thpercentile,emissionsincreasedramaticallywhichresultsintheFEmodelpredictingunrealisticallyhighemissionsincomparisontowhathasbeenobserved,whileincome'squantilesobviouslydonotfollowthatincrease.Thiscanalsoseenfromthedifferencebetweenmeanandmedianwherethemedianisapproximately0.09whilemeanis0.16. 71
PAGE 72
HighquantilesforSO2 Figure5-3. EKCxedandrandomeffectsforSO2usingfullsample.A)Fixedeffects.B)Randomeffects. Figure5-4. EKCxedandrandomeffectswithscatterplotforSO2usingfullsample.A)Fixedeffects.B)Randomeffects. 72
PAGE 73
Panelquantileregressionresultsrevealhighlysignicantincomecoefcientsforallquantiles,asshownintable 5-6 5-5 theestimatedEKCcurvesfromthequantileregressionareplacedinthesamediagramfortworeasons:(1)sothatthecurvesfromdifferentquantilescanbecomparedtoseeiftherearedifferentslopesbetweenthemand(2)tobeabletoseeiftheycrossed.Theoretically,allcurvesshouldnotcrosseachotherbecause,forexample,ifthecurveof90thpercentilecrossedthecurveofthe95thpercentile,thenthatwouldmeanthatitispossibleforthe90thpercentilevaluetobelargerthanthe95thpercentilevalue.However,inpracticeitiscommonforcurvestocrossatsomepoints,butthatshouldnothappentoagreatextent( Koenker 2005 ).Aswecanseefromgure 5-5 thequantileregressionconrmstheEKCinthecaseofNOxbutitevidencesadifferentcurvefromboththepreviousREandFEmodels.BecausetheFEandREdiagramshavethesamescalewithquantileregressions'diagram,thecurvesfromthegures 5-2 and 5-5 canbedirectlycompared.Inviewofthefactthatquantileregressionprovidessimilarcurvesforallquantileswithonlydifferencethathigherquantilescurvesareupwardshifted,thepathofpollution'squantileswillbesimilaracrossincome.Atthebeginningofdevelopment,NOxemissionsincreaseatahigherratecomparedtotherateofdecreasewhentheturningpointisattained.Inaddition,thequantileregressions'turningpointsareapproximately$11,000.Athigherincomes,closetothemaximumvalueofoursample,emissionsseemtostabilizeandceasetodecreaseafterwards.InthecaseofNOxtheEKCdoesnotseemtocrosseachothertoagreatextent,apartfromthe25thpercentileandmedianwhich 73
PAGE 74
QuantileregressionforNOxusingfullsample Figure5-6. QuantileregressionforNOxwithscatterplotusingfullsample crosseachotherapproximatelyat$15,000.Wecanseefromgure 5-5 thatthecurvesaremuchcloserforlowerquantiles.Higherquantilesresultsinhigherdistancebetweenthemanditspreviousestimatedquantilewithanexceptionofthe95thand90thquantilethathavealmostidenticalcurves.ThisfactcanleadtoaconclusionthattheNOxdistributionispositiveskewed.Figure 5-6 revealsthatquantileregressiondidnotaffected,atleastinhighextent,fromoutlierssincethecurvesarecrossingorareclosetothemajorityoftheactualobservations. 74
PAGE 75
QuantileregressioncoefcientsforNOx 5thPercentile 10thPercentile 25thPercentile 75thPercentile 90thPercentile 95thPercentile 5-7 .Thecurves,however,donotconvincinglyconrmtheEKC.Nevertheless,itisinterestingtonotethattheFEmodelshowingamonotonicallyincreasingrelationshipchangesdramaticallyinthequantileregression 75
PAGE 76
QuantileregressioncoefcientsforSO2 5thPercentile 10thPercentile 25thPercentile 75thPercentile 90thPercentile 95thPercentile 76
PAGE 77
