Baixe o app para aproveitar ainda mais
Prévia do material em texto
~ ;JJ ~ I -l I l:.. -:] ~ I-f - .. I ~ I ; -1 = = >t...~ ~ c l r -' Ir' ;JJs ~ 1 C I r- I c- ~ ~ c:: I I C -t IIr •• ; -1 l I t ,~ Firstpublished1996 Copyright© PaulR. Krugman,1996 CompositionbyMeganH. Zuckerman v 53 55 61 61 63 69 71 75 1 9 9 13 15 21 31 36 39 43 44 47 47 49 51 PART I Embryos,Earthquakes,andEconomics Self-Organization in Space The vonThiinen-Mills Model CentralPlaceTheory Schelling'sSegregationModel EdgeCity Dynamics Complex Landscapes The Emergenceof Order An Urban Mystery Meteorites,Earthquakes,andCities Simonon Zipf Principles of Self-Organization OrderfromInstability OrderfromRandomGrowth Where We Stand Contents PART II Self-Organizationin Time andSpace Dynamics in Self-Organizing Systems Temporal Self-Organization Two QuestionsabouttheBusinessCycle NonlinearBusinessCycleTheory PercolationEconomics PhaseLockingandtheGlobalBusinessCycle Models of Spatial Self-Organization Preface 6 7 1 8 2 3v: (7') 4 f' ....J ..3J 5 <:> 95-31593 CIP ./ t/ r I (' •.. t:., l., ("' J\ IY\ I':J.() \. 9:,(a- \ Cl Cj Exceptin theUnitedStatesofAmerica,thisbookissoldsubjecttotheconditionthat it shallnot,bywayof tradeor otherwise,belent,resold,hiredout,or otherwisecir- culatedwithoutthepublisher'spriorconsentin anyformof bindingor coverother thanthatin whichit ispublishedandwithoutasimilarconditionincludingthiscon- ditionbeingimposedonthesubsequentpurchaser. I_I Libraryif CongressCataloging-in-PublicationData Theself-organizingeconomyI PaulR. Krugman. p. em. Includesbibliographicalreferencesandindex. ISBN 1-55786-698-8(he) ISBN 1-55786-699-6(pb) 1. Economics. 2. Self-organizingsystems.I. Title. HB199.k75 1995 338.9--dc20 BlackwellPublishers,Ltd. 108CowleyRoad OxfordOX4 1JF UK All rightsreserved.Exceptfor thequotationof shortpassagesforthepurposesof criti- cismandreview,nopartmaybereproduced,storedin aretrievalsystem,or transmit- ted,in anyformorbyanymeans,electronic,mechanical,photocopying,recordingor otherwise,withoutthepriorpermissionof thepublisher. TherightofPaulR. Krugmantobeidentifiedasauthorofthisworkhasbeen assertedin accordancewiththeCopyright,DesignsandPatentsAct 1988. BlackwellPublishers,Inc. 238MainStreet Cambridge,Massachusetts02142 USA BritishLibraryCataloguingin PublicationData A CIP cataloguerecordforthisbookisavailablefromtheBritishLibrary. Printedin theUSA byBookCrafters Thisbookisprintedonacid-freepaper TheEdgeCityModel A CentralPlaceModel Simon'sUrbanGrowthModel 9 ConcludingThoughts 10 Appendix:The Evolutionof CentralPlaces UrbanMorphogenesis:TheEdgeCityModel A CentralPlaceModel References Index IV SELF-ORGANIZING ECONOMY 76 88 92 99 101 101 106 117 119 Preface Every oncein a while theworld economyplungesinto a severe recession.Someof theseinternationalslumpsappearto be caused by particularnoneconomicevents:wars,disruptionsof oil supply. Others,however,haveno obviouscause- andtheirglobalscaleis hardto explainin termsof theconventionallymeasuredlinkages amongnationaleconomies. Everyoncein a (muchlonger)while, thepaleontologiststellus, theworld experiencesmassextinctionsthatwipe out mostof the extantspecies.Someextinctionsappeartohaveexternalcauses,like the comet whose impact coincided with the demiseof the dinosaurs.Others, however,havenot beentied to any obvious cause.Sometheoristswho simulateevolutionon theircomputers claimthatthisis asit shouldbe: theirmodelspredictoccasional, spontaneousmassextinctionsevenin theabsenceof anyexternal shocksto thesystem. Can thesetwo paragraphshaveanythingto do with eachother? Is therea sensein which a globalslumpis somethinglike a mass extinction? Here is anotherparallel.When you look at thesizes(however measured)ofmanyenormouslycomplexphysicalorbiologicalphe- nomena,thedistributionof thosesizesfor somereasonturnsoutto bewelldescribedbyaverysimplepower law: thenumberof objects (earthquakes,meteorites,species,and - perhaps- extinctions) whosesizeexceedsS is proportionalto s-a,wherea is not only a mysteryparameterbutoftenturnsout,weirdly,tobearoundnum- ber,like 1 or 2. Amongthemostspectacularexamplesof apower law,however,is onethatinvolveseconomicsratherthanphysical science:the sizedistributionof cities.In the United States,the " numberof citieswhosepopulationexceedsS is, simply,propor- tionalto l/S: thereare40citieswith morethanamillionpeople,20 with morethan2 million, and9 (Houstonfell a bit short)with morethan4 million! Socialscientistsarenormallysuspiciousof peoplewho wantto importconceptsfromphysicalorbiologicalscience,andwith good reason:thehistoryof suchefforts,fromsocialDarwinismtosystems dynamics,hasbeenlittleshortof disastrous.Nonetheless,thistime thingsmaybedifferent;thereis agenuinelyinterestinginterdisci- plinarymovementofwhicheconomicsoughttobeapart. In thelastfewyearstheconceptofself-organizingsystems- ofcom- plexsystemsinwhichrandomnessandchaosseemspontaneouslyto evolveinto unexpectedorder- hasbecomeanincreasinglyinflu- entialideathatlinkstogetherresearchersin manyfields,fromarti- ficial intelligenceto chemistry,from evolutionto geology.For whateverreason,however,thismovementhassofarlargelypassed economictheoryby. It is timeto seehow thenew ideascanuse fullybeappliedto thatimmenselycomplex,but indisputablyself- organizingsystemwe calltheeconomy. In thisbookI trytoshowhowmodelsof self-organizationcanbe appliedto manyeconomicphenomena- how the principleof "orderfrom instability,"which explainsthegrowthof hurricanes andembryos,canalsoexplaintheformationof citiesandbusiness cycles;how the principle of "order from randomgrowth" can explainthestrangelysimplerulesthatdescribethesizesof earth- quakes,meteorites,andmetropolitanareas.I believethattheideas of self-organizationtheorycanaddsubstantiallytoourunderstand- ingof theeconomy;whatevertheirultimateusefulness,theseideas areveryexciting,andplayingaroundwith themis tremendousfun. Finally,anoteon style.This bookbeganastheMitsui lectures, which I gaveat the Universityof Birminghamin March 1994. Although I havenot triedto maintainthe lectureformat,I have allowedmyselfto retainsomeof thelicenseusuallygrantedin such alectureseries:thisbookiswrittenin aninformalstyleandcontains morethanafewwild speculations.Nonetheless,I havetriedtoget thingsrightwhenI can;andI amgratefulfordiscussionswith par- ticipants in seminarsat Birmingham, UCLA, Chicago, and Stanfordthathelpedmecorrectsomeseriouserrors.In particular,I wouldlike to thankMike WoodfordatChicagofor pushingmeto testapethypothesisto itswell-deserveddestruction. vi SELF-ORGANIZING ECONOMY PART-----1----- Embryos,Earthquakes,andEconomics Isaneconomicslumplikeahurricane,or isit morelikeanearth- quake?Is a growing city like an embryo,or is it more like a meteorite? Thesequestionsmaysoundlike children'sriddles,to whichthe answeris somekind of badpun; or maybetheysoundlike Zen koansalongthelinesof "What is thesoundof onehandclapping?" in which theabsurdityof thequestionis meanttojolt thelistener intoahigherstateof awareness.But myaimis neitherto telljokes nor tohelpyouachieveanenlightenmentthattranscendsrationali- ty.On thecontrary,in thecourseof thisbook I hopeto convince you thattheseareperfectlyreasonablequestionsto ask,thatthere maywellbeamorethanpoeticsensein whichadeepeningslump resemblesanemerginghurricane,in whichagrowingcityisquitea lot likeadevelopingembryo. In makingthesekindsof analogiesI am not, of course,being completelyoriginal.Thereis abroadandgrowinginterdisciplinary movementin thephysicalandbiologicalsciences- oftenreferredto as the studyof complexity- that looks for exactlysuchparallels betweenseeminglydisparatephenomena.For example,in the courseof my preparationsfor thisbook I ranacrossanarticleon"percolationtheory"thatcasuallylisted15areastowhichthebasic approachapplied,rangingfromatomicnucleitogalaxies.But sofar thismovementhaslargelypassedeconomicsby. Actually,letmequalitythatstatement.Peoplewho writebooks or conveneconferenceson complexityalmostinvariablyassertthat the emergingfield will makegreatcontributionsto the studyof thatimmenselycomplicatedsystemwe calltheeconomy.Indeed, theSantaFe Institute,oneof thehotbedsof thisstyleof research, wasinitiallyfundedby Citibank largelybecauseJohn Reed, the bank's CEO, hoped that researchon complexsystemswould improveeconomicforecasting.For whateverreason,however,the authorsof articlesandbookson complexityalmostnevertalk to seriouseconomistsor readwhat seriouseconomistswrite; as a result,claimsabouttheapplicabilityof thenewideasto economics areusuallycoupledwith statementsabouthow economieswork (andwhateconomistsknow)thatseemsoill-informedasto make anyeconomistwho happensto encounterthemdismissthewhole enterprise. But it doesnot haveto be thatway.What I amgoingto claim, andI hopedemonstrate,in theselecturesis thatsomeof theideas thatcomeout of theinterdisciplinarystudyof complexsystems- the attemptto find commonprinciplesthatapplyacrossa wide varietyof scientificfields,fromneuroscienceto condensedmatter physics- are,in fact,usefulin economicsaswell.That is,you can understandandrespecttheeconomictheorywe alreadyhaveand stillfindwaysbothto improveit andtobuildbridgestootherfields bytakingintoaccounttheideasof theseinterdisciplinarytheorists. BeforeI goanyfurther,I hadbetterexplainalittlebetterwhatI amtalkingabout.What is this interdisciplinaryeffortthatI have beenalludingto,andhowcanit teacheconomistsanythingnew? Manyof thepeoplewho thinkthatearthquakes,embryos,cycles, andcitieshaveall got somethingto do with oneanotherdescribe theirfield asthestudyof "complexity,"basedon theinsightthat complicatedfeedbacksystemshavesurprisingproperties.If that wereall therewereto theprogram,however,economistswouldbe entitledto changethechannel.If economistsdo understandone thingmuchbetterthanthelaypublic,it is thesheercomplexityof the economicsystemandtheimportanceof feedbacks.After all, whatisgeneralequilibriumtheorybutaformalizationofthepropo- sitionthateverythingin theeconomyaffectseverythingelse,in at leasttwo ways?If you haveevertriedto explainto a roomfulof engineerswhy highermanufacturingproductivitywill probably reduce,not increase,manufacturingemployment- andwhy high- er productivitywill not necessarilyreducethe tradedeficit- you quicklyrealizethateconomistsarebetter,not worse,thanmost physicalscientistsatunderstandingtheimportanceof feedbackin complexsystems. A seconddefinition,by Philip Anderson,the Nobel laureate physicistwho mayperhapsberegardedasthefatherof thefield, is thatcomplexityis thescienceof "emergence."That is, it is about how largeinteractingensembles- wheretheunitsmaybewater molecules,neurons,magneticdipoles,or consumers- exhibitcol- lectivebehaviorthatisverydifferentfromanythingyoumighthave expectedfrom simplyscalingup the behaviorof the individual units.