版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领
文档简介
FromDeepLearningtoLLMs:AsurveyofAIinQuantitativeInvestment
arXiv:2503.21422v1[q-fin.CP]27Mar2025
BOKAICAO,TheHongKongUniversityofScienceandTechnology(Guangzhou),ChinaandIDEAResearch,InternationalDigitalEconomyAcademy,China
SAIZHUOWANG,TheHongKongUniversityofScienceandTechnology,HongKongandIDEAResearch,InternationalDigitalEconomyAcademy,China
XINYILIN,XIAOJUNWU,andHAOHANZHANG,TheHongKongUniversityofScienceandTechnology(Guangzhou),ChinaandIDEAResearch,InternationalDigitalEconomyAcademy,China
LIONELM.NI,TheHongKongUniversityofScienceandTechnology(Guangzhou),China
JIANGUO∗,IDEAResearch,InternationalDigitalEconomyAcademy,China
Quantitativeinvestment(quant)isanemerging,technology-drivenapproachinassetmanagement,increasinglyshapedbyadvancementsinartificialintelligence.Recentadvancesindeeplearningandlargelanguagemodels(LLMs)forquantfinancehaveimprovedpredictivemodelingandenabledagent-basedautomation,suggestingapotentialparadigmshiftinthisfield.Inthissurvey,takingalphastrategyasarepresentativeexample,weexplorehowAIcontributestothequantitativeinvestmentpipeline.Wefirstexaminetheearlystageofquantresearch,centeredonhuman-craftedfeaturesandtraditionalstatisticalmodelswithanestablishedalphapipeline.Wethendiscusstheriseofdeeplearning,whichenabledscalablemodelingacrosstheentirepipelinefromdataprocessingtoorderexecution.Buildingonthis,wehighlighttheemergingroleofLLMsinextendingAIbeyondprediction,empoweringautonomousagentstoprocessunstructureddata,generatealphas,andsupportself-iterativeworkflows.
CCSConcepts:•Generalandreference→Surveysandoverviews;•Appliedcomputing→Economics;
•Computingmethodologies→Artificialintelligence;Machinelearning.
ACMReferenceFormat:
BokaiCao,SaizhuoWang,XinyiLin,XiaojunWu,HaohanZhang,LionelM.Ni,andJianGuo.2018.FromDeepLearningtoLLMs:AsurveyofAIinQuantitativeInvestment.ACM/IMSJ.DataSci.37,4,Article111(August2018),
37
pages.
/XXXXXXX.XXXXXXX
1INTRODUCTION
Assetmanagementisacrucialandexpandingsegmentofthefinancialindustry,withQuantitativeInvestment(Quant)emergingasakeyapproachwithinit.Quantitativeinvestmentstrategies
*CorrespondingAuthor.
Authors’addresses:BokaiCao,mabkcao@,TheHongKongUniversityofScienceandTechnology(Guangzhou),ChinaandIDEAResearch,InternationalDigitalEconomyAcademy,China;SaizhuoWang,swangeh@connect.ust.hk,TheHongKongUniversityofScienceandTechnology,HongKongandIDEAResearch,InternationalDigitalEconomyAcademy,China;XinyiLin,xlin652@;XiaojunWu,xwu647@;HaohanZhang,hzhang760@,TheHongKongUniversityofScienceandTechnology(Guangzhou),ChinaandIDEAResearch,InternationalDigitalEconomyAcademy,China;LionelM.Ni,TheHongKongUniversityofScienceandTechnology(Guangzhou),China,ni@ust.hk;JianGuo,IDEAResearch,InternationalDigitalEconomyAcademy,China,guojian@.
Permissiontomakedigitalorhardcopiesofallorpartofthisworkforpersonalorclassroomuseisgrantedwithoutfeeprovidedthatcopiesarenotmadeordistributedforprofitorcommercialadvantageandthatcopiesbearthisnoticeandthefullcitationonthefirstpage.Copyrightsforcomponentsofthisworkownedbyothersthantheauthor(s)mustbehonored.Abstractingwithcreditispermitted.Tocopyotherwise,orrepublish,topostonserversortoredistributetolists,requirespriorspecificpermissionand/orafee.Requestpermissionsfrompermissions@.
