UCD CENTRE FOR ECONOMIC RESEARCH WORKING PAPER SERIES 2010 The Economic Impact of the Little Ice Age Morgan Kelly and Cormac Ó Gráda, University College Dublin WP10/14 April 2010 UCD SCHOOL OF ECONOMICS UNIVERSITY COLLEGE DUBLIN BELFIELD DUBLIN 4
ÓGráda∗April16,,,an-nualsummertemperaturereconstructionsbetweenthefourteenthandtwen-tiethcenturiesbehaveasalmostindependentdrawsfromadistributionwithaconstantmeanbuttimevaryingvolatility;whilewintertemperaturesbe-havesimilarlyuntilthelatenineteenthcenturywhentheyrisemarkedly,,,withepisodesofdeepcoldcausingglacierstoadvance,theThamesinLondontofreeze,∗SchoolofEconomics,-POD(HistoricalPatternsofDevelopmentandUnderdevelopment:OriginsandPersistenceoftheGreatDivergence)ProjectsupportedbytheEuropeanCommission’
climatologistsisthattheNorthernHemisphereabovethetropicsexperiencedsus-tainedepisodesofreducedtemperaturesbetweenthefifteenthandnineteenthcen-turies,withparticularlymarkedfallsinEurope(Mann2002,MatthewsandBriffa2005,:byhowmuchdidtheworseclimateoftheperiodreducetheharvestsonwhichpre-industrialEuropeansreliedforsurvival?Althoughwefindthatcropyieldswerestronglyaffectedbyweather,wefindlittleevidenceofvariationinEuro-peanclimate:thedistributionofsummertemperaturesappearsunchangedbe-tweenthefourteenthandtwentiethcenturies,whilewintertemperaturesremainconstantuntilthelatenineteenthcentury,afterwhichtheyrisemarkedly,(2007).Manorsbelongingtoreligiousandotherinstitutionskeptdetailedannualaccountsthatgivethemostaccurateinformationonharvestyields,outsideChina,:LowCountriessummertemperatures,andthethicknessofannualgrowthringsofIrishoaks(whichcorrelatewithsummerprecipitation).AonedegreeCelsiusfallinsummertemperaturereducedyieldsofwheatbyaround5percent,,theirpricescloselyfol-lowedyieldsofwheat,:a10percentfallinrealwagescausedbyabadharvestresultedina7percentriseinmortalityamongbothunfreetenants1Strictlyspeaking,ourresultsgiveanupperboundfortheeconomiccostoftheLittleIceAge,,becausetheupperboundforthecostoftheLittleIceAgeappearstobezero,
:TheSlutskyeffectinLowCountriessummertemperature,AD1301–;themiddlepanelshowstherawseries;(KellyandÓGráda,2010).-turesince1300,followingthestandardclimatologicalpracticeofsmoothingthedata, CDegrees CDegrees
showacoolingtrendfromthemid-fifteenthtotheearlynineteenthcenturies,withmarkedcoldepisodesinthelatesixteenth,lateseventeenth,andearlynineteenthcenturies,,whenwelookinsteadattheunsmootheddata,inthemiddlepanelofFigure1,theimpressionisoneofrandomnesswithoutstructuralbreaks,cyclesortrends,,,,andalsohistoricaltemperatureestimatesforSwitzerland,FranceandEngland;Irishoaks,,,,then,,whereasthecurrentpracticeinclimatologyistosmoothdatausingamoving4
,asannualEuropeanweatherseriesappeartobe,smoothingcanintroducespuriouscycles,aphenomenonfirstdescribedbySlutsky(19372).TheintuitivereasonfortheSlutskyeffectisstraightforward:justastossingafaircoinleadstolongsequenceswithanexcessofheadsortails,,likethebadweatherinthe1590sor1690sinFigure1,,,,:asociety’sdivisionoflabourisconstrainedbyitspopulationandbyitssurplusofresourcesoverbiologicalsubsistence;,seeSargent(1979,248–249).IntheclimatologyliteratureBurroughs(2003,24)brieflydiscussestheSlutskyeffectinanearlychapteronstatisticalbackgroundandgivesadiagramillustratinghowapplyingamovingaveragetoaseriesofrandomnumberswillgivetheappearanceofcycles,