QuantileregressionforSO2usingfullsample Figure5-8. QuantileregressionforSO2withscatterplotusingfullsample Curvesfromthequantileregressionstillseemtobehavewellintermsofcrossingeachotherandalsoseemtohavesimilarslopesforeveryquantile.Asclearlydemonstratedingure 5-8 ,thequantileregressionisnotaffectedmuchbyoutliers,atleastnottotheextentseeninmodelsfortheconditionalmean.DifferentlyfromthecaseofNOx,SO2'squantilecurvesarenotcloseratlowerquantilescomparetothehigher.Besidesmedianand75thquantileregressioncurvethatareclose,therestcurvesseemtobesymmetricdistance,forinstance,5thand10thquantilecurve 77
PAGE 78
5-8 78
PAGE 79
Millimet,List,andStengos 2003 ).Inordertoseeifthosetwomethodologiesyielddifferentresults,weestimatethepreviousmodelsfortherstsubsampleof1929-1984andthesecond1985-1994.Wepresenttheresultsforthosetwosubsamplestoseeifthereisadifferenceintheestimatedcurves. 5-8 showsthatallcoefcientsarestatisticallysignicantforboththeFEandREmodels.IntheREusingthefullsamplethesquareofincomecoefcientwashighlyinsignicant(pvalue=0:76),whileinthesmallersampletheincomecoefcientissignicant.However,theHausmanteststillindicatesthattheREmodelestimatesareinconsistent.R2washigherinthesmallersamplethaninbothofthefullsamplemodels.Evenso,testsforserialcorrelationstillprovidethesameoutcomeasinthefullsample.Ingure 5-9 wecanseethatbothmodelsstillconrmtheEKCbuttheyprovidemorepessimisticresultsinthesensethattheturningpointoccursatahigherincome.Sinceourincomerangeissmallerinthissample,weproduceacurveonlyforthatrangeofincomewhichexplainswhythecurvedoesnotshowthedecreaseitshowedusingthefullsampleshows. Usingthequantileregression,wheremoreobservationsareneededtoprovidepreciseresults,wecanseethatthereisadifferenceinthestatisticalsignicanceofthecoefcients.Thereareveinsignicantcoefcientswherethreeofthemareofthecubeincometermandtwoofthemconcernthe95thpercentileregression.Ingure 5-10 wecanseethatthe95thpercentile'scurvecrossesthe90thpercentilecurvetoagreatextentbutalsothemedian'scurvecrossesthe 79
PAGE 80
FixedandrandomeffectsmodelcoefcientsforNOxusingsubsample1929-1984 Fixedeffectsmodel CoefcientValueStd.errort-valuep-value Randomeffectsmodel FixedandrandomeffectsmodelsummarystatisticsforNOxusingsubsample1929-1984 StatisticFERE Sumofsquaredresiduals0.0416180.046085R20.65910.26134Durbin-Watson0.2616,p-value<2.2e-160.2665,p-value<2.2e-16F-statistic48(105,2582),p-value:<2.2e-16316(2684,3),p-value=2.45e-5 HausmantestofH0:REvs.FE:2=204,df=3,p-value<2.2e-16 FixedandrandomeffectsforNOxusingsubsample1929-1984.A)Fixedeffects.B)Randomeffects. 80
PAGE 81
QuantileregressioncoefcientsforNOxusingsubsample1929-1984 CoefcientValueStd.errortvaluePr(>jtj) 5thPercentile 10thPercentile 25thPercentile 75thPercentile 90thPercentile 95thPercentile 25th'scurveforawiderangeofincomes.Ingeneral,thoughthecurves,aswellastheturningpoints,aresimilartothoseusingfullsample. EstimatesforSO2usingthe1929-1984sampleandtheFEandREmodelshaveexactlythesamecharacteristicsastheestimatesforthefullsample(tables 5-11 and 5-12 ).Allcoefcients 81
PAGE 82
QuantileregressionforNOxusingsubsample1929-1984 Table5-11. FixedanfrandomeffectsmodelcoefcientsforSO2usingsubsample1929-1984 FixedEffectsModel CoefcientValueStd.Errort-valuep-value FixedandrandomeffectsmodelsummarystatisticsforSO2usingsubsample1929-1984 StatisticFERE Sumofsquaredresiduals0.2670.267R20.700.05Durbin-Watson0.13p-value<2.2e-160.14p-value<2.2e-16F-statistic62(115,3052),p-value<2.2e-1649(2582,3),p-value=0.004 HausmantestofH0:REvs.FE:2=68.3023,df=3p-value=9.857e-15