©2018Copyrightheldbytheowner/author(s).PublicationrightslicensedtoACM.
2831-3194/2018/8-ART111$15.00
/XXXXXXX.XXXXXXX
ACM/IMSJ.DataSci.,Vol.37,No.4,Article111.Publicationdate:August2018.
111
ACM/IMSJ.DataSci.,Vol.37,No.4,Article111.Publicationdate:August2018.
111:2BokaiCaoetal.
leveragestatisticalanalysis,optimizationtechniques,andincreasingly,AIalgorithmstoidentifyandexploitmarketinefficiencies.Benefitingfromtheexponentialgrowthindataavailability,computa-tionalpower,andtechnologicalinnovations,theseapproachessignificantlyimproveinvestmentdecision-makingandprovideacompetitiveedgeinthefinancialmarket.
Amongvariousquantitativeinvestmentapproaches,alphastrategyhasreceivedconsiderableattentionforitsstrongcapacitytocapturemarketinefficienciesanditsnaturalalignmentwithAI-drivenpredictivemethods.Thepursuitof’alpha’referstopredictingindividualasset’sexcessreturnsoverthemarket’soverallperformance,suchasastockindex,andisthecentralfocusofportfoliomanagers.Thedevelopmentofalphastrategiestypicallyincludesfoursteps:datapro-cessing,modelprediction,portfoliooptimization,andorderexecution(asintroducedinsubsection
2.2
).Thesefoursub-tasks,thoughdistinct,arecloselyinterconnected,allworkingtowardsthecommongoalofmaximizingexcessreturnswhilecontrollingrisks.Comparedtootherquantitativeinvestmentstrategies,suchashigh-frequencytradingorarbitrage,alphastrategieshavebeenshowntohavegreatcapacityandeffectivenessbyexploitingmarketmispricings.Asaresult,alphastrategiesreceivethehighestattention,researchfocus,andmarketshareofresearchersandinvestors,representingthecoretechnologyinquantitativeinvestment.Inthissurvey,wetakethealphastrategyasarepresentativeexampleofquantitativeinvestmentandcenterourdiscussiononhowAIplaysaroleinthisfield.
Inrecentyears,theapplicationofdeeplearning(DL)techniquesinalphastrategieshasshownpromisingresults,demonstratingtheabilitytoidentifycomplexpatternsandrelationshipsinfinancialdatathataredifficulttodetectusingtraditionalquantitativemethods.Meanwhile,largelanguagemodels(LLMs),suchasGPT-series[
4
],BERT[
38
],andtheirfinancialvariants,haveshownremarkablepowerinunderstandingcontextualdata,generatingaccurateinterpretations,andreasoninglikehumananalysts.Therefore,theirapplicationinfinance,particularlyinquantitativeinvestment,hassparkedendlesspossibilities.
Thispaperfocusesontheevolution,application,andrespectiveadvantagesofDLandLLMsinquantitativeinvestment,withaspecificemphasisonalphastrategies,providingacomprehensivereviewoftheexistingliterature,anddiscussingthepotential,challenges,andlimitationsofhowLLMscouldenhanceDL-basedapproaches.
1.1EvolutionofAlphaStrategyInvestment
Theevolutionofthealphastrategycouldbeencharacterizedbyathree-stageprogressionfrommanuallabelingoftradingsignalstotheuseofdeeplearningmodelsandultimatelytoaneraofagentinteractionanddecision-makingbetweenLLMagents(Figure
1
).Intheearlystages,thefocuswasontraditionalstatisticalmodelingofmarketpatterns,relyingontheexpertiseofindividualresearcherstoidentifyprofitabletradingsignalsanddevelopcorrespondingmodels.However,thisapproachhadlimitationsduetothecomplexityoffinancialmarketsandthedifficultyofcapturingallrelevantfactorsinamodel.Itstillreliesheavilyontheskillandexperienceofhumanresearcherstoevaluateandexecutetradingstrategies.
