原文翻譯：監(jiān)管科技工具在審慎監(jiān)管及疫情期間的應(yīng)用-bis

dentialsupervisionandtheiruseandemic疫情期間的應(yīng)用iBFinancialauthoritiesusesuptechtoolsforarangeofactivities,includingdataanalyticsforprudentialsupervisionwhoseusecaseshaverecentlygrown.AnearlierFinancialStabilityInstitute(FSI)publicationfoundthatmostsuptechtoolswereusedforreportingandmisconductlyfewdeployedforprudentialsupervisiondiCastrietal(2019)).TheFinancialStabilityBoard(FSB(2020))foundsimilarresults,thoughitobservedariseinsuptechusecasesforprudentialpurposes.TheFSBattributedtheincreasetotheautomationofcertainrepetitivetasksinprudentialsupervision.——包括以前在現(xiàn)場(chǎng)執(zhí)行的一些任務(wù)——擁有現(xiàn)有超級(jí)技術(shù)工具的Thepandemicpromptedauthoritiestoleveragemoresuptechtoolsinday-to-daysupervision.Travelrestrictionsandsocialdistancingprotocolsseverelycurtailedon-siteinspectionsandledtoasimultaneousshiftofnearlyallsupervisoryactivitiestoanoff-sitesurveillancepproachTohelpsupervisorsassesstheprudentialsoundnessoffinancialinstitutionsremotely–includingsometasksthatwerepreviouslyconductedon-site–authoritieswithexistingsuptechtoolsusedthemmoreextensively;atthesametime,theyalsorecognisedtheneedtodevelopnewdataanalyticstoolsforprudentialpurposes.Therefore,itisnotsurprisingthatauthoritiesreportedusing,developingorexperimentingwith71discreteprudentialsupervisorytoolsasofthispublicationupfromonly12toolsin2019.如人工智能和機(jī)器學(xué)習(xí)——為當(dāng)局提供了部署一系列范圍的機(jī)會(huì)。Broadertechnologicaldevelopmentsfacilitatedthemigrationofsupervisoryactivitiestoavirtualenvironmentandunderpinnedthewideruseofsuptechtoolsforprudentialpurposes.Datamanagementplatforms,fileexchangeprotocols,collaborationsoftwareandcommunicationtoolsenabledtheshifttovirtualsupervision,partiallyoffsettinglimitedon-siteinspections.Meanwhile,thegrowthofnon-traditionaldatasourcesthatcanhaveabearingonafirm’sriskprofileandtheadventofnewanalyticaltoolstohelpprocessandanalysedata–suchasartificialintelligenceandmachinelearning–providedauthoritieswithopportunitiestodeployarangeofsuptechtoolsforprudentialsupervision.本文盤(pán)點(diǎn)了20個(gè)司法管轄區(qū)用于審慎目的的超科技數(shù)據(jù)分析hThispapertakesstockofsuptechdataanalyticstoolsusedforprudentialpurposesin20jurisdictionsandexplorestheassociatedbenefits,risksandimplementationchallenges.ThefindingsarebasedonresponsestoanFSIsurveybymembersofitsInformalSuptechNetwork,combinedwithfollow-upinterviewswithselectedjurisdictions.Suptechdataanalyticsforprudentialsupervisionincludetoolstosupportsupervisoryriskassessments,suchascredit,market,liquidityandoperationalrisksandtheirimplicationsforfirm-wideearnings,capitaladequacyandgovernance.本文考察的71種審慎的suptech工具分為三類(lèi)，隨后又分為T(mén)he71prudentialsuptechtoolsexaminedinthispaperareclassifiedintothreecategoriesandsubsequentlydividedintosubcategories.Thetop-tiercategoriesarebasedonthetypeofdatathetoolsscrutiniseandtivedataiitoolsforquantitativedata”;and(iii)“toolsforqualitativeandquantitativedata”.Withineachofthethreecategoriesarevarioussubcategoriesthatclassifyhowthetoolsareused.Toolsthatrelyonmainlyqualitativedatarepresentslightlymorethanhalfofthoseexamined;thesetoolsareusedfortextanalysis,textsummarisation,informationclassificationorsentimentanalysis.Toolsthatmainlylookatquantitativedataandthosethatutilisebothquantitativeandqualitativedataaccountforapproximately25%ofusecaseseach.Theformerisusedforriskidentification,whilethelattermaybeusedfornetworkanalysis,peergroupidentificationorautomationofinspections.效率。掃描定性數(shù)據(jù)的工具通常使用自然語(yǔ)言處理(NLP)和其他人化檢查過(guò)程的各個(gè)方面，為更高級(jí)別的任務(wù)釋放監(jiān)督資源。