(1981,624)claimedthathistorianswere‘psychologicallyready,eveneager’toacceptclimatechangeas‘avehicleoflong-termhistoricalexplanation’.Onlymorerecently,however,economichistoriansbeguntocombinehistorical,economic,andmeteorologicaldatainarguingforalinkbetweensecularclimatechangeandeconomictrendsinEurope(forinstanceSteckel2004,Campbell2009,KoepkeandBaten2005).,000years,theterm‘LittleIceAge’,thereisaconsensusthatmuchoftheNorthernHemisphereabovethetropicsexperi-encedseveralcenturiesofreducedmeansummertemperatures,althoughthereissomevariationoverdateswithMann(2002)suggestingtheperiodbetweenthefifteenthandnineteenthcenturies,MatthewsandBriffa(2005)1570–1900,andMannetal.(2009)(1995)’sVikingcolonyandtheendofgrape-growinginsouthernEnglandinthefourteenthcentury;theDutchwinterlandscapepaintingsofPieterBruegel(1525-69)andHendrikAvercamp(1585-1634);theperiodic‘icefairs’onLondon’sThames,endingin1814;and,astheLittleIceAgewaned,thecontractionofEurope’,howmuchdiditmatter?TheLittleIceAge’(1995,318)hasdrawnattentiontothealleged‘parallelismofclimaticandculturalcurves’asthe3ForausefulsurveyoftheoriesthatinvokeclimatetoexplainthecollapseofhistoricalsocietiesseeTainter(1988).6
(2004)haslinkedthediscoveryofadown-wardtrendinaverageadultheightstoacoolingtrendthat‘causedhavoc’innorth-ernEuropeforseveralcenturies,whileKomlos(2003)attributeshisfindingofa‘verylarge’increaseinFrenchheightsintheearlyeighteenthcenturyto‘averysubstantialriseintemperatures’.Otherhistorianswhohaveassertedastronglinkbetweenclimatechangeandeconomicconditionsinthiserainclude(Cameron,1993,74)andParker(2001,5–6).Againstsuchclaims,LeRoyLadurie(1971)anddeVries(1981)havearguedthattheeconomicand(byimplication),(2007).Whileaccountsgobackto1211,Campbell(2007),,andPostan(1975,43),ifweregresspricesonmedianyieldsandlaggedyields,−−;butfortheperiod1211––1450topredictpricefromyieldsfortheearlierperiodandcomparethesewithactualprices,wefindconsiderableunderestimatesofpriceinyearswithhighrecordedyields,supportingCampbell’ÓGráda(2010)findthatrecordsofpropertytransfersontheWinchestermanorsarefragmentaryandunreliablebefore1269,
101088664422-2-10123-3-2-10123Oak ring thicknessSummer temperature(a)Oakrings.(b):Impactofweatheronwheatyields,1270–1450Toestimatetheimpactofweatheronyieldswerunaregressionoflogyieldratiosonweather,=(β0+β0i)+(β1+β1i)(st−s¯)+(β2+β2i)rt+εitwhereyitisgrossyieldperseedonmanoriinyeart,st−s¯isthedeviationofestimatedsummertemperaturefromitsmeanvalue,βji∼N(0,σ2β).Itfollowsjthattheinterceptisthelogyieldratioinayearwithaverageweather,(PinheiroandBates,2000,).Plotsofthequantilesoftheregressionresidualsagainstquantilesofthenormaldistributionindicatethatthissemi-logspecificiationappearsadequate, yieldWheat yield
InterceptSummerRingsLoglikR˜2σα∗∗∗∗−∗∗−()()()∗∗∗∗−∗∗−()()()∗∗−∗∗−()()()∗∗−−()()()Mixedeffectsregressionoflogcerealyieldratioononestimatedsummertemperatures():σα˜.*denotessignificanceat5percent,**:Cerealyieldsandweather,1211–1500Whileintercepts,whichdenoteaverageyield,varywidelyacrossmanors,thereislittleevidenceforvariationofslopes:comparinglog-likelihoodofregres-sionswithfixedandvariableslopesproducedanimprovementinfitthatwassig-nificantatconventionallevelsonlyforthecaseofsummertemperatureonwheat,,-tieswithconvergence,,ourexpectationwasthatslightvariationsintemperatureandrainfalleachyearwouldhardlyaffectyieldsoutsidewellknownperiodsofsevereweathersuchastheheavyrainsof1315–,wheatyieldturnsouttobestronglyaffectedbyweather:aonedegreeriseinsummertemperature(),,theeffectofsummertemperaturevariesconsiderablyacrossmanors:-estyields:-9