(Thebehaviormayalsobeoddlysimilarto thatof ensembles ofotherwiseverydifferentunits:collectionsofneuronsmaybehave alot likecollectionsof magneticdipoles.)Anderson'sprimeexam- pleof anemergentpropertyis theliquidnessof water,which is in no senseanextrapolationof someprimordialliquidnessof individ- ualwatermolecules. Hereagain,however,we haveadefinitionthatsoundslikewhat economistsalreadyunderstandprettywell. When Adam Smith wroteof thewaythatmarketsleadtheirparticipants,"asif by an invisiblehand,"to outcomesthatnobodyintended,whatwashe describingbutanemergentproperty?And examplesof emergence aboundin economictheory- weneedonlynotethewaythatcom- petitivemarkets,in whicheachindividualis strivingonlyforhisor herownprofit,actasif theparticipantswerecollectivelytryingto maximizethesumof consumerandproducersurplus,conceptsof whichtheyaregenerallyunaware.(Nor is thisonlytheory:experi- mentalmarketsin which theparticipantsareassignedpayoffsand thenmakebidsforandoffersof unitsof anotionalcommoditydo, in practice,comeverycloseto maximizingaggregatesurplus,even thoughthe participantsnot only are not trying to achievethat objective,theydonotevenknowwhateachothers'payoffsare.) There is, however,a thirddefinitionof thisfield thatdoesnot soundlike somethingthateconomistsalreadydo, or at leastnot whattheydoroutinely.This istheviewthatwhatlinksthestudyof embryosandhurricanes,of magneticmaterialsandcollectionsof neurons,is thattheyareallself-organizingsystems:systemsthat,even when theystartfrom an almosthomogeneousor almostrandom state,spontaneouslyformlarge-scalepatterns.One daytheairover aparticularpatchof tropicaloceanis no differentin behaviorfrom theairoveranyotherpatch;maybethepressureis abit lower,but 2 SELF-ORGANIZING ECONOMY Embryos,Earthquakes,andEconomics 3 thedifferenceisnothingdramatic.Overthecourseof thenextfew days,however,that slight dip in pressurebecomesmagnified throughaprocessof self-reinforcement:risingairpullswatervapor up to analtitudeatwhichit condenses,releasingheatthatreduces thepressurefurtherandmakesmoreair rise,until thatparticular pieceof theatmospherehasbecomeahuge,spinningvortex.Early in theprocessof growthanembryois acollectionof nearlyidenti- calcells,but(oratleastsomanybiologistsbelieve)thesecellscom- municatewith eachotherthroughsubtlechemicalsignalsthatrein- forceandinhibiteachother,leadingto the"decision"of somecells tobecomepartsof awing,otherspartsof aleg. Is theeconomya self-organizingsystemin thissense?Of course it is.Think aboutametropolitanarea- even,or ratherespecially,a modernmetropolitanarealikegreaterLosAngeles,with no clearly definedcenter.Is anurbansprawllike LA ahomogeneous,undif- ferentiatedmass?No - it is a patchworkof areasof verydistinct character,rangingfromKoreatowntoHollywood,WattstoBeverly Hills. And it contains(accordingto therecentbook Edge City by Joel Garreau)no fewerthan16"edgecities,"newlyemergedbusi- nesscenters,eachof whichincludesatleast5 million squarefeetof office spaceand20,000workers,wherethelow-rise sprawlsud- denlygivesway to tall buildingsandmultistoryparkinggarages. What issostrikingaboutthisdifferentiationisthatit issoindepen- dentofphysicalgeography:therearenoriverstosetboundaries,no bigdowntowntodefineagradientofaccessibility.(OK, thebeach- esandthefreewayscreateabitofexogenousstructure,butgrantme alittlelicense.)The strongorganizationof spacewithin metroLos Angelesis clearlysomethingthathasemerged,notbecauseof any inherentqualitiesofdifferentsites,butratherthroughself-reinforc- ing processes:Koreansmoveto Koreatownto be with Koreans, beautifulpeoplemoveto BeverlyHills to bewith otherbeautiful people.And whenanobserverlikeGarreautracesthedevelopment of edgecities,heis immediatelydrawnto metaphorslike "sponta- neouscombustion"and"criticalmass"- clearindicatorsthatheis tryingto describea self-organizingsystem. I havestartedwith thewaythatcitiesspontaneouslyevolveapat- ternof sharplydistinctdistricts,becausetheeconomicself-organi- zationof spaceis somethingwe can all relateto our immediate experience.But I wouldassertthatthereareprocessesin theecon- omy thatproducetemporalself-organizationaswell. I referto the businesscycle:thepulsesofexpansionandcontractionaroundarel- ativelystablelong-run trend.(No, I don'tbelievein Kondratieff waves.)Somerecessionsandrecoveriesareclearlysetoffby specif- ic, essentiallyexogenouseventslikeoil crises.Overthelongsweep of history,however,mostboomsandslumpshavehadno obvious externalcause.Most notably,themotherof all economicslumps, thecontractionfrom 1929to 1933,cameasit wereout of a clear bluesky.But then,sodohurricanes. In fact,letmeexplainwhyaslumpmaybequitealot likeahur- ricane.A hurricaneisaself-reinforcingprocess,in whichanupdraft thatpullswatervaportoalevelatwhichit condensestherebyreleas- esheat,which in turn reinforcestheupdraft.Aslumpis alsoself- reinforcing:fallingoutputcausesfirmstoslashtheirinvestmentand consumersto reducetheirspending,therebyreducingoutputstill further.But wait,thereismore.A hurricanecannotgoon forever. The fuel for its violenceis thewatervaporthatexistsin relative abundanceabovea warmtropicalocean.But the hurricaneitself coolsthatoceansurface- indeed,in awaythewholepointof trop- icalstormsis thattheyareNature'swayof transferringsolarenergy fromthetropicstohigherlatitudes.And sohurricanes,eventhough theyareself-reinforcingin the shortrun, areself-limitingin the longrun.What aboutaneconomicslump?Well, it leadsto falling prices,or at leastdisinflation,which graduallyincreasesthe real moneysupply.The low or negativenetinvestmentin adeepslump mayalsoleadtoagrowingbacklogofpotentiallyprofitableprojects. And sowemayarguethataneconomicslumpisalsoaself-limiting process,evenif thereis no deliberateor effectivegovernmentpoli- cytogettheeconomymovingagain. By theway,Americanswho watchtheweathernewsknow that hurricanessustainthemselvesby moving,gainingstrengthwhen theypassoverfreshtractsof warmwater;if thishasanyanalogyin businesscycles,I amunawareof it. But I didnotclaimthatthepar- allelwasperfect.And you mayarguethatmy descriptionof what goeson in atypicalrecessionisnottooaccurate- ormaybethereis no suchthingasa "typical"recession,in which caserecessionsare morelikeearthquakesthanlikehurricanes.But we shallgetto that in PartII. While I ammakingasides,let mealsotakethe opportunityto makeanotherpoint: self-organizationis not necessarily,or even presumptively,agoodthing.I think it is fairto accusemanyof the 4 SELF-ORGANIZING ECONOMY Embryos,Earthquakes,andEconomics 5 t " writerson complexity,especiallybut not only the morepopular ones,of fallingintothisfallacy.Book titleslikeOrderoutif Chaos(by Nobel laureateIlyaPrigogineandI. Stenger,butwith a Foreword by,believeit or not,Alvin Toffler),or Complexity:Life at theEdgeif Chaos(byR. Lewin) comeperilouslycloseto makingself-organi- zationa kind of mysticalgoal.Even StuartKauffman,whoseThe Originsif Orderis a seriousandstimulatingtract(if we ignorethe inevitablycarelessandill-informedsectionon economics),talksfar too casuallyaboutcoevolutionarysystemsthatmaximize"average fitness"- a surelymeaninglessconceptwhen the fitnessof each speciesis definedatleastpartlyin termsof howwell it copeswith competitionandpredationfromothers.Luckily,if weareatallseri- ousabouttheeconomicsof self-organizationweimmediatelyreal- ize thatno valuejudgmentis implied.An economywith a strong businesscycleexhibitsmore temporalself-organizationthan an economythatgrowssmoothly,butmostof uswouldratherlivein the latter.A city whoseraciallyintegratedcommunitiesunravel, producinghugesegregateddomains,becomesmorespatiallyorga- nized,butnotbetter,in theprocess.Self-organizationis something we observeandtry to understand,not necessarilysomethingwe want. I haveassertedthatthestudyof self-organizationis something thateconomistsdo notdo, or atleastdo not do routinely.What I havejust saidabouteconomicslumpsis,however,notatallnewor original.On thecontrary,I havejust givenyou a looseversionof thenonlinearbusinesscycleliteraturethatflourishedin the 1940s and1950s,with contributorsincludingeconomistsof thestatureof John Hicks,RichardGoodwin,andJamesTobin. For thatmatter, urbaneconomistshavehardlybeenunawareof theself-organization of metropolitanareas,andtherehavebeensomenotableeffortsto modelthecreationof urbansubcenters,asedgecitiesarecolorless- ly knownamongtheprofessionals.(I havein mindparticularlythe pioneeringwork of FujitaandOgawa.) And yetI thinkit is fairto saythatfeweconomistshaveexplicit- ly realizedthattheyweretrying to modelself-organization;few havetriedto drawtheparallelbetweenself-organizationin space and self-organizationin time;few havetried to usesomeof the techniquesforunderstandingself-organizationthathaveevolvedin otherfields.Furthermore,althoughsomedeeplyinsightfulthinkers moreor lessconsciouslyhavewrittenaboutself-organizationin the economy,theirwork hasbeenneglectedby theprofession(andis completelyunknownoutsideof it). Nonlinearbusinesscycletheo- ry, for example,wastechnicallyfaraheadof itstimeandmakesthe latereffortsto shoehorn"catastrophetheory"intoeconomicslook primitive.But who remembersit? (I maybetheonlyeconomistin mygenerationwho hasevenheardof it.) ThomasSchellingwrotea remarkableessayon thedynamicsof segregationin hiswonderful 1978book MicromotivesandMacrobehavior,but thebook hadlittle impactatthetimeandis stillunderappreciated. I thinkthatI canexplainwhythepioneersof self-organizationin economicshavebeeneitherneglectedor forgotten,andI will talk aboutthosereasonslaterin theselectures.For now,however,letus putintellectualhistoryasideandturnto doingsomeeconomics. Here is myplanof action.In thisfirstpart,I amgoingto do a quickrun-throughof stories- I do not want to dignifythemby callingthemmodels- aboutwaysin whichtheeconomyorganizes itselfin space.(I startwith spatialself-organizationfor two main reasons.First,asanempiricalmatterI amonmoresolidground:the self-organizingspatialcharacterof theeconomyisobvioustoevery- one,althoughnotusuallyunderthatname.Second,I knowwhatI believeaboutspatialeconomics,but I amstillfairlyagnosticabout the macroeconomicsof businesscycles.)Along the way I shall describetwo broad,andseeminglyparadoxical,principlesof self- organizationthataresuggestedby thesestoriesandthatarealso commonin much of the literatureon self-organizationin other fields.Just towhetyourappetite,letmegivetheseprinciplesnames: "orderfrominstability"and"orderfromrandomgrowth." In thesecondpart,I will startby showinghow theseprinciples mayapplynot only to spacebut to time: in particular,how we might think of the businesscycleas temporalself-organization. Thenweshallcomebacktothespatialeconomy.I will describetwo moreorlessfull-fledgedmodelsthatillustratetheprincipleoforder frominstabilityandonethatillustratestheprincipleof orderfrom randomgrowth.EventhereI will golighton theequations,stress- inganintuitiveexplanation.But therewill beatechnicalappendix, whichderivestheresultsin all thegorydetailI canmanage(which isnotmuch). So letusbeginour tourwith a look atthewaythateconomies organizespace,particularlythespacewithinmetropolitanareas. 6 SELF-ORGANIZING ECONOMY Embryos,Earthquakes,andEconomics 7 -----1----- Self-Organizationin Space THE VON THUNEN-MILLS MODEL How do economistsroutinelydealwith thequestionof how the economyorganizesitsuseof space?The shortansweris thatmost- ly theydo not dealwith thequestionatall. Indeed,thereis some- thingstrangeabouttheway thatmostof our professionneglects anythinghavingto dowith whereeconomicactivitieshappen.For example,theverypopular(andalmost900-page-Iong)economics principlestextbookbyWilliam BaumolandAlanBlindercontains notasinglereferenceto "cities,""location,"or "space"in itsindex. The rival,1100-pagetextbyJosephStiglitzdoescontainonerefer- enceto cities,which turnsout to occurin a brief discussionof rural-urbanmigrationin less-developedcountries.Considering how muchtimemostpeoplespendin trafficjams,how manyfor- tunesaremadeandlostin realestate,thewaythatwe turnablind eyetospatialeconomicsislittleshortofeerie.I stronglysuspectthat thisneglectiscloselyrelatedto thethemeof thisbook:asaprofes- sionwe areimplicitlyawarethatto understandcitiesandspatial economicsgenerallywe mustcopewith issuesof self-organization andthatratherthanfacewhatseemtobeintractableissueswesim- plyavertourgaze. Anyway,whenwe do dealwith thequestionof theorganization of space,asurbaneconomistsatleastmust,we generallyturn to a classofmodelspioneeredin theearly19thcenturybyvonThiinen. Manyofyouareprobablyfamiliarwith thevonThiinenmodel,but I wantto run throughit brieflyto makea coupleof pointsabout complexity,emergence,andself-organization. 9 Von Thiinen envisagedanisolatedtownsuppliedby farmersinthesurroundingcountryside.He supposedthatcropsdifferin both theiryieldperacreandtheirtransportationcosts,aswell asallow- ingfor thepossibilitythateachcropcouldbeproducedwith differ- entintensitiesofcultivation.And heaskedtwoquestionsthatmight seemtobeverydifferent:How shouldthelandaroundthetownbe allocatedto minimizethecombinedcostsof producingandtrans- portinga givensupplyof food to the town?How will the land actuallybe allocatedif thereis an unplannedcompetitionamong farmersandlandowners,with eachindividualactingin his or her perceivedself-interest? We all know the answerto the secondquestion.Competition amongthefarmerswill leadtoagradientoflandrentsthatdeclines from amaximumatthetowntozeroattheoutermostlimit of cul- tivation.Each farmerwill be facedwith a tradeoffbetweenland rentsandtransportationcosts;sincetransportationcostsandyields differamongcrops,theresultwill beapatternof concentricrings of production.In equilibriumthelandrentgradientmustbe such asto inducefarmersto growjust enoughof eachcropto meetthe demand,andit turnsoutthatthisconditiontogetherwith thecon- ditionthatrentsbezerofor theoutermostfarmersufficesto fully determinetheoutcome. Figure1.1illustratesschematicallythetypicaloutcomeof avon Thiinenmodel.The upperpartof thefigureshowstheequilibrium "bid-rent" curves,therentthatfarmerswouldbewilling topayat anygivendistancefromthetown,for threecrops.The heavyline, the envelopeof the bid-rent curves,definesthe rent gradient. Along eachof thethreesegmentsof thatlinegrowersof oneof the cropsarewillingtopaymoreforlandthantheothers.Thusonegets concentricringsof cultivation,with a quartersectionof thelayout shownin thebottomhalfof thefigure. It isworththinkingaboutthisoutcomeforamomentin termsof thevariousdefinitionsthathavebeenofferedof whatcomplexity theoryisallabout.If it is reallyjust aboutcomplexity,vonThiinen modelsareprobablycomplicatedenoughto qualify.After all, the problemof whichcropstogrowwhereis not thateasy:byallocat- ing anacreof landnearthecity to someonecrop,you indirectly affectthecostsof deliveringallothercrops,becauseyouforcethem tobegrownfurtheraway.Exceptin thecasewherethereisnopos- sibilityofvaryingthelandintensityof cultivation,it isbynomeans Figure1.1 The Von ThiinenModel. Competitionfor landarounda townleadsto theemergenceofconcentricringsofproduction 11 Cattle Bidrent Distance from center Wheat Vegetables Self-Organizationin Space trivialtodetermineeitherwhatshouldbedoneorwhatwill happen in anunplannedmarket. SupposethatyouinsteaddefinethesubjectasAndersondoes,as astudyofemergence.Then surelythevonThiinenmodelqualifies. At the mostobviouslevel,the concentricring patternis hardly somethingthatis immanentin themotivesof thefarmers.Indeed, theconcentricringswill emergeevenif no farmerknowswhatany- oneelseisgrowing,sothatnobodyisawarethattheringsarethere. Evenmorestrikingis theresultthateconomicshastrainedusto expectbut that remains startling (and implausible)to most noneconomists:theunplannedoutcomeis efficient,is indeedthe sameastheoptimalplan.More specifically,unplannedcompetition will allocatecropsto landin a waythatminimizesthetotalcom- SELF-ORGANIZING ECONOMY10 binedcostof producingandtransportingthecrops- notincluding thelandrent.This is surelyasniceanexampleof emergenceasyou couldwant.All farmersaretryingtomaximizetheirincomeandare thereforeverymuchconcernedwith landrents,yettheircollective behaviorminimizesafunctionin whichlandrentsdo notappear. On two of thecriteriathathavebeenusedto definethisenter- prisecalledcomplexity,then,thevon Thiinen modelqualifies.And yetI doubtthatmanycomplexitytheoristswouldagreethatthiswas thekind of economicstheyhadin mind.Avon Thiinen modelis not like a hurricane,or an embryo,or a neuralnetwork.Why? Becauseit isnotself-organizing.The concentricringsofproduction formaroundatownwhoseexistenceis simplyassumed.That does notmakeit abadmodel,butit doesmakeit alimitedone.If your questionisnotsimplyhowlanduseisdeterminedgivenapre-exist- ing town,butratherhowlanduseis determinedwhenthelocation of thetownor towns- indeed,theirnumberandsize- arethem- selvesendogenous,thevonThiinen modeloffersno help. This limitationbecamepainfullyclearwhen the von Thiinen modelwaspresentedin a newform in the1960sby Edwin Mills, who substitutedcommutersfor farmersand the centralbusiness districtfor theisolatedtown,toarriveathisnow classicmodelof a monocentriccity. Again, you areprobablyfamiliarwith thebasicideaif not the details.We imagineacitypopulatedby commuterswho wantland to liveon butmustwork in acentralbusinessdistrict;theremaybe differenttypesof peoplewho varyin thevaluetheyplaceon the timetheyspendcommuting,theamountof landtheywantto live on, theirwillingnessto substitutefancyhousesfor largebackyards, andsoon. As in thevonThiinen model,competitionestablishesa landrentgradient,which sortspeopleout into a structureof con- centricrings;andasin thatmodel,decentralizedlocationchoices lead,throughnobody'sintention,to anefficientoutcome. Mills's 1967paper introducingthe monocentric city model launchedahugetheoreticalandempiricalliterature.And yetin the end thatliteratureprovedrathersterile.Part of theproblemwas purelyaesthetic:amodelthatsimplyassumesthatthereis acentral businessdistrictisdeeplyunsatisfYingif youwanttoexplaincitiesas opposedto describingthem.But urbaneconomistsmight have swallowedtheirdisappointmentif it hadnotbeenforanotherprob- CENTRAL PLACE THEORY Economicsaswe knowit is largely,thoughnotentirely,anAnglo- Saxontradition.Locationtheory,however,waslongaGermantra- dition, containingatleastthreestreams.One streamfollowsfrom lem:cities,donot look like that,andtheylook lesslike thatwith everypassingdecade. Rememberwhat I saidaboutgreaterLos Angeles,with its 16 edge cities far overshadowingits two traditionaldowntowns. Sophisticatedpeopleusedto turnuptheirnosesatCaliforniacities, with theirlackof clearlydefinedcenters- GertrudeSteindeclared of Oaklandthat"there'sno therethere."Shewaswrong,ofcourse: therearelotsof quitedistincttheresthere,justno onebigthereyou cancall thecenter.And increasinglythatis thewayall our cities look. I wrotethedraftof thisbookin PaloAlto, whichispartof the SanFranciscometropolitanarea.SanFranciscoproperisacompact city,whichstillepitomizesacertainkindof urbanity.But PaloAlto haslittletodowith thatcenter.One thinksof oneselfasbeing,not aSanFranciscan,butaresidentof SiliconValley;onereadstheSan JoseMercuryNews ratherthantheSanFranciscoChronicle(it is abet- terpaperanyway),andpeoplewho livein PaloAlto aremuchmore likelyto commuteto theedgecityin Sunnyvalethanto thevicini- tyof theGoldenGate.The monocentriccitymodelpicturesamet- ropolitanareaassomethinglike a slicefrom an onion,with rings arrayedaroundasinglecenter.The realityof all largemetropolitan areasin the United Statestoday,eventhoselike New York or Chicagothatstillhavehuge,vitaldowntownofficedistricts,is that theyarelesslike anonion sliceandmorelikeJack Horner'splum pudding,in whichedgecitiescorrespondto theplums. The problemof modelingthestructureof themodernpolycen- tric urbanareais in largepartoneof eXplainingtheformationand locationof theseplums.That is, we cannotavoidtheproblemof understandingtheurbanarea'sself-organization.Partof theprocess bywhichthatself-organizationtakesplaceis,of course,acompeti- tion for landthatestablishesa land rentsurfaceacrosstheurban landscape;in thatsensethevon Thiinen-Mills approachremains essential.But it isatbesthalfthestory- andarguablythelessinter- estinghalf 13Self-Organizationin SpaceSELF-ORGANIZING ECONOMY12 15Self-Organizationin Space The answer,I believe,is thatanyeconomistwho thinkshard aboutcentralplacetheory realizesthat it doesnot quite hang togetherasaneconomicmodel.What do we look for in aneco- nomicmodel- or toputit differently,whatconstitutesan"expla- nation"fromthepointof view of economists?I do not thinkyoucansumit up betterthanThomasSchellingdid in thetitleof his book, Micromotivesand Macrobehavior.We feel thatwe havereally managedto shedlighton aphenomenonwhenwe showhow that phenomenon,the"macrobehavior,"emerges(thereis oneof those wordsagain)fromtheinteractionofdecisionsbyindividualfamilies or firms;themostsatisfyingmodelsarethosein which theemer- gentbehavioris mostsurprisinggiventhe "micromotives"of the players.What is thereforedeeplydisappointingaboutcentralplace theoryis thatit givesno accountalongtheselines.Loschshowed thata hexagonallatticeis efficient;he did not showthatit would tendto emergeout of anydecentralizedprocess.Christallersug- gestedtheplausibilityofahierarchicalstructure;hegavenoaccount of how individualactionswouldproducesuchahierarchy(oreven sustainoneonceit hadbeensomehowcreated). What,then,is centralplacetheory?It is notacausalmodel.It is probablybestto thinkof it asaclassificationscheme,awayof orga- nizingourperceptionsandour data.Seenin thatlight,it is awor- thyenterpriseindeed.After all, in thephysicalandbiologicalsci- encesclassificationschemeshaverepeatedlyservedasthebasisfor greatinsights- thinkof theLinnaeanclassificationof speciesor the periodictable.The point,however,isthatclassificationschemesare onlyastepon theway:theytellyouwhat,but theydonottellyou why.So centralplacetheoryis adescriptionbut not reallyanexpla- nationof self-organization. SCHELLING'S SEGREGATION MODEL The twoapproachesto theorganizationof spacejust describeddo not quite makeit: the von Thiinen-Mills approach,which is a modelof spatialorganizationbutnotof self-organization,andcen- tralplacetheory,which is a usefulclassificationschemebut not a causalmodel.Well, enoughof frustration:nowletusturnto some approachesthat do help explainthe self-organizationof spatial econormes. Thiinen analysisofland rentandlanduse,whichwe have lussed.A secondstream,associatedwith AlfredWeberand ers,focusedontheissueofoptimalplantlocation;thatlit- "11playno rolein mydiscussion.But thereisathirdtradi- ichatfirstsightseemstoofferananswerto theproblemof If-organization:thecentralplacetheoryof Christallerand SELF-ORGANIZING ECONOMY asicideasof centralplacetheoryseempowerfullyintuitive. a featurelessplain,inhabitedby anevenlyspreadpopula- "rmers.Imaginealsothattherearesomeactivitiesthatserve ersbutcannotbeevenlyspreadbecausetheyaresubjectto o[lliesof scale- manufacturing,administration,and so on. seemsobviousthatthetradeoffbetweenscaleeconomies r~[lllsportationcostswill leadto theemergenceof a latticeof places,"eachservingthesurroundingfarmers. bvious,but still intuitivelypersuasiveoncepresented,are en~ementsintroducedby ChristallerandLosch.Christaller",I~ndproducedevidencein support,thatcentralplacesform hy: therearealargenumberof markettowns,everygroup t townsis focusedon alargeradministrativecenter(which arkettown),andsoon. Loschpointedoutthatif alattice o minimizetransportationcostsforagivendensityofcen- s, the marketareasmustbe hexagonal.And thusevery on locationtheorycontainsapictureof anidealizedcen- systemin whichahierarchyofcentralplacesoccupiesaset hexagons. iginalcentralplacetheorystoryappliedto townsservinga ketoBut it isobviousthata similarstorycanbeappliedto districtswithin a metropolitanarea.Smallneighborhood districtsarescatteredacrossthebasinsthatsurroundlarg- tswith morespecializedstores,alleventuallycenteringon townwith itsgreatdepartmentstoresandhigh-endbou- Illdeed,thehierarchicalimageis sonaturalthatit is hardto .cribingthingsthatway. \MHyis centralplacetheorynot a standardpartof theecono- lkit?Why do theintroductorytextsfind no room for a In thatusescentralplacetheoryto explainwhy we havecenters,thenusesthevonThiinen-Mills modeltoexplain n ofland usearoundthosecenters? 14 1. You shouldnot,bytheway,concludethatI prefer"commonsense"toaca- demicdiscourse.In the realworld of affairs,an economicideais mostlikelyto succeedif it isnaivelywrong- onlythencanit appealto theprejudicesof impor- tantpeople. I havealreadyexpressedmyadmirationfor thework of Thomas Schelling.He is bestknown for his famoustreatiseon non-zero- sumgames,TheStrategyof Conflict.But I thinkthathisbestbookis MicromotivesandMacrobehavior,anunderappreciatedclassicthathada deepimpacton mewhen I first readit asa wet-behind-the-ears assistantprofessor.The firstchapterof thebook is surelythebest essayon what economicanalysisis about,on the natureof eco- nomicreasoning,thathaseverbeenwritten.And thetwochapters on "sortingandmixing"areawonderfulintroductiontotheideaof self-organizationin economics. If thereis anyflaw in Schelling'swork, it is thathe is socleara thinkerthathecanoftenreachdeepconclusionswith almostnovis- ible technicalapparatusandsogracefulawriter thathe canoften maketheseconclusionsseemintuitivelyobvious.Thesevirtues,I believe,haveworkedagainsthim. As an amateuranthropologist who haslongstudiedthatpeculiarcultureknownasacademiceco- nomics,it seemstomethataneconomicideaflourishesbestif it is expressedin arathertechnicalway,evenif thetechnicaldifficultyis largelyspurious.!After all,a teacherwantssomethingto do atthe blackboard,and a cleverstudentwantssomethingon which to demonstratehisor hercleverness.If a deepideais conveyedwith simpleexamplesandelegantparables,ratherthanwith hardmath, it tendsto getignored. In anycase,however,in MicromotivesandMacrobehaviorSchelling presenteda simpleyetprofoundmodelof segregation.The basic ideasoundstrivial:segregationresultswhenpeopleprefernot to havetoo manyneighborswho aredifferentfrom themselves.But Schellingmadetwo muchlessobviouspoints.First,mild prefer- encesaboutthecolor or cultureof your neighbors- preferences thatseemon thefaceof themtobeconsistentwith maintainingan integratedresidentialpattern- in facttypicallyleadtoahighdegree of segregation.Why?Because,evenwhenpeoplehavemildprefer- encesof theform"I don'tmindhavingsomeneighborsof adiffer- entcolor,aslongasI'm not toomuchin theminority,"integrated 2. An equivalentstatementof therule is thateachindividualrequiresthatat least37percentof theneighborsbeof hisor herown type. 17Self-Organizationin Space residentialpatternstendtobeunstablein thefaceof randompertur- bations.Second,evenif theconcernsof individualsareverylocal- theycareaboutonlytheirimmediateneighbors- whatemergeare largesegregatedneighborhoods.Thus Schellingderived,without anyfanfare,a themeof manywriterson complexity:local,short- rangeinteractionscancreatelarge-scalestructure. Characteristically,hemadethesepointswith afewsimpleexam- plesratherthanafullyworked-outmathematicalmodel.It ispossi- ble to setup suchamodel,andI amsurethatonecanevenprove theoremsaboutit. I shallreservemylimitedmathematicalfirepow- er, however,for my own models. For our current purposes, Schelling'sapproachwill begoodenough. Let us,then,imaginea "city" thatconsistsof anumberofloca- tionslaidout in a squarelattice,like achessboard.For illustration, indeed, let us actuallyuse an 8-by-8 chessboard-sizelattice, althoughthesameprincipleswouldapplytoamuchlargerdomain. And letus supposethattherearetwo kindsof people- callthem blackandwhite,buttheycouldrepresentanykind of racialor cul- turalgroups,orforthatmatterdifferenttypesofbusinessesthattend to repeleachothers'customers,like boutiquesand auto supply stores. We now assumethatblacksandwhitescareaboutthecolor of theirimmediateneighbors.(Thatmeanstheabuttingsquaresonthe chessboard.)Their preferencesareassumedtobenotsomuchalik- ing for neighborsthe samecolor as a fear of being isolated. Specifically,Schellingsuggestedthe followingrule: an individual with oneneighborwill try to moveif thatneighboris a different color;onewith twoneighborswantsatleastoneof themtobethe samecolor;onewith threetofiveneighborswantsatleasttwotobe hisor hercolor;andonewith six to eightneighborswantsatleast threeof themtobelikehim or her.2 Thesepreferencesareconsistentwith an integratedresidentialpattern.ConsiderFigure1.2,in which#and@ signifYthetwodif- ferentgroups.Herewe havemanagedto place60 individualsin a completelyintegratedpattern,without violatinganyone'scon- straints.That is,completeintegrationisanequilibrium. SELF-ORGANIZING ECONOMY16 But arewelikelyto geta resultlike thisin practice?No, argued Schelling.Althoughhedidnotputit quitethisway,anequilibrium like theone shownin Figure1.2will be unstablewith respectto somerandomshufflingandwill thereforetendto unravel. To showthis,Schellingtookthepatternin Figure1.2andmessed it up a bit. (His instinctwasunerring.My first inclinationwould havebeento startfromacompletelyrandomallocationof peopleto locations.It turnsout, however,thatself-organizingspatialsystems yieldthegreatestorderwhentheinitialconditionis a smallpertur- bationawayfromtheunstableintegratedorflatequilibrium.Wewill seewhyin thesecondpart.)Specifically,heextracted20individuals atrandom,bothdisruptingthepatternandfreeingupsomeroomfor discontentedindividualsto move,thendisruptedthepatternabit moreby filling fiveemptysquaresatrandomwith #sor @s.Figure 1.3showstheresultwhenI andmyrandomnumbergeneratordoit. It isprettyobviousthatin Figure1.3someof thepeopleareno longercontentwith their locationsand will move.When they move,however,theywill in turnoftenmakethosewho stayunhap- py- eitherby deprivingsame-colorindividualsof neighborsor by shiftingthebalanceagainstnewneighborsof adifferentcolor.Soa chainreactionbegins.To simulatethatchainreactiononacomput- eryouwouldneedto specifytheorderin whichpeoplemoveand how theypick amongavailablelocations.If you aredoing it by hand,you canbe looseraboutit - it doesnot mattermuch.Like Schelling,I did thisbyhand(usingaspreadsheetinsteadof pennies and dimeson a realchessboard)andjust watchedthe structure evolve.When thingssettledown,yougetFigure1.4. Guesswhat: eventhough individualsare tolerantenoughto acceptan integratedpattern,theyendup with moreor lesstotal segregation.And, eventhoughindividualscareaboutonly their PerturbingtheEquilibrium.If thepatternis givensome randomscrambling,someindividualsareno longercontent withtheirlocation. 19Self-OrganizationinSpace #@ @#@ # @#@#@# @ #@#@#@# # ##@# @ @@@@@# # @ @#@#@ @ # # @ @ @ Figure1.3 #@#@#@ # @#@#@#@ @ #@#@#@# # @#@#@#@ @ #@#@#@# # @#@#@#@ @ #@#@#@# @ #@#@# An IntegratedCity. Evenif peopleinsistthataminimum fractionoftheirneighborsresemblethemselves,it ispossibleto createanequilibriumresidentialpatternthatishighlyintegrat- SELF-ORGANIZING ECONOMY Figure1.2 18 21Self-Organizationin Space Asyoucantell,I loveSchelling'smodel.It isaperfectexampleof how to wearyour sophisticationlightly;thewaythatbig conclu- sionsarederivedfromplayingwith coinsonachessboardhasaspe- cialcharm.As amodelof theself-organizationofurbanareas,how- ever,it hassomelimitations.Someof themareproblemsthatI will nottrytofix in thisbook;forexample,thereisnoexplicithandling of themarketfor land,which surelyplaysacrucialrolein mediat- ingthechoicesofhouseholdsandfirmsaboutwheretolocate.One problemthatI do want to dealwith, however,is thatSchelling's model is too one-sided:it tellsus why birds of a featherflock togetherbut offersno reasonwhy thereshouldeverbemorethan twoflocks. To seewhat I mean,askwhatwould happenif I replacedmy 8-by-8chessboardwith amuchbiggerone,say1000by1000.What wouldaSchelling-typemodelpredictasanoutcome?Well, ahuge literaturein physicsdealswith systemsnot too differentfrom Schelling'smodel,calledspin-glasses,andanumberof formalresults havebeenderived;but I havetoadmitthatI havenotyettakenthe timeto try to masterthisliterature.Still, a few thingsseemfairly obvious.The tendencyof Schelling'smodelis alwaysto dividethe whole city into two vast#and@territories.That doesnot mean thatwewill necessarilygetthere:if we startwith arandompattern, the chainreactionof movinghouseholdswill typicallydie out at somepoint,leavingthecity"frozen"into#and@ domainsofvary- ing sizes.(If we settheparametersright, thesizedistributionwill surelyobeyapowerlaw- butmoreaboutthatkind of thinglater.) We can,however,"melt"thesedomainsif weaddabit of "temper- ature"tothestory,bygivingevencontentedhouseholdssomesmall probabilityof movingjust for the hell of it. In thatcasewe will indeedeventuallyfind ourselveswith atwo-neighborhoodcity. Now thatis too stronga result.Let usgo backto Los Angeles. Therearesomewaysin whichthecityhasatwo-zonestructure:all thebeautifulpeoplelivein BeverlyHills, all thenonbeautifulpeo- plesomeplaceelse.But whatissostrikingaboutLA comparedwith atraditionalcityispreciselyitsmultipolarity.Thereisno dominant downtownofficedistrict:instead,thereare16edgecities,spreadin a not-too-irregularfashionacrossthe metropolitanarea- and, accordingto Garreau,8 morein theprocessof coalescing. # # #### # ## ### # ###@@ # # ##@@@@ @ @@@@@@ @ @@@@@@ @ @ @ @ @ SELF-ORGANIZING ECONOMY 1.4 A SegregatedCity.Theresultisachainreaction,in which eachmoveprovokesothermoves;intheendmildconcernsabout beinginalocalminorityproduceacompletelysegregatedcity. 20 calneighborhood,thewholechessboardgetsorganizedinto eareaandablackarea. and@ reallydorepresentwhiteandblack,it isanastyoutcome. That, owever,isbesidethepointforthisbook(exceptasareminder that . anizeddoesnot mean"good").What I wantto emphasizeis thatt ischessboardcityhasengagedin aprocessof self-organization. Larg scaleorder- not a nice order,but ordernonetheless- has emeredfromabasicallydisorderedinitialcondition. T .s large-scaleorderemergesbecausea disorderedstate,in whicl #'sand@'sareevenlymixed,is unstable:scrambleit abitand yous .rt adynamicprocessthatproducessegregation.This isthere- fore y firstillustrationof theprincipleof orderfrominstability- butd notworry,therewill bealot more. r·· II DGE CITY DYNAMICS 23Self-Organizationin Space 3. Alternatively,one canimaginea city of infinite extent.Los Angelesmay approximatethiscondition. theends.Thatisnotsimplyaformalissue:in realityanofficecom- plexnearthemiddleof a metropolitanareawill be differentfrom onenearthepointwheresubdivisionsbegintobemixedwithfarm- land.BecauseI do notwantto dealwith thatissue,I needto make asillyassumption:mycityisnotonlyonedimensionalbutcircular, sothereareno ends.3 A secondissueis thetreatmentof landandlandrents.Modern urbaneconomistsaregenerallywilling to playgameswith geome- try,but theytendto be quitefanaticalaboutexplicitlymodeling landrent- when I presenteda modelwithout landrentsto one group,adisgruntledurbantheoristtoldmethatasfarashewascon- cerned,urbaneconomicswasessentiallyaboutlandrent.It is obvi- ous where thatattitudecomesfrom: in the von Thiinen-Mills modellandrentis indeedthecrucialingredient.But I amaskinga differentquestionandwill askyou to bearwith anapproachthat recognizeslandscarcityatbestin animplicit,reduced-formway. A thirdissueis thetreatmentof expectations.I will be tellinga dynamicstory;it isastoryabouthowedgecitiesevolve.But I donot wantto worry aboutforward-lookingbehavior.To manyecono- mists,raisedin anenvironmentof rationalexpectationstheory,that byitselfdisqualifiesamodel.Most ofyouprobablydonotcare,but for thoseofyouwho do, I urgeyoutoputyourprejudicesonhold. Finally,I haveapersonalrulethatI amabouttobreak.In gener- al, I do not like simply assumingthe existenceof external economiesanddiseconomies.It oftenseemstoocloseto assuming your conclusions.(I know of oneeconomistwho triedto explain hisworktoagroupofphysicists,oneofwhomsarcasticallysaid,"So whatyou'retellingusis thatfirmsagglomeratebecauseof agglom- erationeconomies.")Most ofmyworkoneconomicgeographyhas focusedon tryingto deriveexternaleconomiesout of theinterac- tions among scaleeconomies,transportationcosts,and factor mobility- to makeexternaleconomiesan emergentproperty.I actuallyarrivedattheapproachI amgettingto in thecontextof suchamodel,onethatI shallpresentin thesecondpart.But it turns out thatit is easiestto explainthebasicideasimplyby assuming externaleconomies,andsoI will temporarilysuspendmyrule. SELF-ORGANIZING ECONOMY So if we wanta storyaboutthe self-organizationof theurban l~ndscapethatreflectsthepolycentric,plum-puddingmetropolitan reaswe increasinglyinhabit,we needamodelthatspontaneously roducesnot only orderbut somekind of more or lessregular epetitivepattern.Let usseeif wecanconstructone. IinamomentI will describeanapproachthat,it seemstome,sheds onsiderablelightonhowofficedistricts- edgecities- mayemerge ih a polycentricmetropolitanarea.I believeand hope that the pproachwill provideausefulmetaphorfor theemergenceof spa- tlal structurein a varietyof othercontexts.But anyexperienced conomistknowsthatin presentinganewapproachonemustbegin with somepreemptiveexcusesfor thequestionablesimplifications nehasmade.SobeforeI evenstartto describemyframework,let eoffersomejustificationsfor thewayit is done. The modelsthatI will describelooselynow andpresentmore lly in thenextpartwill, not surprisingly,representakind of ide- allizationof reality.This will not shockanyone:bothvon Thiinenandcentralplacetheorybeginwith theidealizationof ahomoge- eousagriculturalplain,which neverboremuchresemblanceto alityandbearsevenlessnow thanit did whentheywrote.What ind of idealizationseemslegitimateis,however,in theeyeof the eholder.My guessis thatthereareatleastfourwaysin whichthe a~proachI amabouttopresentwill botherpeople. First, thereis the issueof geometry.The Germantraditionin locationtheorywascheerfulaboutassumingawayrivers,roads,and ariationsin landqualitybutrigorousaboutfacingupto theconse- uencesof thefactthattheEarth'ssurfaceis twodimensional.It is cfmmon in modernurbaneconomictheory,however,to analyze "long, narrow"citiesthatareeffectivelyone dimensional(as,for ebmple, in Fujita1988),a simplificationthatseemsreasonableto ebutthatwouldprobablyhavehorrifiedmanyof theGermans. Well, it turnsout thatformypurposeseventheone-dimension- cityis notquitesimpleenough,becauseI wantto focuson self- 0rganizat.•ion:theemergenceof structurethatarisesnotfrominher- e t differencesamonglocationsbut from theinternallogicof the s stem.And evenin a one-dimensionalcity locationsaredistin- g ishedfromoneanotherbyonecrucialaspect:theirdistancefrom 4. Joel Garreauwrites:"Fivemillionsquarefeetis a pointof spontaneous combustion.It turnsouttobeexactlyenoughtosupportthebuildingofaluxu- ryhotel.It causessecondaryexplosions;businessesbegintoflocktothelocation toservethebusinessesalreadythere." Now thattheexcusesareout of theway,letusgetto thesub- stance. Imagineametropolitanarea,whichwe canthink of asahomo- geneousexpanseof identicalhousingdevelopments- exceptthat, ratheroddly,thepopulationisdistributednotin atwo-dimension- alsprawlbutin anarrowring,andtravelispossibleonlyalongthat ring'scircumference.And supposethatthereissomebusinessactiv- ity- it couldbeofficework,it couldberetailing- thatdependsOn thespread-outpopulationbothasamarketandasasourceoflabor. Let usalso,unobjectionably,supposethatthedecisionsbybusi- nessesaboutwhereto locateareinterdependent.That is, thedesir- abilityof anyonesiteasabusinesslocationdependsOnwhereallof theotherbusinessesarelocated.And we mayalsosafelysuppose thatbusinessesmigrateovertimefromlessto moredesirablesites. Clearlythedynamicsof thisprocessdependOnthenatureof this interdependence.One canimaginetwogeneralsortsof interdepen- dence.On the Onehand,businessesmight dislikehavingother businessesnearby,becausetheycompeteforcustomers,workers,or land.Call theseconsiderationscentrifugalforces,forcesthatpromote dispersionof business.On theotherhand,businessesmightlike to haveotherbusinessesclose,becausetheyattractcustomersto the areaor helpsupporta greatervarietyof localservices.4Call these centripetalforces,forcesthattendto makebusinessesclumptogeth- er.If onlycentrifugalforcesexisted,businesseswouldspreadthem- selvesevenlyacrossthe landscape.If therewereonly centripetal forces,theywouldrushtogetherinto Onebigclump. But whatexplainsthepolycentric,plum-puddingpatternof the modernmetropolis?In general,amodelthatwouldexplainthispat- ternmustmeettwo criteria: Criterion 1.Theremustbe a tensionbetweencentripetaland centrifugalforces,with neithertoostrong. Criterion 2. The rangeof thecentripetalforcesmustbeshorter thanthatof thecentrifugalforces:businessesmustliketohave otherbusinessesnearby,but dislikehavingthema littleway away.(A specialtystorelikesit whenotherstoresmoveinto its 25Self-Organizationin Space "oJ" "l/l'17e -"" - c) CIl l.- llS "l.J:: " II> c:::. II> II> CIl c: '" "Vi c)::l cO 't c:i I/) _ c:i r___________. -"-~, shoppingmall,becausetheypull in morepotentialcustomers; it doesnot like it whenstoresmoveinto a rivalmall10miles away.) And that'sall thatwe need.In anymodelmeetingthesecriteria, anyinitialdistributionof businessacrossthelandscape,nOmatter howeven(orrandom),will spontaneouslyorganizeitselfintoapat- ternwith multiple,clearlyseparatedbusinesscenters. Doesthispropositionsoundobvious?Maybeso (althoughI did nothaveit clearin myOWnminduntil afterI startedplayingwith mathematicalmodels).But thereismore.For awidevarietyof spe- cific modelsin which criteria1and2 aremet,anyinitialdistribu- tion of businessacrossthelandscapewill evolvenot merelyinto a patternwith severalbusinesscentersbut into a patternin which thesecentersareroughlyevenlyspaced,with a characteristicdis- tancethatdependsOnthedetailsandparametersof themodelbut notOntheinitialdistribution.And thesmootheris theinitialspa- tial distributionof the businesses,the more eventheir eventual spacmg. Figure1.5 The Evolutionof EdgeCities.An ititiallyalmostuniform distributionof businessaccrossthelandscapeevolvessponta- neouslyintoa higWystructuredmetropoliswith twoconcen- tratedbusinessdistricts. SELF-ORGANIZING ECONOMY24 5. Eachlocationreceivedaweight5 +Ui, whereUi wasalocation-specificran- domvariablebetween0 and1, ands wasa"smoothing"parameter,setequalto 5 forthisrun.Then eachlocationwasassignedashareofthebusinessesproportional to itsweight, 6. Robert Gordonof NorthwesternUniversityhasdubbedFigure1.5the'59 Cadillacmodel. At thispoint I hadbettershowyou anillustration.It turnsout, not too surprisingly,thatcomputersimulation- and computer graphics- arean invaluableaid to thinkingaboutself-organizing systems.Figure1.5isasamplerun of amodelof a "city" consisting of 24 locationsaroundacircle.The locationsareshownon theX axis;bearin mindthatlocation24isnexttolocation1.I startedthe runwith afairlybutnotperfectlyevenallocationof businessacross theselocations,5thenlet it evolveaccordingto a rule thatcaused businessesto movetowardlocationsthatwere highly desirable. "Desirability"of a locationwas both positivelyand negatively affectedby thenumberof businessesatotherlocations,with both effectsdecliningwith distancebutwith thepositiveeffectsdeclin- ing fasterthanthenegativeeffects.The Y axisof thefigureshows thepassageof time;theZ axisshowstheshareof thebusinessesin eachlocationateachpointin time.Thus thefiguregivesyouasort of frozenportraitof thewholesimulatedhistory. Let uslook abit atFigure1.5;thepicturecontainsquitea lot of information.The rightedgeof thecalculatedsurface,whichrepre- sentsthe initial geographicaldistributionof business,is almosta horizontalline.That is, I havestartedmy"city" offwith almostno spatialorganization.But eventuallythesurfacerearsup into apair of dorsalfins:6 all of thebusinessesendup in locations8 and20. The cityhasspontaneouslydevelopedastrongspatialstructure. This maynotbesurprising.But now noticewhich locationsget thebusinesses.In a24-locationcircularcity,locations8 and20are exactlyoppositeoneanother.That is notanartifactof theparticu- lar startingposition:ifyou run the modelrepeatedlywith these parameters,butwith a differentinitial spatialdistributionof busi- nesseachtime,you will consistentlygettwo businessconcentra- tionsoppositeeachother. By theway,I do notwantyouto gettheimpressionthatthereis somethingspecialaboutcitieswith twobusinessdistricts.Figure1.6 showsa typicalrun of the samemodelwith somewhatdifferent 27Self-Organization in Space ...•o ti Q)O \.. '" ~tiIII Ill'"III - ~ti 'iij ::::l <:> d) -ti II) '" ci o 10) ci parameters.In thiscasewegetfourbusinessdistricts,equallyspaced aroundthecircle.The generalprinciple,then,is thatthecityorga- nizesitselfinto a structurewith a characteristicdistancebetween businessdistricts;if thereareonly two, thisimpliesthattheyface eachother. Letusgobackto Figure1.5andlook atonemorething.Look at how thealmostflat initial surfaceevolvesovertime.Althoughin theendonly twowidelyseparatedlocationsendup with business concentrations,in theearlystagesof self-organizationit doesnot look asif thewinninglocationsaregrowingattheexpenseof their neighbors.On thecontrary,initiallynotonlythewinninglocations butthosenearbygrow.The surfaceseemsto undulate,with waves risingupoutof theplain.Only aftersometimehaspasseddothese wavesgatherthemselvesinto dorsalfins,with theeventualcenters cannibalizingtheirneighbors.Again,thisimpressionis notunique toaparticularrun:it isaconsistentfeature,whatevertheinitialdis- tribution. Figure1.6 SameStory,DifferentParameters.With somewhatdiffer- entparameters,themetropolisevolvesfour businessdistricts. SELF-ORGANIZING ECONOMY26 7. Rememberthatthesearespatialfrequencies:thesearewobblesin space,not time. It turnsout thatthis imageof an undulatingsurface,with the wavesgrowingovertime,isthekeytounderstandingtheprocessof self-organization.To fullyexplainwhywill takealittletime,andI'll reservethatforthesecondpart(andthetechnicalappendix).But let meofferapreliminaryview. RememberthatI startedthe simulationin Figure1.5with an almostbutnotexactlyflatdistributionofbusinessesaroundthecir- cle.The deviationof thatdistributionfromperfectflatnessmaybe representedasanirregular,wobblyline.But it isgenerallytruethat, to usetechnicallanguage,anirregularwobblecanbethoughtof as thesumofmanyregularwobblesatdifferentfrequencies.In partic- ular,anirregularwobblearoundacirclecanbeexpressedasthesum of oneline thatwobblesonceasit goesaroundthecircle,another thatwobblestwice,a third thatwobblesthreetimes,andso on. (Someofyouknow thatI amtalkingaboutaFourierseries.) Now hereis thepoint:in anymodelthatsatisfiesmytwocrite- ria, someof thesecomponentwobbles- wobblesthatgo certain particularnumbersof timesaroundthecircle- will tendto grow overtime(althoughwobblesatotherfrequencies7maytendto die out).And, becausethedecompositionof anirregularwobblewill ordinarilycontainwobblesof allpossiblefrequencies,anevendis- tributionofbusinessaroundthecircleisunstable.Someof thecom- ponentwobbleswill grow,creatinganincreasinglyunevenspatial distributionof business.Orderfrominstability! Why, however,is theeventualspacingof businessso regular? Becausewobblesthatgodifferentnumbersof timesaroundthecir- clewill growatdifferentrates.If theinitialdistributionof business is sufficientlysmooth,afterawhile thedeviationfromsmoothness will bedominatedbywhicheverfrequencywobblegrowsthefastest - bythemostunstablewobble.And thefrequencyof thiswobble,the numberof timesit goesaroundthecircle,will determinewherethe peakbusinessconcentrationsarelocated.The undulationsyou see in Figures1.5and1.6arethatmostunstablewobbletakingoverthe distributionof businesses.For the parametersusedto generate Figure1.5,themostunstablewobblegoesaroundthecircletwice; for thoseusedto generateFigure1.6,themostunstablewobble goesaroundthecirclefour times.Not only doesinstabilitycreate order,theform of thatorderis dictatedby a sortof principleof maximuminstability. Whataboutthelastpartof thefigures,wherethewobblesgath- erthemselvesup intospikes?Well, theformalansweris thatevery- thingthatI havesaidisvalidonlyforalinearapproximationto the model,whichbreaksdownwhenthespatialdistributionofbusiness getstoouneven.Lessformally,in theearlystagesof self-organiza- tionthemostfavoredlocationscangrowbypullingbusinessesaway from distantlocations.Once thereareno morebusinessesin the largegapsbetweenbusinesscenters,theycancontinuetogrowonly byeatingtheirneighbors.The importantpoint,however,isthatthe locationsof thewinnersaredeterminedin theearlierstage:thedis- tancebetweenedgecitiesis determinedby thewavelengthof the mostunstablewobble. OK, letusstopandtakeadeepbreath.EventhoughI haveavoid- ed any formal modeling,you may at this point havesuddenly realizedhow seeminglyabstract,how unrelatedto the detailsof freewaysandshoppingmalls,skyscraperconstructionandfastfood consumption,thishasgotten.That isverymuchthestyleof com- plexitytheorists:indeed,thewholerationaleof thefieldis theidea thatcommonprinciplesmayapplyto subjectswith verydifferent details.Still, hastheabstractionled us into a storythatconflicts badlywith reality? My guessis thatmanyreaderswill objectto theimpliedregular- ity of theresult- thoseequal-size,regularlyspacedbusinesscon- centrations.In realityedgecitiesarenotallthesamesizeor equally spacedacrossthe landscape.That doesnot worry me, however. Mter all, thereallandscapeis not homogeneous.Therearehigh- ways,whoseintersectionsmakeparticularlyfavorablesitesforbusi- ness,variationsin thepleasantnessor buildabilityof sites,andfor thatmatterrealmetropolitanareasdo havecentersandedgesand thusareprimafacienot undifferentiated.Furthermore,theresult thatbusinessdistrictsareevenlyspacedistrueonlyforanimaginary historyinwhichbusinessstartsoutspreadalmostevenlyacrossloca- tions;becausetherealhistoriesarenotlikethat,we shouldexpecta moreirregularresult. But if realcitiesdo not evolvein thewayI havejust described, doesthatmeanthatthewholeapproachis irrelevant?I think not, but I needto introducesomemore conceptsand examplesto explainwhy. 29Self-Organizationin SpaceSELF-ORGANIZING ECONOMY28 -----2----- ComplexLandscapes In hiswonderfulbookNature'sMetropolis:ChicagoandtheGreatWest, thehistorianWilliamCronondistinguishesbetweentwolandscapes in whichurbanevolutiontakesplace.One is thenaturallandscape: themountainranges,rivers,andlakesthataregivensof theenvi- ronment.The secondis the createdlandscapeof railroadlines, canals,farmingpatterns,and citiesthemselvesthat resultsfrom humandecisions.Crononarguesthatin themodernworldthecre- atedlandscape,which he refersto assecondnature,hasbecomefar moreimportantasadeterminantoflocationthanthe"firstnature" in which it is embedded:Chicago'sroleasa GreatLakesportwas quicklyovershadowedbyitsroleasarailhub.And secondnatureis oftenself-reinforcing:railroadsaimedatChicagobecauseit wasthe economiccenterof itsregionandtherebymadeitscentralityallthe greater.That is, the landscapeof secondnature is inherently dynamic. But how doesonevisualizea landscapethatchangesovertime? One answeris ahistoricalatlas:a sequenceof mapsthattracesout thechanges.In somecasesit maybepossibleto do abit betterby drawingakindof temporalreliefmapinwhichthetimedimension, andthustheevolutionof thelandscape,ismoreor lessimperfectly represented.Indeed,Figures1.5and 1.6arejust that:theyshow how someimaginarylandscapeschangeoverthe courseof their imaginaryhistories.But evenwhenyoucandothis,allthatyouget isadescriptionofwhathappened;whyit happenedisatbestimplic- it. Sowhatdoyoudo?Anyonewho hasworkedon formalmodels of dynamicsystemsknowstheanswer:youtryto drawapictureof 11 1. Strictlyspeaking,onecanrepresentatwo-dimensionaldynamicsystemasa three-dimensionalreliefmaponlyif thesystemactsasif it werefollowingthegra- dientof somepotentialfunction.This is sometimesreasonable,but not always: theevolutionof asinglespeciesisin effectmaximizingsomethingwecancallfit- ness,but thecoevolutionofpredatorsandpreyis not.Nonetheless,thelanguage commonlyusedtodescribephasespace,with itsbasinsof attraction,saddlepaths, andsoon, drawsheavilyon thereliefmapmetaphor;andI atleastviewthepoet- ry asworth thepotentialconfusion. 2. See,for example,theinfluentialpaperby Hopfield (1982)andcompareit with the"NK model"ofKauffinan (1993). yetathirdlandscape,a "phasespace"in whicheachpointsumma- rizesthepositionof thesystematapointin timeandin whichthe rulesthatgovernthesystem'sevolutionaretranslatedinto "lawsof motion"in thatabstractlandscape. Phasespacerepresentationsof dynamicsystemsareextremely commonin moderneconomicanalysis.In general,however,we tendtofocuson onlyanarrowrangeofpossibletypesoflandscapes - indeed,basicallyononlytwofairlysimpleforms.Beforethemid- 1970snearlyalldynamicmodelsin economicsweregloballystable; thatis, thephaselandscapewasassumedto belikeabowl,a single basinof attractionin which all pointsdrainto a singlelong-run equilibrium.Sinceabout1975it hasbecomecommonalsotowork with modelsin whichthephaselandscapelookslikeasaddle- and in whichsomesetof forward-lookingvariables,suchasassetprices, is determinedbytheassumptionthattheeconomyisalwayson the ridgethatis theonlypathto long-runequilibrium.1 The literatureon complexity,however,is largelyconcernedwith systemsin which thedynamiclandscapelookslike neitherabowl norasaddle;indeed,it isoftenconcernedwith "ruggedlandscapes" (asStuartKauffmanputsit) thatlook like theSouthDakotabad- lands. What aspectsof a dynamicsystemleadto a ruggedphaseland- scape?I haveno generalanswer,andI do notknow if thereis one. Many of themodelsin thecomplexityliterature,however,havea similarsetup.2The modelerrepresentshis or her systemas an ensembleof manycomponents,eachof which is atanyparticular timein oneof severalstates:magneticdipolesthatareorientedup or down,neuronsthatarefiring or quiescent,genesthathaveone characteror theother,andso forth.And futurechangesin these statesarelinkedin someway:theenergyofaspin-glass,andthusits likelydirectionof change,dependson whethernearbydipolesare orientedin thesamedirectionornot;neuronsexciteor inhibiteach other;theeffectof changingan individualgeneon anorganism's fitness,and thus the likelihood that a mutationwill survive, dependsonwhatothergenesit has.Suchasystem,it turnsout,will produceacomplexdynamiclandscapeaslongastwoconditionsare met:theresponsesof theindividualunitsmustbediscrete(neurons eitherfiring or not),andtheremustbe a mixtureof positiveand negativefeedback(neuronsbothexciteandinhibiteachother). Now think aboutspatialeconomies.They aresystemsin which manycomponents(firms)areat anyparticulartime in particular states(locations)and in which changesin thesestatesarelinked (throughagglomerationeconomiesanddiseconomies).If thereare significanteconomiesof scale,firmswill chooseonlyafewdiscrete locations;andtherewill usuallybeamixtureof positiveandnega- tive feedbackbetweenthesechoices.Thereforewe might well expectspatialeconomiesto havecomplex,ruggeddynamicland- scapes. And indeedthatis whatyou find whenyou look atevenquite simplemodelsof spatialeconomies.Over thelastfewyears,I have spenta lot of time tryingto understandthebehaviorof a model almostasminimalistasthe edgecity model I introduceda little while ago.(In fact,asI shallexplainin Part II, the two models behavein verysimilarways.)In thisstructureweimaginethatthere aretwo factorsof production:immobileagriculturalworkersand mobilemanufacturingworkers.Manufacturingisamonopolistical- ly competitivesectorcharacterizedbybothincreasingreturnsatthe levelof thefirm andtransportcosts.The interactionamongfactor mobility,increasingreturns,andtransportcostsgeneratesforcesfor agglomeration:firms tendto concentrateproductionin locations with goodaccesstomarkets,butaccesstomarketsisgoodprecisely whereotherfirmsareconcentrated.Working againstthese"cen- tripetal"tendencies,however,is the"centrifugal"pull providedby thegeographicallydispersedagriculturalsector. If weaddsomerudimentarylawsof motion,saytheassumption thatmanufacturingworkerstendtomovetolocationsthatofferrel- ativelyhigh realwages,we geta dynamicstoryin a phasespace 33Complex LandscapesSELF-ORGANIZING ECONOMY32 3. The kind of complexlandscapethatcanarisein modelsof economicgeog- raphycan,of course,arisein manyothereconomiccontextsaswell.Mostnotably, thechoiceamongseveraltechnologiessubjecttonetworkexternalitieswill present averysimilarpicture,with thelandscapecomplexif noneof thetechnologieshas too stronganinherentadvantageandtheexternalitiesaresufficientlypowerful. definedbytheallocationofmanufacturingworkersacrosslocations. And thisdynamiclandscapecaneasilybeverycomplex.3 An easilyshownexampleis thethree-regioncase.Supposethat therearethreeequidistantlocations,with equalagriculturallabor forces.The allocationof themanufacturingwork forceamongthe threelocationscanberepresentedasapointon the"unit simplex": atrianglewhosecornersareatthepoints(1,0, 0), (0, 1,0),and(0, 0, 1)in a spacewhoseaxesaretheshareof manufacturingin each locationbut thatcanconvenientlybe pastedonto a two-dimen- sionalpage.Startingatanygivenpointon thatsimplex,onecanlet themodelevolveandseewhereit endsup.To drawthepictureana- lyticallyis extremelydifficult,but it is straightforwardto compute numericalexamples. Figure2.1showswhatI getforthemostinterestingrangeofpara- meters.(Thereareonlythreeparametersin themodel:theelastici- ty of substitutionamongproductsin themanufacturingsector,set for thisexampleat4; theshareof manufacturesin expenditure,set at0.2;andthetransportcostbetweenanytwolocations,setatOA.) At eachof anumberof pointson thesimplex,representinganini- tialallocationofmanufacturingworkers,I drawanarrowrepresent- ing thedirectionandspeedof "flow". It turnsout thatthereare fourequilibria:threein whichallmanufacturingis concentratedin onelocation,onein whichthereis anequaldistributionof manu- facturingacrossthelocations.Therearecorrespondinglyfourbasins of attraction:a centralbasinthatleadsto theequaldivisionout- come,andthreeflankingbasinsthatleadto concentration. The landscapecanbecomefar morecomplexwhen thereare morelocations.Suppose,forexample,thatweconsideranexample in whichthereare12locations,laidoutin acirclelikethenumbers on aclock.(Twelveis aparticularlyconvenientnumberbecauseit is a fairly smallnumberwith a largenumberof divisors.)The dynamicsonceagaintakeplaceon aunitsimplex- butthistimean ll-dimensionalone.This is hardfor mostof us to visualize.We can,however,getagoodideaof thepropertiesof themodelexper- Figure 2.1 Basins of Attraction. A three-regioneconomicmodelcanend up with four differentlocationalpatterns,dependingon initial conditions. 35 , "-- \. '" ...•..•...•. '\..- _0 ,, .•....• " "...•. -.•.. - '" o/, / t , / \ I , "'" J \ \ 1 I- " ,/ - ~---+- • ........-""""'" '" - I •••••" ./'" .......•. /" ••.....••.. ,__/t" '" ......•..•...•• Complex Landscapes '" ~ ,/ ~ / - /~ ..-•.......--+-- +- imentally,bystartingwith anumberof randomallocationsof man- ufacturingacrosslocationsandseeinghow theyevolve. Figure2.2illustratesa typicalrun. The initialrandomallocation ofmanufacturingeventuallyorganizesitselfintotwomanufacturing concentrations,atlocations6 and11;thatis,5 apart.In thecourse of anumberof runswith theseparametervalues,I gottwoconcen- trations5 apartabout60percentof thetime,twoconcentrations6 aparton almostall otheroccasions.At rareintervalsa run would leadto threeequallyspacedconcentrations. On a circlewith 12locations,thereare12waysto placetwo markers5 apart,6waystoplacethem6apart,and4waystoplace3 equidistantmarkers.So it appearsthatwith theseparametersthe modelimpliesalandscapewith 22basinsofattraction- ruggedter- rainindeed.In suchaworld thelocationof economicactivity,and to someextenteventhestructureof theresultingeconomicgeog- raphy,woulddependcruciallyon initialconditions,whichisto say on historicalcontingency. Now, finally,we getbackto thequestionthatledusoff on this sidetrip throughphasespace.The beautifullysimple,aesthetically pleasing"'59 Cadillac"picturesin Figures1.5and1.6werebased SELF-ORGANIZING ECONOMY34 THEEMrnRGENCE OF ORDER 37ComplexLandscapes Let us look againatwhat I saidaboutthe 12-regionmodelof economicgeographywhoseoutcomeis representedin Figure2.2. The modelappearsto havea dynamiclandscapewith 22basinsof attraction.And yet,lookedata differentway,the resultsaresur- prisinglyordered.Startingwith arandomallocationofmanufactur- ing acrossspace,the modelalwaysorganizesitselfinto a highly orderedstructurein whichmanufacturingisconcentratedin twoor threeequal-sizeconcentrations.Furthermore,althoughthereare manysuchequilibriumstructures,theysharestrongsimilarities:all involveroughlyequidistantcitylocations,withafairlynarrowrange of typicaldistancesbetweencities. Whatwe sawin theedgecitymodelwasthesecharacteristicfea- turesofequilibriumstructuresin anextremeform:ifwestartedthat modelfromanalmostuniforminitialdistributionofbusiness(cor- respondingto a startingpoint nearthe centerof the simplexin Figure2.1),we gotbusinessconcentrationsexactlyevenlyspaced aroundthe circle,with an invariantdistancebetweenconcentra- tions.The sameoccursin thismodel:if I taketheparametersused to generateFigure 2.2 but restrictmyselfto initial positionsin which thedistributionof manufacturingis sufficientlyflat, I will consistentlygetapicturelike Figure1.5- theeconomyorganizes itselfinto astructurewith twocities,exactlyoppositeoneanother. Whatthissuggeststomeis thatthefakehistory,in whichahigh- lyregularspatialstructureemergesfromanalmostunstructuredini- tialposition,canbeviewedasasortofmodelof themodel.Thatis, thepreciseregularitiesof thatspecialcasehelpusunderstandthe roughregularitiesof themoregeneralcase. Nor do I think thatthis insightis merelyaboutmodeling.It seemsreasonableto speculatethatthe immenselymorecomplex landscapethatdeterminestherealgeographyof theworldeconomy hasitsownunderlyingapproximatesimplicities.That is, theremay be manypossibleoutcomes,dependingon initial conditions- SiliconValleymight,givena slightlydifferentsequenceof events, havebeenin LosAngeles,Massachusetts,or evenOxfordshire.But somebroaderfeaturesmaybemoreor lessindependentof histori- calcontingency. And in factoneregularityin spatialeconomicsissospectacularin itsexactnessanduniversalitythatit ispositivelyspooky.Thatregu- larityinvolvesthesizedistributionof cities,andit leadsus to our otherprinciple:orderfromrandomgrowth. SELF-ORGANIZING ECONOMY 0.0 2 3 4 5 6 7 8 9 10 II 12 Location I_Initial _Final I Figure 2.2. Exploring a Complex Landscape. Models with more regionscanbeexploredviasimulations;hereisatypicalrunfor a12-regionexample. 0.1 0.5 0.4 '0 0.2 Q) L. ..s ..r::. II) 36 on theassumptionthatthespatialeconomystartedwith analmost uniformdistributionof businessacrossspace.But thatis not how realeconomichistoryworks- andwe havejust seenthatspatial economiestypicallymust havecomplexdynamiclandscapesin whichwhereyouendupdependsalotonwhereyoustart.Soisthe simplepicturebasicallyirrelevant? I think not - becausealthoughspatialmodelseasilygenerate complexdynamics,lurkingwithin thatcomplexitywe oftenfind surprisingsimplicity. The mostprovocativeclaimof theprophetsof complexityis that complexsystemsoftenexhibitspontaneouspropertiesof self-orga- nization,in atleasttwo senses:startingfromdisorderedinitialcon- ditionstheytendtomovetohighlyorderedbehavior,andatleastin astatisticalsensethisbehaviorexhibitssurprisinglysimpleregulari- ties:forexample,apowerlawdistributionrelatingthesizesandfre- quenciesof earthquakes. t>O c: 'C a ~ 0.3 :::l c: ..s E ,I i I i 11 I I -----3----- An UrbanMystery Let usstartwith apicture.Supposethatyou takeallof themetro- politanareaslistedin theStatisticalAbstractif theUnitedStates,130in number,andrankthembypopulation.And supposethatyouplot theranksof theseareasagainsttheirpopulations,usingalogarith- micscale.Whatyougetis Figure3.1. What is interestingaboutthisfigure?Of coursetheline relating rankto populationis,by definition,downwardsloping.But there aretwo thingsaboutthatline thatdid nothaveto betrue.First,it isprettyclosetoastraightline:thereseemstobesomethingcloseto alog-linearrelationshipbetweenrankandpopulation.Thatin itself isprettyinteresting;surelyit suggeststhatsomehiddenprincipleis at work. But evenmoreinterestingis the slopeof thatapproxi- matelylinearrelationship:it isverycloseto-1. Thereisanotherwayofputtingtheseresults,whichisalsoanoth- er nameforwhatis sometimescalledZipj's law: it is the rank-size rule.This rulesaysthatthepopulationof acityis inverselypropor- tionalto its rank.If therule heldexactly,thenumber2 city in a countrywould havehalf the populationof the biggestcity, the number3 cityone-thirdthatpopulation,andsoon. Obviouslythe ruledoesnotholdexactly.If youlook atonlythelargestmetropol- itanareas,you maybe unconvinced:Los Angelesis considerably morethanhalfaspopulousasNew York. But onceyou getdown therankingabit, thefit startstobecomealmostterrifyinglyexact. For example,the 10thrankedmetropolitanareain the United StatesisHouston,with 3.85millionpeople.The 100thrankedarea is Spokane,Washington,with 370,000people- closeenoughto 39 Figure3.1Zipt~sLaw:Thesizedistributionofmetropolitanareasin the UnitedStatesisstartlinglywelldescribedbyapowerlaw,withan exponentverycloseto1. Source:StatisticalAbstractif theUnitedStates,1993. 7.0 41 6.56.05.55.0 An UrbanMystery 4.5 In(S) I • 1990 +1940 * 1890 I 4.0 1 3.5 3 6 7 ~ -----6 4~----c Zipfs law is not quiteasneatin othercountriesasit is in the UnitedStates,butit stillseemsto holdin mostplaces,if youmake one modification:manycountries,for example,Franceand the United Kingdom,havea single"primatecity" thatis muchlarger thanalinedrawnthroughthedistributionofothercitieswouldlead you to expect.Theseprimatecitiesaretypicallypoliticalcapitals;it is easyto imaginethattheyareessentiallydifferentcreaturesfrom therestof theurbansample. What couldexplaintheexistenceof somethingthatlookssuspi- ciouslylikeauniversallawoncitysizes?Bearin mindthatthereare two thingswe needto explain.We needto understandwhy the rank-sizerelationshipis nearlylinearin thelogs,andwe needto understandwhy it stayssooddlyclosetoaslopeof -1. When urbaneconomistsattemptto explaintherank-sizerule, theyusuallyarguethatit reflectssomesortof hierarchyof central places.Hereis how thisreasoninggoes.First,wepointout thatan Figure3.2Zipf'sLawoverTime. Thesamepowerlawhasworkedrea- sonablywellforatleastacentury. Source:HistoricalStatisticsif theUnitedStates. 2~------------------------------------------------------------------------ 109.0 9.57.0 7.5 8.0 8.5 Log of citysize SELF-ORGANIZING ECONOMY 5.0 4.54.0~ 3.5 c '" 3.0 L.. ~·u 2.5.•.... 0 tl() 2.00...J 1.5 1.00.50.0 5.5 6.06.5 40 one-tenthof Houstonto bewell within reasonableuncertaintyof definitionandmeasurement.If you regressthelog of rankon the log of population,you geta coefficientof -1.003,with a standard errorof only0.01- aslopeverycloseto 1andverytightlyfitted. We areunusedto seeingregularitiesthisexactin economics- it is so exactthatI find it spooky.The picturegetsevenspookier whenyou find out thatthe relationshipis not somethingnew- indeed,therank-sizerule seemsto haveappliedto u.s.citiesat leastsince1890!Figure3.2showsdatatakenfromHistoricalStatistics of theUnitedStates,whichreportsthenumberof "urbanplaces"in specifiedsizeranges;I showthenumberwith morethan100,000, morethan250,000,morethan500,000,andmorethan1million for 1890,1940,and1990.(Thesedataarenot quitecomparableto thosein Figure3.1,becauseof thedifferencein definitionbetween anurbanplaceandametropolitanarea.)The pictureis notperfect,
Compartilhar