PAGE 83
FixedandrandomeffectsforSO2usingsubsample1929-1984.A)Fixedeffects.B)Randomeffects. Figure5-12. QuantileregressionforSO2usingsample1929-1984 aresignicant,themodelssufferfromautocorrelation,andtheHausmantestrejectsthenullhypothesisinfavoroftheFEmodel.Nevertheless,theshapeofthecurve(gure 5-11 )issomehowdifferentfromthoseusingfullsample.TheFEestimatesseemtobeaffectedlesssinceitsmaximumvalueisabout0.6shortton,whereasitwasabout0.8forthefullsample.Additionally,thecurve'sslopeseemsmuchatteratahighincomelevel(thoughitisstillmonotonicallyincreasing). ContrarytothemethodsestimatingtheconditionalmeanofSO2,quantilemethodshavedifferentresultswhenusingtherstpartialsample,atleastintermsofthestatisticalsignicance 83
PAGE 84
QuantileregressioncoefcientsforSO2usingsubsample1929-1984 CoefcientValueStd.errortvaluePr(>jtj) 5thPercentile 10thPercentile 25thPercentile 75thPercentile 90thPercentile 95thPercentile 5-13 showsthat11outof21coefcientsareinsignicant.Allthecubeincometermsareinsignicantbutnoneoftherstpowerofincome.Althoughitisremarkablethatthecurvesbarelycrosseachotheringure 5-12 ,theyseemalmostatandparalleltoeachother,apartfromthe$15000levelofincomewherepollutionstartstodecrease.Whilethis 84
PAGE 85
85
PAGE 86
Koenker 2004 )orusingthesamexedeffectsforeveryquantileandestimatingallthequantileregressionssimultaneously.Nevertheless,weleavetheseexercisesforfuturework. TheFEandREmodelscollapseforthissampleinthecaseofNOx.(table 5-14 )Tosomeextentthatwasexpectedduetothelackofdegreesoffreedom,especiallyintheFEmodel.Allthecoefcientsareinsignicantinbothmodels.However,itshouldbepointedoutthatintheFEmodeltheyarejointlysignicant,whileintheREmodeltheyarejointlyinsignicant.AlsotheHausmantestdoesnotrejecttheREinfavoroftheFEmodel,thoughthep-value(0.057)wasreallycloseto0.05,i.e.itrejectsfor6%level.Sinceallcoefcientsareinsignicant,theshapeandpositionofthecurvesarenotverymeaningful,excepttoshowthatthecoefcientsareaffectedbyoutliersasshowningure 5-2 .Itisalsonoteworthythatthecurvesdonotshowthesamerateofpollutionreductionliketheydidinthefullsample. Table5-14. FixedandrandomeffectsmodelcoefcientsforNOxusingsubsample1985-1994 Fixedeffectsmodel CoefcientValueStd.errort-valuep-value
PAGE 87
FixedandRandomEffectsModelSummaryStatisticsforNOxusingsubsample1985-1994 StatisticFERE Sumofsquaredresiduals0.288630.33228R20.9370.014Durbin-Watson2.2,p-value=0.97681.9,p-value=0.2262F-statistic105(59,420),p-value<2.2e-162.3(476,3),p-value0.28 HausmantestofH0:REvs.FE:2=7.5,df=3,p-value=0.057 FixedandrandomeffectswithscatterplotforNOxusingsample1985-1994.A)Fixedeffects.B)Randomeffects. Forquantileregression,theresultsintable 5-16 seembetterthantheFEandREmodels.However,therearestillmanyinsignicantcoefcients,especiallyatthe90thand95th,per-centiles,whereallcoefcientsareinsignicant.Figure 5-14 conrmsthecollapseofatleastsomeoftheconditionalquantilessincemanycurvescrosseachotherextensively.Additionally,theyfailtoconvincinglyconrmthepollutiondecreaseseeninboththerstsubsampleandthefullsample. InthecaseofSO2thereisacontradictionbetweenthetablesandthegraphs.InboththeFEandREmodels(table 5-17 andtable 5-18 ),allcoefcientsarehighlysignicantbutthecurves 87