Asthefieldhasmatured,theapplicationofdeeplearninghasopenedupnewpossibilitiesforquantitativeinvestment.Particularlyinthecontextofalpharesearch,deeplearninghasbeenshowntobeeffectiveinidentifyinginherentpatterns.Forexample,deeplearningmodelshavebeenusedtoanalyzefactorssuchasspatialinterconnectedness[
142
],long-termtemporaldependence[
178
],andnewssentiment[
68
]topredictthepricemovementsandmanagepositions.Whiletheuseofdeeplearningholdsgreatpromise,therearealsochallengesthatneedtobeaddressed.Onemajorchallengeistheriskofoverfitting,whichcanleadtopoorperformancewhenthemodelisappliedtonewdata.Anotherchallengeliesinimprovinginterpretabilityandaccuracytofurtherunderstand,reason,andinteractwithvastvolumesofmultimodaldata.Despitethesechallenges,
FromDeepLearningtoLLMs:AsurveyofAIinQuantitativeInvestment111:3
Fig.1.TheevolutionaryprocessofAlphainvestmentacrossdifferentstages.
theapplicationofdeeplearninginquantitativeinvestmentisexpectedtocontinuetogrowinbothacademiaandindustry,asinvestorsseektogainanedgeinacompetitivemarket.
Morerecently,LLMshaveemergedasapowerfultoolinquantitativeinvestment,characterizedbyrapiddevelopmentandenormousapplicationpotential,attractingsignificantattention.Theyexcelinunderstandingandprocessingmultimodaldataandpossessthepotentialtoautonomouslyhandlecomplextasksofreception,comprehension,andinferenceoverlarge-scaledatasets.Incurrentmainstreamresearch,LLMsprimarilyservetworolesinalphastrategy:aspredictors(subsection
4.1
)andasagents(subsection
4.2
),handlingvarioustasks.Inbothcases,theycontributehigh-levelinsightsbuiltupondeeplearningframeworksandhavethepotentialtofurtherevolveAI-powereddeeplearninginvestmentmethodsintotheAI-automatedstage.However,thepracticaldeploymentofLLMsisstillinitsearlystages.Wewillhighlighttheircurrentlimitationsanddiscusspotentialfuturedirectionsfortheirdevelopment(subsection
4.3
).
1.2MotivationandContributionofthisSurvey
TheuseofdeeplearningandLLMsinalphastrategiesofquantitativeinvestmenthasrecentlyseenasurgeofinterest,withmanystudiesfocusingontheapplicationinvariousaspectsofthealpharesearchpipeline.However,mostofthesestudiesarerelativelyisolatedinspecifictasksordisciplines,andthereisalackofaunifiedviewofthewholelandscapeofquantitativeinvestment,particularlyinthecontextofalphastrategy.Thissurveywillalsosystematicallysummarizetheevolutionofalpharesearchfromtheperspectiveofphasedalgorithmicevolution.Additionally,quantisafieldwhereresearchandpracticearehighlyinterconnected,whileexistingsurveypapersarelimitedtofillingtechnicalgapsbetweenpracticalopportunitiesandresearchtheoriesrelatedtothecombinationofLLMsandDL-basedalphamodels.There’salackofacomprehensiveframeworkandforward-lookingperspectiveonthefutureofdeeplearningandLLMsresearchworkflowsfromareal-worldviewpoint.
Toaddresstheseissues,thissurveypaperaimstoprovidereaderswithamoreintegratedandcomprehensiveviewofthealphastrategy.Weintendtoachievethisbysurveyingrelevantworks
ACM/IMSJ.DataSci.,Vol.37,No.4,Article111.Publicationdate:August2018
.
111:4BokaiCaoetal.
Fig.2.Theoverallframeworkofthispaper.
37,No.4,Article111.Publicationdate:August2018.
ACM/IMSJ.DataSci.,Vol.
ACM/IMSJ.DataSci.,Vol.37,No.4,Article111.Publicationdate:August2018.
