A Ricardian analysis of the impact of climate change on agriculture in Germany

ClimaticChange(2009)97:593–610

DOI10.1007/s10584-009-9652-9

ARicardiananalysisoftheimpactofclimatechange

onagricultureinGermany

C.Lippert·T.Krimly·J.Aurbacher

Received:14March2008/Accepted:22May2009/Publishedonline:18August2009

©SpringerScience+BusinessMediaB.V.2009

AbstractBasedonaRicardiananalysisaccountingforspatialautocorrelationandrelyingonrecentclimatechangeforecastsatalowspatialscale,thisstudyassessestheimpactofclimatechangeonGermanagriculture.Giventhelimitedavailabilityofdata(e.g.,theunknownaveragesoilqualityatthedistrictlevel),aspatialerrormodelisusedinordertoobtainunbiasedmarginaleffects.TheRicardiananalysisisperformedusingdatafromthe1999agriculturalcensusalongwithdatafromthenetworkofGermanweatherobservationstations.Thecross-sectionalanalysisyieldsanincreaseoflandrentalongwithbotharisingmeantemperatureandadecliningspringprecipitation,exceptforintheEasternpartofthecountry.ThesubsequentsimulationoflocallandrentchangesunderthreedifferentIPCCscenariosisdonebyenteringintotheestimatedregressionequationsspatiallyprocesseddataaveragesfortheperiodbetween2011and2040fromtheregionalclimatemodelREMO.Theresultingexpectedbene?tsarisingfromclimatechangearerepresentedinmapscontainingthe439Germandistricts;thecalculatedoverallrentincreasecorrespondstoapproximately5–6%ofnetGermanagriculturalincome.However,inthelongrun,whentemperatureandprecipitationchangeswillbemoreseverethanthosesimulatedfor2011–2040,incomelossesforGermanagriculturecannotbeexcluded.1Introduction:climatechangeinGermany

Asaconsequenceofthegreenhouseeffect,anon-goingchangeoftheglobalclimateisprojectedforthenextdecades.TheIPCCreport(2007a)expectsanincreaseofthemeanglobaltemperatureby1.8?Cto4.0?C.Furthermore,precipitationandtheoccurrenceofextremeweathereventswillincrease.Overthepast100yearsthe

C.Lippert(B)·T.Krimly·J.Aurbacher

InsititueforFarmManagement(410a),SectionofProductionTheoryandResource

Economics,UniversitätHohenheim,SchlossOsthof-Süd,70593Stuttgart,Germany

e-mail:clippert@uni-hohenheim.de

594ClimaticChange(2009)97:593–610averagetemperatureincreaseinEuropewas1?C,comparedtoaglobalaveragetemperatureincreaseofabout0.7?C(IPCC2007a).ThemeantemperatureinEuropeisexpectedtoincreaseby2.1?Cto5.3?Cbytheendofthiscentury;again,Europeshowsastrongerwarmingtrendthantheglobalaverage.Sinceagricultureisaneconomicactivitywhichstronglydependsontheclimatesettingandisparticularlyresponsivetoclimatechanges,itisimportanttounderstandhowsuchchangesmayaffectagriculturalproductivityandpro?tability.

Inprinciple,therearetwomainapproachestoassessingtheimpactofclimatechange(Mendelsohn2007):onewayistorunsimulationmodels,theparametersofwhichhavetobeobtainedfromcontrolledexperiments;theotherwayistoconductacross-sectionalanalysisobservingthe(economic)systemacrossdifferentlocationsinordertodeterminehowthesystemmayadapttodifferentclimates.Thismethod,usuallyreferredtoasaRicardianapproach,http://wendang.chazidian.comingobservedlandprices,itsbasicpurposeis“[...]toinferthewillingnesstopayinagriculturetoavoida3?Ctemperaturerise(forexample)byexaminingtwoagriculturalareasthatarethesameinallrespectsexceptthatonehasaclimateonaverage3?Cwarmerthantheother”(Kolstad2000:317;forabroaderdescriptionoftheunderlyingtheorycf.Mendelsohnetal.1994;MendelsohnandReinsborough2007:10f.;Lang2007:425f.).Inthecaseofcompetitivemarkets,assumingthatlandpricesatdifferentlocationshavereachedtheirlong-runequilibrium,thisapproachaccountsforboththedirecteffectsofclimateoncropyieldsandtheindirecteffectsresultingfromthesubstitutionoradaptationoffarmingactivities.