PAGE 88
QuantileregressioncoefcientsforNOxusingsubsample1985-1994 CoefcientValueStd.errortvaluePr(>jtj) 5thPercentile Income4:8203e051:7521e052:751186:1951e03Income22:3770e091:0385e092:289012:2574e02Income33:8158e142:0159e141:892895:9060e02 10thPercentile Income4:8203e051:6449e052:930453:5694e03Income22:3770e099:8291e102:418351:6015e02Income33:8158e141:9167e141:990874:7143e02 25thPercentile Income4:1707e051:4612e052:85433494:5262e03Income22:1416e098:6721e102:46954291:3925e02Income33:5732e141:6952e142:10787863:5633e02 Income3:4614e051:2673e052:731286:5747e03Income21:7281e097:7934e102:217392:7131e02Income32:8050e141:5743e141:781807:5504e02 75thPercentile Income3:4857e051:2418e052:80695:2351e03Income21:7416e097:5952e102:29302:2343e02Income32:8945e141:5132e141:91295:6446e02 90thPercentile Income2:1402e051:7125e051:249742:1209e01Income21:1226e091:0025e091:119832:6343e01Income31:9050e141:9200e140:992183:2168e01 95thPercentile Income2:1402e051:6674e051:283522:0002e01Income21:1226e099:7655e101:149592:5097e01Income31:9050e141:8704e141:018473:0904e01 5-15 obviouslydonotprovideagoodt,despitethefactthatR2ishighintheFEmodel.TheHausmantestindicatesthatFEperformsbetterandthusREprovidesinconsistentresults. 88
PAGE 89
QuantileregressionforNOxusingsubsample1985-1994 Table5-17. FixedandrandomeffectsmodelcoefcientsforSO2usingsubsample1985-1994 Fixedeffectsmodel CoefcientValueStd.Errort-valuep-value Randomeffectsmodel FixedandrandomeffectsmodelsummarystatisticsforSO2usingsubsample1985-1994 StatisticFERE Sumofsquaredresiduals9.2164e-060.11083R20.98440.09Durbin-Watson1.2,p-value<2.2e-161.1,p-value<2.2e-16F-statistic20.4(420,3),p-value0.01416.4(476,3),p-value:0.02 HausmantestofH0:REvs.FE:2=24.3576,df=3p-value=2.103e-05
PAGE 90
FixedandrandomeffectswithscatterplotforSO2usingsample1985-1994.A)Fixedeffects.B)Randomeffects. Figure5-16. QuantileregressionsforSO2usingsample1985-1994 Figure5-17. QuantileregressionsforSO2usingsample1985-1994 90
PAGE 91
QuantileregressioncoefcientsforSO2usingsubsample1985-1994 CoefcientValueStd.errortvaluePr(>jtj) 5thPercentile Income1:0982e043:1375e053:5001555:1475e04Income26:2864e091:9290e093:2588751:2095e03Income31:1853e133:9111e143:0305982:5916e03 10thPercentile Income1:0982e043:2556e053:3732318:1185e04Income26:2864e091:9924e093:1550991:7198e03Income31:1853e134:0220e142:9470483:3870e03 25thPercentile Income9:3764e053:1512e052:975483:0942e03Income25:2530e091:9357e092:713686:9279e03Income39:5942e143:8824e142:471211:3862e02 Income7:2118e053:1872e052:262742:4162e02Income23:9430e091:9458e092:026384:3357e02Income37:2773e143:8277e141:901265:7953e02 75thPercentile Income7:3487e052:9164e052:51971:2114e02Income24:0908e091:7721e092:30852:1457e02Income37:6014e143:5071e142:16743:0764e02 90thPercentile Income7:4006e053:7258e051:986294:7651e02Income24:2652e092:2685e091:880166:0778e02Income38:2738e144:4992e141:838966:6627e02 95thPercentile Income7:4006e053:9180e051:8888865:9596e02Income24:2652e092:3968e091:7795757:5869e02Income38:2738e144:7597e141:7382908:2893e02 5-19 )butalsointhegraphs(gure 5-16 ).Despitethefactthat6coefcientsareinsignicant,theirp-valuesareverycloseto0.05.Additionally,curvesdonotcrosseachothertoagreatextentand,aswecansee 91