FromDeepLearningtoLLMs:AsurveyofAIinQuantitativeInvestment111:5
thatcoveralltypesofdeeplearningandLLMstasksinthewholealphapipeline,offeringaholisticandinterconnectedviewofthefield.Moreover,oursurveypaperseekstoprovideabroaderresearchperspectivebystartingfromreal-worldapplications,highlightingthepracticalissuesandchallengesfacedbyinvestorstorevealthegenericresearchproblemsandpromotefuturestudie.TheoverallframeworkofthispaperisshownasFigure
2
.
Keycontributionsofthispaperinclude:
•Itprovidesacomprehensivesurveyoftheexistingresearchontheuseofdeeplearningandlargelanguagemodelsinalphastrategies,connectingdifferentworksinaconcreteresearchpipeline.Thissurveyisthefirstofitskindtocoverthetopiccomprehensively,offeringaholisticviewofthefield.
•ItIntroducesthedomainfromaninterdisciplinarystandpoint,emphasizingpracticalap-plicationstoderivekeyresearchquestionsforquantitativeinvestigations.Additionally,itdiscussesthemostchallengingproblemsfromapracticalperspectiveandoffersinsightsintopotentialfutureresearchdirections.
•Itsystematicallycomparesthetechnicalapproaches,strengths,andweaknessesacrossthethreestagesofquantitativeinvestment:traditionalstatisticalmodels,DL-basedmethods,andLLM-basedapproaches.Buildingontheiterativedevelopment,itidentifieskeygapsandleadsalphastrategiestothenextstage.
2BACKGROUND
Inthissection,weprovideabriefoverviewofalphastrategyanditsroleinquantitativeinvestment.Specifically,webeginbyintroducingalphastrategiesandthendiscussthealphapipelineinquant,whichservesastheframeworkfordevelopingandimplementingtheinvestmentprocess.
2.1AlphaStrategy
Alphastrategiesareinvestmentapproachesthatfocusonidentifyingopportunitiesyieldingreturnsthatexceedthemarketbenchmarkbyexploitinginefficienciesormispricings.Thesestrategiestypicallyconsistoftwocomponents:thealphasideandthehedgingside.
Thealphasideofanalphastrategyfocusesongeneratingexcessreturnsbyidentifyingprofitableinvestmentopportunities.Toachievethis,itaimstomakeaccuratepredictionsaboutthetrendsofindividualinstruments,sectors,ortheoverallmarket.Thesepredictionsarethenusedtoallocatecapitaltodifferentinvestmentinstrumentstomaximizereturns.Thealphasidefacesseveralcommonchallenges.Forexample,predictingassetpricesinvolveshandlinghigh-dimensional,noisydatawithcomplexpatterns,whileportfolioallocationrequiresoptimizingthetrade-offsbetweenriskandreturn.Deeplearningtechniqueshavebeenshowntobeeffectiveinaddressingthesechallenges,whichiswhyrecentworkshaveapplieddeeplearningtoalphastrategies.
Thehedgingsideofanalphastrategyfocusesonmanagingtherisksassociatedwithmarketmovements.Itsgoalistomitigatemarketrisks,ensuringthattheexcessreturnsgeneratedbythealphasidearenoteroded.Toachievethis,thehedgingsidetypicallyemployshedgingportfolios,suchasderivativeslikestockindexfuturesoroptions,whileaimingtominimizehedgingcosts.Thehedgingsidealsofacesseveralcommonchallenges.Forexample,selectingtheappropriatehedgingportfoliorequiresunderstandingtherelationshipbetweentheportfolioandtheunderlyinginvestments.Additionally,hedgingportfoliosmayincurtransactioncosts,whichcanaffectreturns.
Whileboththealphasideandthehedgingsideofanalphastrategyaimtogeneratereturnswhilecontrollingrisks,theyfacedifferentchallengesandrequiredistinctsolutions.Inthispaper,weprimarilyfocusonthealphasideofalphastrategiesanddiscussrecentworksthathaveemployeddeeplearningtechniquesandlargelanguagemodelstoaddressthechallengesfacedbythealpha
111:6BokaiCaoetal.
Adjustments
Monitor
InvestmentResearchExecution
RiskanalysisDecomposition
Pre-processingFactorminingModellingPortfoliooptimizationOrderexecution
ImputationCombinationLearningReturnObjectiveScheduling
Measurement
Trading
StandardizationSelectionReasoningOptimizationNetwork
CleaningExtractionInferenceRiskConstraintsAlgorithmic
Management
Programming
Returns
DataFactorsAttributesPositionsOrders
QuantitativeInvestmentSystem
RawData
Feedback
Exchange
Fig.3.Atypicalpipelineofquantitativeinvestment.
side.However,itisimportanttonotethatthehedgingsideisalsoanessentialcomponentofalphastrategies,andinterestedreaderscanrefertorelevantliteratureforfurtherstudy.
2.2AlphaPipeline
Initsessence,analphastrategyaimstoidentifywhichinstrumentstotrade,andactuallytradethemtoearnprofits.Initsmostsimpleform,thisideacanbepracticedinaone-stepway.Theinvestorcanjustlookatthemarket,findsomeassetshewantstotrade,andputthecorrespondingordersonthemarket.
However,thingsbecomemuchmorecomplexwhenthecapacityofthestrategygrowsintermsofboththeamountofmoneyandthenumberofassetsinvolvedintheinvestment.Inthiscase,tradingdecisionscannotbecarriedoutjustwithsimplehumanoperations.Instead,thewholestrategyneedstobestandardizedintoseveralsub-tasks,whereeachtaskiswell-formulatedandhasawell-developedtoolboxtodealwithit.AsshowninFigure
3
,analphastrategycanbedecomposedintoapipelineconsistingofseveralsub-tasks,whichwewillillustrateinthefollowing.
•Dataprocessing:Thewholepipelinestartsfromanalyzingthedatafromthemarket,andthedatacomeinvariousforms.Toperformmoderndataminingtechniquesonthesedata,thedatamustbefirstcleanedandstandardizedintounifiedforms.Thepre-processingstepishenceinvolvedtodothecleaning,standardization,andimputationofdata.Meanwhile,pre-processeddatacandevelopedintofeaturestobetterintegratemarketinformationandserveasinputforthenextstageofthemodel.
•Modelprediction:Althoughinvestmentscanbedonewithvariousmotivations,themostimportantoneshouldbetheexpectedfutureperformanceoftheassetprice.Therefore,toformaninvestmentdecision,weneedtofirstpredictthefutureoftheassetsofinterest,andthenmakedecisionsaccordingly.Pricepredictionaimstomakepredictionsaboutassets,suchastheirfuturepricechange,volatility,etc.Thispredictiontaskiswhatdeeplearningisgoodat,soabunchofdeeplearningtechniqueshavebeenstudiedinmakingbetterandmoreaccuratepredictions.
•Portfoliooptimization:Modelpredictionsthemselvescannotbedirectlyusedasinvestmentdecisions.Theyareutilizedintheportfoliooptimizationstagetogenerateinvestmentdecisions.Essentially,theportfoliooptimizationsteptakesthevariouskindsofmarketpredictions(e.g.
37,No.4,Article111.Publicationdate:August2018.
ACM/IMSJ.DataSci.,Vol.
FromDeepLearningtoLLMs:AsurveyofAIinQuantitativeInvestment111:7
ACM/IMSJ.DataSci.,Vol.37,No.4,Article111.Publicationdate:August2018.
Table1.Categorizationoffinancialdata.
Modality
Features
Example
References
Numerical
Quotedata
Regularinterval
Quotesthataregeneratedatregulartimeintervals
1-minutecandlestickchart
[131,
149]
Irregularinterval
Quotesthataregeneratedatirregulartimeintervals
Tick-leve
l1
orderbookdata
[16]
Fundamentaldata
Fundamentaldatasuchasrevenuesandprofits.
Financialstatements
[134]
Relational
Pairwiseedges
Therelationbetweenapairofentities
Knowledgegraph
[46,
75,
159,
160]
Hyperedges
Therelationinvolvingasetofentities
Sectorcategorization
[127]
Alternative
Text
Informationexpressedinnaturallanguage.