WhereastheRicardianapproachhasbeenfrequentlyusedforNorthernAmerica(e.g.,Mendelsohnetal.1994;PolskyandEasterling2001;Schlenkeretal.2005,2006;DeschênesandGreenstone2007;MendelsohnandReinsborough2007),http://wendang.chazidian.comng(2007)analysedweatherdataalongwith1990through1994paneldatafromfarmersinformerWestGermany,andfound,amongotherresults,aninverselyu-shapedrelationshipbetweenthelocaltemperaturesumduringthegrowingseasonandlandrentalprices.Hepredictedthat“[...]Germanfarmerswillbewinnersofclimaticchangeintheshortrun,withmaximumgainsoccurringatatemperatureincreaseof+0.6?Cagainstcurrentlevels”(Lang2007:423).

Whencomparedwiththementionedexperimental-simulationapproach,onead-vantageofaRicardiananalysisisthatitisbasedonreal-worldadaptationmeasureswhichhavebeenbroughtaboutbyatrial-and-errorprocessinvolvingmanyfarmerswellacquaintedwiththeirspeci?clocalproductionconditions.AmajorweaknessoftheRicardianapproachconsistsintheinadequacyofextrapolatingitforclimaticsettings(e.g.,temperature,CO2-fertilisation)whichhavenotbeenobservedsofar(i.e.,settingswhicharenotcoveredbythedatasetusedtoestimatetheHedonicPricingfunction).Furthermore,theapproach“mustworkhardnottobebiasedbyomittedvariablesthatarecorrelatedwithclimate”(Mendelsohn2007:2).

OnepromisingwaytocopewiththeproblemofspatialautocorrelationistoexplicitlyconsiderspatialautocorrelationoftheresidualswhenestimatingtheparametersoftheHedonicPricingmodel.Uptonow,thishasrarelybeendoneinthecontextofclimatechangeimpactassessment.ExceptionsareSchlenkeretal.(2006:116),DeschênesandGreenstone(2007:366),whoadjustedthestandarderrorsoftheirestimatedmodelsforspatialdependence,andtoacertainextentPolskyand

ClimaticChange(2009)97:593–610595Easterling(2001),whoincludedadditionalexplanatoryvariablesreferringtoalargerspatialscale(districts)intheircounty-basedanalysis.

Ourapproachtakesonlylong-termclimaticvariablesintoconsiderationalthoughSchlenkerandRoberts(2006)indicatethatalreadysingledayeventscanhavesigni?cantin?uenceonyields.However,dailyweatherdataforGermanywasnotavailabletous.DeschênesandGreenstone(2007)criticizedtheRicardianapproachwhichintheiranalysisturnedouttobestronglyin?uencedamongotherthingsbythechoiceofvariablesincludedintotheestimatedequation.Alternatively,theysuggestedandappliedanapproachwheretheyusedtheobservedyear-to-yearvariationofprecipitationandtemperaturetoexplainagriculturalpro?tsintheUnitedStates.However,astheyadmit,indoingsofarmers’damagesduetoclimaticchangearesystematicallyoverstatedbecausethestatisticalmodelthendoesnotaccountforcompleteadaptationwhichisimpossiblewhenonlyreactingtotheweathereventsofsingleyears.

TheobjectiveofthispaperistoassesstheimpactofclimatechangeonGermanagricultureusingrecentclimatechangeforecastsatalowspatialscale,relyingonaRicardiananalysiswhichaccountsforspatialautocorrelation.Inthenextsectionwewillpresentanappropriatestatisticalmodelrelyingonaspatialweightmatrix(Section2.1)aswellasthedata(Section2.2)thatwill?nallybeusedtoestimatetwoHedonicPricingfunctions(Section2.3).Then,bymeansofthesefunctions,theeconomicimpactofthreedifferentclimatechangescenarios(Section3.1)onthepro?tabilityofGermanAgriculturewillbepresented(Section3.2)anddiscussed(Section4).