PAGE 92
5-17 wherethescaleismuchclosertothefullsamplegraphs,thecurvehasasimilarshapetothatofonefromthefullsample. Overall,theresultsfromthesubsamplesshowthat,whiletherstsubsamplecloselyfollowstheresultsofthefullsample,inthesecondsubsamplethatanyhappensforthequantileregres-sionofSO2.Theseresultsmightbeexpectedsincethesecondsubsampleisrelativelysmallandmostobservationscomefromtherstsubsample.Inanycase,theresultsforthesubsamplesarenotdramaticallydifferent,whichmaybeanindicationthattheswitchinmethodologiestomeasureemissionsbyEPAhaslittleimpactonthefullsampleresults. InthecaseofNOx,modelsforbothquantilesandthemeanconrmtheEKChypothesis,althoughinthecaseofquantileregressionthereisnotactuallyaninvertedUshapesincetherateofdecreaseafterthepeak(turningpoint)ofthefunctionisslowerthantherateofincreasebeforethepeak,atleastinthesample.Ontheotherhand,inboththeREandFEforthe 92
PAGE 93
CurvesfromquantileregressionandxedeffectsmodelforconditionalmeanforNOx CurvesfrommedianregressionandxedeffectsmodelforconditionalmeanforNOx 5-18 whereboththecurvesfromquantilesregressionandtheFEmodelforconditionalmeanarepresented.Alsoingure 5-19 wecandirectlycomparethecurvesofthetwomomentsthatmeasurethecentraltendency,themean(fromFEmodel)andthemedian.Inthisgurethetwoverticallinesshowstheturningpointsofthecurves. Devisingaprecisemethodologyiscrucialbecausegovernmentsmaybasetheirenvironmen-talpoliciesonresultsfromtheconditionalmeanmethodsandthereforebelievethatpollution 93
PAGE 94
CurvesfromquantileregressionandxedeffectsmodelforconditionalmeanforSO2 OurresultsfromquantileregressionarenotveryoptimistictowardtheEKChypothesis.Therefore,economistsshouldbemorecarefulaboutanalyzingthebenetsfromeconomicdevelopmentsinceafterthepeakofenvironmentaldegradationoccurs,itisnotfollowedbyaconsiderablereductioninpollution.Sinceourresultsarefromareduced-formequation,wehavetorefrainfrommakingcausalinferences.Furthermore,wecannotbesureaboutthecausesofthereductionaftertheturningpoint,whichcouldbetheresultoftechnologicalprogressorstricterenvironmentalpolicies.Inthatregard,apossiblewaytoincreasethereductionofpollutionaftertheturningpointisthroughtechnologyincentivesand/orstricterenvironmentalpolicies. InthecaseofSO2theresultsaremoreheterogeneousNOx.SincetheHausmantestindicatesthattheREmodelisinconsistent,weonlyconsidertheFEmodelfortheestimationoftheconditionalmeanequation.Inthatcase,itisclearfromgure 5-4 thattheFEequationfortheSO2hasbeenshiftedbecauseoftheoutliers.Additionally,itsshapemonotonicallyincreases, 94
PAGE 95
CurvesfrommedianregressionandxedeffectsmodelforconditionalmeanforSO2 5-7 ).Thesecomparisonsaregraphicallypresentedingures 5-20 and 5-21 .BothcasesarenotoptimisticintermsoftheEKChypothesisandtheresultsfromtheestimationofNOxequation.Thequantileregressionimpliesthatadditionalenvironmentaldegradationwillnotbeproducedbyeconomicgrowth,while,insteadtheFEconditionalmeanresultsindicatethatpollutionwillcontinuetoincreasewitheconomicgrowth.Therefore,basedontheseresults,animplicationisthatstricterenvironmentalregulationshouldbeimposedtodecreaseSO2pollution.Incaseswherethepollutionlevelisatacceptablelevels,accordingtothequantileregressionresultsnofurtherpolicyrestrictionsshouldbeimposedthatcouldaffecteconomicactivity.Ontheotherhand,iflesspollutionisdesired,stricterenvironmentalregulationsarenecessary. 95