Socialmediaposts
[59,
87,
128,
162]
Images
Imagesthatarerelatedtothetradedasset
Satelliteimages
[101,
118]
Othermodalities
Anything
WiFitraffics,cellphonesignals
[64]
Simulation
Timeseries
Syntheticquotedata
Simulatedorders
[50,
84,
183,
184]
Tabular
Databasestructuredinatabularform
Simulatedfinancialstatements
[47,
126,
133]
pricechange,volatility)asinput,andoutputsthecorrespondinginvestmentdecisionssuchastheamountofmoneyallocatedforeachassetinthenextholdingperiod(alsoknownaspositions).Thecoreprobleminthisstageistofindanoptimalportfoliooptimizationthatmaximizestheobjectivedefinedbyinvestorssuchasmaximizingtheexpectedrisk-adjustedreturns,subjecttotheconstraintsdefinedsimultaneously,suchasthemaximumvolatilityconstraintorthediversityconstraintdefinedforriskcontrol.Traditionally,thisproblemisformulatedasanoptimizationproblemandthecorrespondingoptimizationtechniquescanbeappliedtosolvethisproblem.
•Orderexecution:Theportfolioallocationsneedtobeimplementedbyactuallyputtingordersonthemarketandmakingthedeal.Andthisorderexecutionstepisbynomeansaneasytaskforalphastrategiesthatinvolvealargeamountofmoney.Thisisbecauseordersplacedonthemarketwillinevitablybringfluctuationsonthemarket,drivingthepricetotheoppositedirectionthatisbeneficialfortheinvestor.Andwhenthetradevolumebecomeslarge,thiseffectisalsomagnifiedandmightintroducegreatlosstothestrategy.Hence,theorderexecutionstageisinvolvedtominimizesuchlossbysplittingbigordersintosmalleronesanddecidingtheappropriatetimetoexecutethem.Traditionally,thisproblemhasbeenformulatedasanoptimalcontrolproblemwherethedynamicsofthelimitorderbookareextensivelystudied.
Theinvestmentpipelineisnotanopen-loopsystem.Instead,itneedsfeedbackfromthemarkettoadjustitsbehaviors.Themonitormoduleisthereforeinvolvedintheentirepipelinetoconductriskanalysis,collectfeedbacksignals,andmakecorrespondingadjustmentstothepreviousprocesses.Forexample,theriskmanagementmodulecalculatestheriskexposureofthecurrentportfoliotodifferentsectors.Then,fromtheperspectiveoftheoverallprocess,itmakesbalancingorneutralizationadjustmentsbyinjectingconstraintsoradjustingtheobjectivefunctionsintheportfoliooptimizationmodule.
3DEEPLEARNINGINALPHAPIPELINE
Deeplearninghasbeenwidelyappliedthroughoutthewholealphapipeline.Inthissection,wesystematicallyanalyzehowdeeplearningisutilizedtoenhancetraditionalalpharesearchateachsub-taskofthepreviouslydiscussedpipeline.
111:8BokaiCaoetal.
High
Uppershadow
Amount(BillionDollar)
Open
Real
Body
Close
Lowershadow
Low
(a)Candlestickchart[
1
](b)LimitOrderBook[
153
]
60
50
40
30
Q3/FY2021Q2/FY2021
Q1/FY2021
20
10
0
TotalRevenueGrossProfitNetincome
(c)FundamentaldataofMicrosoft.
Fig.4.Examplesofnumericaldata.
3.1DataProcessing
Asstatedbefore,thedatausedforalpharesearchneedtobefirstpre-processedintounifiedformat,andthenusedforprediction.Herewedividedatabeforeandafterpre-processingintotwotypes,namelyrawdata,andfeatures.
3.1.1RawData.Financialmarketsconstantlygenerateheterogeneousdatawithvariousmodalities.Basedontheirsourceandmodality,wecategorizefinancialdatausedinquantstrategiesasnumericaldata,relationaldata,alternativedataandsimulationdata.AsummaryandacomparisonofdifferenttypesofdataarepresentedinTable
1
andwewillelaborateoneachoftheminthefollowing.
Numericaldata.Numericaldataarethemostwidespreadtypeofdatainthefinancialworld,itcanbecategorizedasquotedatathatindicatethemovementofassetpricesofarbitraryfrequency,andfundamentaldatathatreflecttheoperationsoftheunderlyingeconomicentitiesofsecurities.
Quotedatatypicallyincludescandlestickchartandlimitorderbook,asillustratedinFigure
4
.Candlestickchartisawaytoillustratethepricemovementofafinancialinstrument,whichconsistspricesatfourdifferentinformationdimensioninantimeinterval,namelytheopen,close,high,andlowprices.Alimitorderbookreferstoanelectroniclistofbuyandselllimitordersorganizedbypricelevels.Fundamentaldataiswidelyusedbyanalyststodeterminetheintrinsicvalueofafinancialinstrument.Importantsourcesoffundamentaldataarebalancesheets,incomestatements,andcashflowstatementsfromfinancialstatements.
Financialmarketsareintrinsicallydynamic,numericaldataareoftentreatedastimeseries.
Undersuchcircumstances,averyimportantpropertyistheirfrequency,i.e.,theintervalatwhichnewdatapointsaregenerated.Generallyspeaking,fundamentaldatahavethelowestfrequency,changingeveryfewmonths.Incontrast,quotedatahavefrequenciesatmultiplelevels,rangingfromtheloweronessuchasweek-orday-leveltothehigheronessuchasminute-orsecond-levelfrequencies.Moreextremely,thequotedatawiththemaximumtemporalresolution,namelytick-leveldata,recordsthehistoryofeachorderandtransactionsequentially.
Relationaldata.Whilenumericaldatadescribesindividualfinancialentities,therelationshipsbetweenthemcanalsoimpactmarkettrends.Werefertosuchdataasrelationaldata,whichdescribestheubiquitousrelationshipsbetweentwoormorefinancialentities.Formally,relationaldataareusuallyrepresentedasagraphG=v×E,wherevisthesetofnodesandEisthesetofedges.Theedgescanbeeitherpairwiseedgesorhyperedges[
147
],basedonthenumberofentitiesinvolvedintherelations.Inpractice,relationaldatahaverichsemantics,rangingfromthebusinessrelationshipsbetweencompanies,tovariouseventsinvolvingfinancialentities,tothecorrelations
37,No.4,Article111.Publicationdate:August2018.
ACM/IMSJ.DataSci.,Vol.
ACM/IMSJ.DataSci.,Vol.37,No.4,Article111.Publicationdate:August2018.
FromDeepLearningtoLLMs:AsurveyofAIinQuantitativeInvestment111:9
(a)Pairwiseedges[
46
](b)Hyperedges[
127
]
Fig.5.Relationaldataexamples.
andcausalrelationshipsbetweenentitiesfromastatisticalperspective.AnillustrationofpairwiseedgesandhyperedgesispresentinFigure
5
.
(1)Pairwi
温馨提示
- 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
- 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
- 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
- 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
- 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
- 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
- 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。
最新文档
- 小绞车司机操作规程
- 2026年本溪市高考临考冲刺语文试卷含解析
- 医学26年:外科抗菌药物应用 查房课件
- 【2026】公司治理、企业社会责任和财务绩效关系实证研究8400字(论文)
- 26年失能老人心理服务参考课件
- 医学26年:神经重症中心建设 查房课件
- 26年护理服务用语规范课件
- 年产450片板束新建项目、年产管束板片4000片项目可行性研究报告模板-备案审批
- 医学26年:淋巴管肌瘤病诊疗 查房课件
- 26年老年节日话题沟通技巧课件
- 鸡鸭冻品专业知识培训课件
- 出货检验流程标准作业指导书
- 2025年中医全科医生转岗培训考试综合能力测试题及答案
- 医学课题申报书技术指标
- 交通安全协管员考试题库及答案解析
- 地铁区间高架桥施工安全风险评估及改进方案
- 2024煤矿地质工作细则
- 苏州文华东方酒店公区概念设计方案文本
- 2025年安徽中烟工业公司岗位招聘考试笔试试卷(附答案)
- 2025中小学教师考试《教育综合知识》试题及答案
- 暖通可行性研究报告
评论
0/150
提交评论