:Correlationbetweenannualyields(abovediagonal),andnominalprices(belowdiagonal)ofcereals,1270–,toallowforpossiblenon-lineareffectsofweather,,,basedontheknownrelationshipofthevariablesbetween1766and1900,aregressionusingDutchsummertemperatureandoakringthicknessasproxiesforEnglishsummertemperatureandrainfallrespectivelywillproducecoefficientsthatare70percentand48percentrespectivelyoftheirtruevalues,andtotheextentthatmedievaltemperatureestimatesarelessaccuratethantheselaterobservations,,,whilethespringgrainsbarleyandoatsshownomeasurableeffectofrainfall,,intermsofweatherrisk,springgrainsofferedthebestinsurancetosubsistencefarmers,andhavetheaddedadvantagesofgrowingonpoorersoilthanwheat,,outsideharvesttime,wasdredge,amix-tureofbarleyandoats,KellyandÓGráda(2010)
themselves,,yieldsofotherce-realsarepoorlycorrelatedwithwheat,,sum-marizedinTable3:LowCountriesSummerandWintertemperaturefrom1301;5Frenchsummertemperaturefrom1370;Swisssummerandwintertemperatureandprecipitationfrom1525;Englishsummerandwintertemperaturefrom1660,::-ron(1998)proceduretodetectbreakpointsinaregressionofannualweatheronaconstant(includingautoregressivetermsandatrenddidnotalterourresultsma-terially).,withabreakoccurringineverycasearoundtheendofthenineteenthcentury,–,wintersappeartobecomewetteratthistime,,SwissorEnglishdata,,thereareincreasingnumbersofmissingobservationsaswegobackpast1301andtheauthorsarelessconfidentoftheiraccuracy,,missingobservationsduringthe14thandearly15thcenturies(30forwinter,11forsummer)
StartMeanSDK-Wρ∗∗−∗∗−−∗∗−−−∗∗−∗−−∗−−−−−−−−−∗∗ρandtrendcoefficients,andsquaredcorrelationR2forAR(1)regressionwithtimetrend.**denotesacoefficientsignificantat1percent,,Irishoaksin1993,:,,,theresultsineachcasearethesameasthep-valueofaregressionofannualtemperatureondummy12
,theoutcomeisfarfromsignificance:,SwitzerlandandFranceaf-ter1728,thetestissignificanthoweverat1percent,’shonestsignificantdifferencecalculationofp-values,wefindthattheEnglishresultisdrivenbytheearliest,andpossiblylessaccurate,observationsinthelateseven-teenthcentury,:,significanceoccursbecauseofthedifferencebetweentheearlyeighteenthandnineteenthcenturies;-centuries,usingaFligner-Killeentest,ledtostrongrejectionineverycase:,althoughtheeffectissmall:—exceptIrishoakswhicharelarge,slowgrowingorganisms—,apartfromoaks,,,-turebefore1728,althoughthispossiblyreflectsautocorrelationinthestartdates6TheexceptionisIrishoaks,
,onlySwisssummersbefore1812showsignificantautoregression,,theyarenotentirelyrandom:,Swissglaciersfluctuateno-tably,expandingfromthemid-fifteenthcenturyuntil1650,contractinguntil1750andthenexpandingagainuntil1850(MatthewsandBriffa,2005,18–19).GlacierscanbeseenasaphysicalexampleofaSlutskyeffect:theirextentrepresentsamovingaverageprocessoftemperatureandprecipitationoverprecedingyears,-tweenannualtemperatureandprecipitation,,(calculatedusingdetrendedfluctuationanalysis),withhighervaluesindicatingtrending,,withthesummertemperatureseriesshowingsomesmalldegreeoftrending,.(1996),only14
HσHBDSTLGGARCHσσHaretheestimatedHurstexponentanditsstandarderror;BDSisthep-valueofaBDStestwithembeddingdimensionof3withanepsilonvalueof2standarddeviations;TLGisthep-valueofaTerasvirta-Lin-Grangertestfornon-linearityinmeans;GARCHisthecoefficientonlaggedvarianceinanAR1modelwithGARCH(1,1)errors,andσ:,LinandGranger(1993)testfornon-linearityinmeansisreportedand,inthiscase,-sion,:yt=β0+β1yt−1+εtwherethevarianceσ2ofεttfollowsaGARCH(1,1)processσ2=αt0+α1ε2t−1+α2σ2t−α1wasclosetozero,andthefinaltwocolumnsofTable4reportthecoefficientandstandarder-rorofthevariancecoefficientα,withtheexceptionagain15
oftheunusualFrenchsummerestimates,-ances;,annualweathershowsweakdependenceinlevels,-runweatherconstructionsforEuropeindicatethatannualweathershowslittledependencefromyeartoyear,-tionsrelyattimesonstrongassumptions,,-ropeancitiesintheCDIAC“UpdatedGlobalGridPointSurfaceAirTemperatureAnomalyDataSet:1851-1990,”,-runwinterreconstructionsinTable3,thewintertemperatureseriestendtobreakinthelatenineteenthorearlytwentiethcenturies,withahighertemperatureafterthebreak,,,thedataappearaconsiderablynoisierwithbreaksoccurringinonecitybutnotinnearbyneighbours,