PAGE 96
96
PAGE 97
GrossmanandKrueger 1991 )intheirstudyoftheenvironmentalimpactsoftheNorthAmericanFreeTradeAgreement(NAFTA).ThisstudywasbaptizedastheEKCinPanayotou'sstudy( Panayotoy 1993 ).However,despiteitsmanysupportersandsomeempiricalevidencethatsupportsthevalidityoftheEKC,manyresearchersopposethisconceptandhaveproducedstudiesthatcontradictit.Duetothesecontradictionsothertheorieswerecreatedtodescribetheincome-pollutionrelationship,suchasthenewtoxicstheoryandtheracetothebottomscenario.InadditiontothetheoreticalargumentsopposingtheEKChypothesis,EKCstudiesreceiveeconometriccritiques.Someoftheproblemsthattheseapplicationstendtofaceareomittedvariablebiases,autocorrelationandcointegration.TheEKCstudiesintheliteraturehavegreatdisparitiesintermsoftheirresults,butsomeofthereasonsforthiscanbeattributedtotheuseofdifferentdata,differentmodels,differentpollutants,differentvariables,anddifferentmeasurementsofincome. Inthisstudywefocusonadifferentaspectoftheincome-pollutionrelationshipbyconcen-tratingondifferentaspectsofthedistribution.Namely,thepanelquantileregressionmethodisemployedtoestimatetherelationshipatdifferentpointsofemission'sdistributionasrelatedtoincome.Quantileregressionprovidessomeadvantagescomparedtotheusualmethodsemployedtoestimatetheconditionalmean.Forexample,theestimatesaremorerobustandinferencesareunencumberedbythemorerestrictiveassumptionsthatunderpinotherestimationmethods.Ad-ditionally,quantileregressionestimatesotherconditionalquantilesthatmightrevealadifferent 97
PAGE 98
Theestimatesfortheconditionalmeanandtheconditionalquantilesprovidedifferentresults.WhileinNOxemissionsbothFEmodelsconrmtheEKChypothesis,theFEmodelforthequantilesresultedinamuchlowerpollutionreductionthantheFEmodelforthemean.Inthatsense,theestimatesforthequantilescanbeconsideredaslessoptimistic.Inadditiontoadifferentshapeofthecurve,resultsshowadifferentturningpointsinceareductionforthemeanpollutionisnoticedafter$8600,whereasallquantilesshowtheturningpointatapproximately$11000.TheoppositehappenedwhenSO2isusedasaresponsevariable.InthecaseofFEfortheconditionalmean,itshowsapessimisticestimateinwhichtheseisamonotonicallyincreasingrelationshipwhileFEforquantileregressionshowedarelationshipthatasymptoticallyapproachesamaximum.Anotherndingisthatthemodelfortheconditionalmeanishighlyaffectedbyoutliers.Therefore,estimatesfortheconditionalmedianappearmorevaluablethantheusualmethods. PossibleexplanationsforthedifferentresultsthatthetwoemissionsprovideisthateithertheEKChypothesiscannotdescriberealisticallytheincome-pollutionrelationship,soitis 98
PAGE 99
99
PAGE 100
100
PAGE 101
101
PAGE 102
102
PAGE 103
103
PAGE 104
104
PAGE 105
DimitriosKapetanakisreceivedhisMasterofScienceatUniversityofFloridainFoodandResourceEconomicsDepartment.HereceivedhisBachelordegreeatAgriculturalUniversityofAthensmajoredinruraleconomicsandeconomicdevelopment. 105
|