2Empiricalanalysis

2.1Statisticalmodel

Inthefollowing,afunctionalrelationshipbetweentherentalpriceriforfarmlandatlocationianddifferentexogenousfactorsxcandxncisassumed:

ri=f(xc,xnc),(1)

wherexcisavectorofclimatecharacteristicssuchasmeanannualtemperatureoraverageprecipitationindifferentmonths,andxncstandsforavectorofnon-climatevariablessuchasgrasslandshareofoverallagriculturallandorsoilquality.Sinceitisimpossibletoobtainsuf?cientdataforallrelevantvariablesxncwhenestimatingtheHedonicPricingfunction1,weexplicitlyconsideredspatialautocorrelation.Equations2and3outlinethegeneralversionofacorrespondingspatiallyautoregres-sivemodel(Anselin1988:34ff.;LeSage1999:52f.)whichaccountsforbothspatiallagdependenceandspatialerrordependence(cf.PattonandMcErlean2003:37):

r=ρW1r+Xβ+u

u=λW2u+ε

with

????ε?N0,σ2I,(2)(3)

596ClimaticChange(2009)97:593–610where

r

X

Ws

I

u

εn×1vectorcontainingthereportedaveragefarmlandrentalprices,eachassociatedwithaspeci?cadministrativedistricti(i=1,...,n);n×(1+k)designmatrixcontainingasetofobservationsforkexplanatoryclimateandnon-climatevariables;givenn×nspatialweightmatrices(s=1,2;W1andW2maybeidentical);n×nidentitymatrix;n×1vectorofthespatiallycorrelatedresiduals;

n×1vectorofnormallydistributederrors(mean=0,variance=σ2).

Theparameterstobeestimatedare

ρβ

λspatiallagcoef?cient;(1+k)×1vectorcontainingtheregressioncoef?cientsfortheexplanatoryvariables;

coef?cientre?ectingthespatialautocorrelationoftheresidualsui.

Forthefollowingestimations,wewillalwaysuseastandardised?rst-ordercontigu-itymatrix(W=W1=W2).Noticethatsuchamatrixre?ectssimpleneighbourhoodalone(inourcasebetweenthen=439districtsofGermany):ineveryrowia0isassignedtoeverydistrictj=ithatdoesnotadjoinwithdistricti;thesameisdoneforalldiagonalelementsofthen×nmatrix.Whentwodistrictsiandjarecontiguous,1/giwillbeassignedtotheintersectionoftheithrowandthejthcolumn(wheregiisthenumberofdistrictswhichhaveacommonborderwithdistricti).WithWsetuplikethis,then×1vectorWrgivesforeverydistrictithemeanrentalpriceobservedinitscontiguousdistricts.Asigni?cantpositiveparameterρwouldhintataself-enforcingeffectofhigherfarmlandrentalprices(acaseofspatialdependencyofrentalprices).SolvingEq.3forthevectorofthespatiallycorrelatedresiduals(u)andenteringtheresultingtermintoEq.2gives:

r=ρWr+Xβ+(I?λW)?1ε

<=>(I?λW)r=(I?λW)ρWr+(I?λW)Xβ+ε

<=>r=ρWr+λW(r?ρWr)+Xβ?λWXβ+ε.(4)(4a)

Aproblemoccurswhenoneormoreoftheoftenspatiallycorrelatedfactorswhichin?uencerentalpricesarenotaccountedforinthestatisticalmodel,ascanbeeasilydemonstratedbylookingatEq.4a:neglectingpossiblespatialdependency(i.e.,assumingρ≈0),Eq.4aisreducedto

ε

<=>ε==r?Xβ?λWr+λWXβu?λWr+λWXβ.(4b*)

IfallrelevantexplanatoryvariableswerecontainedinX,thevaluesriwouldincreaseanddecreaseinlinewithXiβ(apartfromthe“whitenoise”ε;i.e.,ε=u).Theparameterλthenwouldbeclosetozero.Ontheotherhand,ifimportantspatiallycorrelatedexplanatoryvariableswerenotcontainedinXβ,theresidualsuiwouldhavetobecorrectedbyλWrinordertoobtainavectorofnormallydistributed

ClimaticChange(2009)97:593–610597residualsε.(Notethatinthelattercase,λWXβwillcontainonlynormallydistributedelements,whereasλWrwillpositivelydependonu.)Asigni?cantvalueforλmeansthatthereisatleastonespatiallycorrelatedfeaturewhichisnotre?ectedbytheexogenousvariablesusedinthemodel,butwhichaffectstheobservedrentalprice.Subsequentregressionanalysesalwaysyieldedahighlysigni?cantMoran’sIaswellasahighlysigni?cantvalueforλ(whichindicatesthatsomeoftherelevantexplanatoryvariablesxcandxncwerenotincludedinX),whereasasigni?cantparameterρcouldnotbefound.Consequently,ourparameterestimatespresentedinthefollowingarebasedonthesimplespatialerrormodel:

r=Xβ+(I?λW)?1ε.(4b)

TheEq.4bwereestimatedusingMATLABalongwiththe“EconometricsToolbox”byLeSage(2003)(forthespatialerrormodelandtheiterativemaximumlikelihoodestimationemployed,cf.LeSage1999:48f.).

2.2Data

SourcesforlanddataDataregardingthedistricts’utilisableagriculturalarea(UAA)andthegrasslandshareofthatareaweretakenfromthe1999agriculturalcensus(StatistischeÄmterdesBundesundderLänder2001).Alltogether,theUAAofthe439Germandistrictsamountedtoatotalof17,157,906hectares.The1999yearlyrentalpricerforfarmland(inEuroperhectareUAA)bydistrict(Landkreis)waskindlycommunicatedbyStatistischesLandesamtBaden-Württemberg(2007).For14ofthe439Germandistricts,partoftheagriculturalcensusdatawaslackingandhadtobereplacedbycarefulassumptionsbasedonobservationsfromsimilardistricts(usingaspatialweightmatrixW,itwasimpossibletoomitthecorrespondingdistrictsfromtheanalysis).Relyingonrentalpricesinsteadoffarmlandpriceshastheadvantagethatweneednotconsidersomefactorswhichstronglydistortfarmlandprices,especiallyinthedenselypopulatedregionsofCentralEurope.Forexample,thehighpricesforfarmlandwhichmaybecomebuildinglandinthemediumtermhavenothingtodowithrealagriculturalproductivity.RentingfarmlandisquitecommoninGermany:in1999,68.4%ofGermanfarmsrentedatleastpartoftheirland;theshareofrentedlandwasabout50%ofoverallcultivatedfarmlandinWestGermanyandabout90%inEastGermany(BMVEL2001:12).Sincelandrentcontractsalwaysexpireaftersomeyearsthereportedlandrentalpricestobepaidonayearlybasisbythetenant(whoisnotentitledtosellthelandandwhowillnotbene?tfromalandsalebytheowner)donotcontaintheoptionvalueofthecorrespondingplotsoflandforapossibleurbandevelopment.Theyarejustre?ectingtheagriculturalproductivityoftheland.Incontrast,purchasepricesforfarmlandalsoincludethementionedoptionvalue.

SourcesforclimatedataThisanalysisuseddatafromweatherobservationstationsfromtheGermanWeatherService(DWD2007).Thisdatasetcontainsthelatitudeandlongitudeofthestation(ingeographiccoordinates),altitude,andaveragesoftemperature,precipitationandsunshinedurationover30years(1961–1990)forthewholeyearandforeachmonth.Theprecipitationdatasetconsistsof4748stations,whilethedatasetfortemperaturesincludes675stations.Forthefuturescenarios,climatedatafromtheREMOmodel(MPIonbehalfoftheUmweltbundesamt2006)