StartMeanSDK-Wρ−−−−−−−−−−−−−∗∗−−−−−−∗−−−−−−∗∗−−−−−−−−∗∗−ρandtrendcoefficients,andsquaredcorrelationR2forAR(1)regressionwithtimetrend.**denotesacoefficientsignificantat1percent,*at5percent,:,however,
StartMeanSDK-Wρ−∗∗∗−∗−−∗−∗∗−∗∗−∗∗−−−∗∗−−∗∗∗∗∗∗−−∗∗−−−∗∗−−−−−−−−(o1w)edbyautoregressivecoefficientρandtrendcoefficients,andsquaredcorrelationR2Aforregressionwithtimetrend.**denotesacoefficientsignificantat1percent,*at5permceanxt,-Bisthep-valueoftheLjung-Boxtestwithlagof10years,:,-structions,solongastheyarenotsmoothedbeforeanalysis,,annualsummerweatherconditionsinwestern18
Europebetween1300and2000areweaklydependentdrawsfromadistributionwithafixedmeanbutautocorrelatedvariance;whilewinterweatherbehavessim-ilarlyuntilaround1900,(1995)—whichformostpeopleisthemostsalientfactabouttheLittleIceAge—wasduetoOldLondonBridgewhicheffectivelyactedasadam,,evenduring1963whichisthethirdcoldestwinter(after1684and1740),Avercampmadealivingfromhisformulaicwinterscenes,butsuchsnowscapesrarelyfeatureintheworkofhisbetterknowncon-temporaries,suchasAlbertCuyp,JanvanGoyen,’siconic‘HuntersintheSnow’’sVikings,competitionforresourceswiththeindigenousInuit,thedeclineofNorwegiantradeinthefaceofanincreasinglypowerfulGermanHanseaticLeague,thegreateravailabilityofAfricanivoryasacheapersubsti-tuteforwalrusivory,overgrazing,plague,andmaraudingpiratesprobablyallplayedsomeroleinitsdemise(Brown,2000);andevenifweatherdidworsen,themorefundamentalquestionremainsofwhyGreenlandsocietyfailedtoadapt(McGovern,1981).ThedisappearanceofEngland’sfewvineyardsisassociatedwithincreasingwineimportsafterBordeauxpassedtotheEnglishcrownin1152,,thedeclineofwheatandryecultivationinNorwayfromthethir-teenthcenturymayowemoretolowerGermancerealpricesthantemperature19
change(Miskimin,1975,59).Withworseningclimatewewouldexpectwheatyieldstofallrelativetobarleyoats(seeTable1)whereasApostolidesetal.(2008,Tables1A,1B)findthatbetweentheearlyfifteenthandlateseventeenthcentury,wheatyieldsshownotrendrelativetooats,,,,,,,(Maddison,2009).Appendix:DataSourcesandEstimation•MonthlymeanCentralEnglandtemperaturefrom1659arefrom
•EuropeancitiessummerandwintertemperaturesareaveragesofJunetoAugust,“AnUpdatedGlobalGridPointSurfaceAirTemperatureAnomalyDataSet:1851-1990”availableat
Cameron,:,,1211–://,7,21Campbell,.“NatureasHistoricalProtagonist.”,Isabelle,IsabelleYiou,NicolasViovy,BernardSeguin,.“Graperipeningasapastclimateindicator.”Nature432(289–290).20deVries,:,:,7Kelly,MorganandCormacÓGrá,,7,10Koepke,.“TheBiologicalStandardofLivinginEuropeduringtheLastTwoMillennia.”EuropeanReviewofEconomicHistory9(01):61–,.“AnAnthropometricHistoryofEarlyModernFrance.”Eu-ropeanReviewofEconomicHistory70:159–,,:,19LeRoyLadurie,,TimesofFamine::,,GDPandPerCapitaGDP,:
Mann,,,6Mann,MichaelE.,ZhihuaZhang,ScottRutherford,,,DrewShindell,CasparAmmann,.“GlobalSignaturesandDynamicalOriginsoftheLittleIceAgeandMedievalClimateAnomaly.”Science326:1256–,6Matthews,.“The’LittleIceAge’:Re-evaluationofanevolvingconcept.”GeografiskaAnnaler87A:17–,6,14McGovern,,,-bridge:–,,:,::,,:,:,:,:,.“TheSummationofRandomCausesastheSourceofCyclicProcesses.”Econometrica5:105–
Steckel,.“NewLightonthe"DarkAges":TheRemarkablyTallStatureofNorthernEuropeanMenduringtheMedievalEra.”SocialScienceHistory28:211–,7Tainter,:,T,.“PoweroftheNeuralNetworkLinearityTest.”JournalofTimeSeriesAnalysis14:209–,.,,:MemoriesoftheFuture?,,,:
