1Introduction
Mostsocialsystemsinvolvecomplexinteractionsamongmanyindividuals.Muchofeconomicshopesbothtogreatly
simplifyhumanbehavior,andtodoitinsuchawaythataggregatemacrofeaturescanbeeasilycharacterized.The
successofthistraditionalapproachtodescribehumanbehaviorhasonlyhadmixedsuccess.Oneareawherequestions
remainisnancewheremanyempiricalfeaturesaretroublingforexistingtheories.Severalofthefoundationsofthe
eldareinastateofdisarray,andnew,radicallydifferenttheoriesareappearing.
1
Oneofthedirectionsthatresearchershavebeentakingistheuseofagent-basednancialmarkets.Thesebottom-
upmodelsofnancialmarketsstartfromrstprincipalsofagentbehavior.Usingeithercomputational,ormore
sophisticatedmathematicaltoolstheyareabletodescribemacrofeaturesemergingfromasoupofindividualinteract-
ingstrategies.Sincetheearly1990′sI’vebeeninvolvedwithoneoftherstagent-basednancialmarketplatforms,
TheSantaFeArticialStockMarket.Now,withnearlyadecadeofexperienceinlookingatnancialmarketsfroman
agent-basedperspective,IwouldliketoturnmyattentionbacktotheSFImarket.Iwillexploresomeofthemarket’s
earlyhistory,andthedebatesanddesignquestionsthatwentintoitsdevelopment.Fromamodernperspectivethere
isstillmuchtolikeaboutwhattheSFImarketwastryingtodo.However,therearealsothingsthatIbelievemakeita
lessthanperfecttoolformodernresearchonnancialmarkets.
Muchofmymarketdesignphilosophystemsfromadesiretounderstandtheimpactofagentinteractionsand
grouplearningdynamicsinanancialsetting.Whileagent-basedmarketshavemanygoals,Iseetheirrstscientic
useasatoolforunderstandingthedynamicsinrelativelytraditionaleconomicmodels.Itisthesemodelsforwhich
economistsofteninvoketheheroicassumptionofconvergencetorationalexpectationsequilibriumwhereagents’
beliefsandbehaviorhaveconvergedtoaself-consistentworldview.Obviously,thiswouldbeaniceplacetoget
to,butthedynamicsofthisjourneyarerarelyspelledout.Giventhatnancialmarketsappeartothriveondiverse
opinionsandbehavior,arstleveltestofrationalexpectationsfromaheterogeneouslearningperspectivewasalways
needed.
2
ForthisreasonIhaveoftenusedtraditionaleconomicmodelswhichhavewelldenedhomogeneousequilibriaas
thecoreformodelbuilding.Thiscertainlyisn’tnecessaryforanagent-basedapproach,butIthinkatthisstageadeeper
understandingofdynamicsaroundwellstudiedequilibriaisessential.Theexistenceofanequilibriumbenchmarkalso
providesanimportanttestcaseformanyofthecomputationallearningtoolsinuse.Ifforsomecasestheycanactually
learnconvergencetoanequilibriumthentheirabilitiestoperformpurposefulsearchthroughthesetsofrulesaregive
1
SeeHeaton&Korajczyk(2002)alongwitharticlesinthatvolumeforanintroductiontobehavioralnanceissues.Averynicesummaryofthe
eldisShefrin(2000).MuchofthisworkisstillsubjecttocontroversyasshownbyRubenstein(2001).
2
SeeforSargent(1993)forfurtherthoughtsontheimportanceoflearningrationalexpectations.Also,Board(1994)providessomeimportant
incitesintothetheoreticalcomputabilityofequilibria.
1