Whilesuptechtoolsvaryindesignandpurpose,allshareatleastoneoftwooverarchingobjectivesofextractingdeepersupervisoryinsightsandenhancingtheefficiencyofthesupervisionprocess.Toolsthatscanqualitativedataoftenusenaturallanguageprocessing(NLP)andotherartificialintelligencetocombthroughanastonishingarrayofmaterialstoquicklyfind,summarise,classifyandpresentrelevantinformationforfurtherreview.Thesetoolsallowsupervisorstoconsiderabroaderrangeofinformationintheirprudentialriskassessments.Toolsthatrelyonquantitativedatafacilitateidentificationofhigh-riskbanksanddriversofspecificriskswithinbanks,enablingabetterallocationofsupervisoryresources.Toolsthatconsiderqualitativeandquantitativedataallowsupervisorstoassessrelationshipsbetweenentitiesthatmaynotbeapparenttothehumaneye;toenableconstructionofenhancedbankpeergroups,facilitatingmoreconsistentsupervisionoffirmswithsimilarriskprofiles;ortoautomateaspectsoftheinspectionprocess,freeingupsupervisoryresourcesforhigher-ordertasks.風(fēng)險(xiǎn)識(shí)別工具也被用來(lái)發(fā)現(xiàn)可能被錯(cuò)誤分類(lèi)或撥備不足的潛在信用SuptechtoolswerewidelydeployedduringtheCovid-19pandemic,particularlythosethatscrutinisequalitativedataandsupportriskidentification.Themigrationofon-siteactivitiestooff-sitework,inconjunctionwithvariousadhocreportsrequestedduringthepandemic,addedtothemountingstackofexistingstructuredandunstructureddatathatrequiredreview.Inthevirtualenvironment,suptechtoolsprovedindispensable,enablingsupervisoryreviewsofcorporategovernanceandassetquality,bothofwhicharetypicallyassessedon-siteandoftendriveafirm’soverallriskprofile.NLPtoolshelpedsupervisorspinpointcorporategovernancerisksfromvoluminousdocumentsthatmightotherwisenothavebeenpossible.Riskidentificationtoolswerealsoutilisedtospotpotentialcreditexposuresthatmaybemisclassifiedorunderprovisioned,providingsupervisorswithaspecificlistofborrowersforfollow-up.某些suptech工具和驅(qū)動(dòng)suptech輸出的參數(shù)的非結(jié)構(gòu)化數(shù)據(jù)。制定或更新有助于促進(jìn)監(jiān)管支持和指導(dǎo)當(dāng)局部署各種suptech工具Notwithstandingthesetangiblebenefits,formidableimplementationchallengesremain,hamperingwideradoptionandacceptanceofsuptechtools.Akeyissueisthelimiteddatascienceskillsofsupervisors.Toaddresstheskillsgap,continuedtrainingofsupervisorscombinedwithhiringdatascientistsmayhelp.Othercriticalissuesinvolvedataquality,particularlytheunstructureddatawhichunderpinsomesuptechtoolsandtheparametersthatdrivesuptechoutputs.Anoverlytightcalibrationmightleadtothetoolmissingsupervisoryissues,whileaveryloosesettingcanresultinflaggingtoomanyirrelevantissues.Thesechallengesmaypointtoabroaderneedtodeveloporupdateasuptechstrategythathelpstofacilitatesupervisorybuy-inandguideauthorities’deploymentofvarioussuptechtools.監(jiān)管者可以更少地依賴(lài)自己的判斷，而更多地依賴(lài)suptech輸出來(lái)具輸出之間的緊張關(guān)系并概述其各自作用的明確政策可能會(huì)有所幫Assuptechtoolstakeonagreaterroleinprudentialsupervision,supervisoryjudgmentmaydiminish.Suptechtoolsareautomatinglower-value,labour-intensivetasksandsupportinghigher-value,judgment-basedfunctions.Thesetrendsarenowaccelerating,particularlythedevelopmentoftoolsthattargetcomplexriskassessmentsthatentailjudgment.Asthesetoolsgetoperationalised,supervisorscouldrelylessontheirownjudgmentanddependmoreonthesuptechoutputtoidentifykeysupervisoryissues.Ifthistranspires,itmayleadtosupervisoryblindspotsandabroaderlossofinstitutionalknowledgebasedontheartofjudgment-basedsupervision.Whileauthoritieshaveemphasisedthatsuptechtoolssupport,ratherthanreplace,supervisoryjudgment,explicitpoliciesthatacknowledgethetensionsbetween,andoutlinetherespectiverolesof,supervisoryjudgmentandsuptechtooloutputs,couldhelp.虛擬檢查和更廣泛使用超級(jí)技術(shù)工具的經(jīng)驗(yàn)引發(fā)了關(guān)于監(jiān)管未Experiencewithvirtualinspectionsandwideruseofsuptechtoolshavesparkedabroaderdebateonthefutureofsupervision.Duringthepandemic,authoritiesdemonstratedtheabilitytoshiftallsupervisoryactivitiestoanoff-sitestance.Thishasblurredthelinesbetweenon-andoff-siteroles,promptingarethinkonthemodesofsupervisioninthepost-pandemic,digitalera.Theshifttovirtualsupervision,however,wasnotfrictionless.Onthesupervisoryside,managingremoteteamsbecameachallenge;andwhilecommunicationtoolsenabledvirtualmeetings,therearenogoodsubstitutesforin-personmeetingswithbankstaff,whichprovidesupervisorswithcriticalinsightsonthequalityofabank’sinternalcontrolsandriskmanagementpractices.Onthetechnologyfront,thepandemichighlightedsomegapsinauthorities’owntechnologicalinfrastructure,whileexposingvariedtechnologicalcapabilitiesofsupervisedfirms.Whiletherewillalwaysbeacrucialroleforon-siteinspections,theremaybescopeformoresupervisoryworktobeconductedoff-site,depending,inpart,ontheevolutionoftechnologicalinnovations,includingthebroaderdeploymentofsuptechtoolsinprudentialsupervision.一、簡(jiǎn)介具有監(jiān)管和非監(jiān)管職能的機(jī)構(gòu)(即沒(méi)有監(jiān)管授權(quán)的金融情報(bào)機(jī)構(gòu))。展的所有活動(dòng)中都可以找到suptech用例——FSIInsightsno(“Thesuptechgenerations”)definedsuptechastheuseofinnovativetechnologybyfinancialauthoritiestosupporttheirworkForthispurpose,“innovativetechnology”referstotheapplicationofbigdataorartificialintelligenceAItotoolsusedbyfinancialauthoritiesFinancialauthoritiesreferstoauthoritieswithsupervisoryandnonsupervisoryfunctionsiefinancialintelligenceunitswithoutrymandatesAssuchsuptechusecasescanbefoundinthewholerangeofactivitiesthatfinancialauthoritiesundertake–fromdataionincludingdatamanagementtodataanalyticsChartWithindataanalyticssuptechusecasescanhelpinmarketoversight,conductupervisionandprudentialsupervisionThispaperfocusesonsuptechdataanalyticstoolsforprudentialsupervision.用于審慎監(jiān)管的Suptech數(shù)據(jù)分析工具僅占總用例的一小部suptech具相比，Suptechdataanalyticstoolsforprudentialsupervisionmadeuponlyasmallfractionoftotalusecases,butthisproportionmaybechanging.Ofthe99suptechusecasesexaminedinFSIInsightsno19,themajoritywereforreporting(32%)andmisconductanalysis(30%),withonlyafewforprudentialsupervision(12%).3FSB(2020)foundasimilarpatterninthedistributionofsuptechusecasesbutnotedtheincreasedinprudentialusecasesinrecentyears.Itattributedthisincreasetotherelativelyrule-basednatureofsomeprudentialtasks.Authoritiesthereforewereabletoeasilycodifysomeoftheseassessmentsinsuptechtools,thusintroducingefficienciesinthesupervisoryprocesses.Indeed,comparedwiththesuptechdataanalyticstoolsforprudentialsupervisionexaminedin2019,thenumberoftoolsexaminedforthispaperrepresentsasignificantincrease(Chart2).大流行限制了監(jiān)管活動(dòng)，并可能為開(kāi)發(fā)更多用于審慎監(jiān)管的Thepandemichasconstrainedsupervisoryactivitiesandmayhaveprovidedanimpetusforthedevelopmentofmoresuptechusecasesforprudentialsupervision.On-siteinspectionshavebeenseverelylimitedornon-existentinalmostalljurisdictions.Thepandemicforcedsupervisionworktofocusmoreonoff-sitemonitoring,usingwhateverdataandanalyticstoolssupervisorshad.Authoritieswithoperationalsuptechtoolsfoundthemquiteusefulunderthecircumstances.Atthesametime,authoritiesconsideredadditionalusecasesthatwouldhavebeenusefulgivenlimitedon-siteinspections.Theshifttooff-siteactivitiesduringthepandemic,plustheexpectationthatthe“newnormal”mightcontinuetomeanlesson-siteactivities,maypushauthoritiestoleveragemoresuptechtoolsonapermanentbasis.本文概述了suptech數(shù)據(jù)分析工具在全球多個(gè)司法管轄區(qū)進(jìn)行的調(diào)查的21份答復(fù)(有關(guān)對(duì)調(diào)查作出答復(fù)的當(dāng)局列表，請(qǐng)參見(jiàn)附件1)。調(diào)查答復(fù)補(bǔ)充了對(duì)一些答復(fù)當(dāng)局的采訪(fǎng)，以詳細(xì)討論他們的suptech中的實(shí)踐的一些觀察，包括治理、用例識(shí)別、向主管部署和有效性衡量。第4節(jié)研究了在大流行期間如節(jié)結(jié)束。Thispaperprovidesanoverviewofthestateofplayofsuptechdataanalyticstoolsforprudentialsupervisioninanumberofjurisdictionsaroundtheworld.Itbenefitedfrom21responsestoasurveysenttomembersoftheFSI’sInformalSuptechNetwork(seeAnnex1foralistofauthoritiesthatrespondedtothesurvey).Surveyresponsesweresupplementedwithinterviewsofsomerespondingauthorities,todiscusstheirsuptechtoolsindetailand/orclarifytheirresponses.Section2describesandclassifiesthesetoolsaccordingtothedatatheyanalyseand/ortheirobjectives.Section3offerssomeobservationsonauthorities’practicesthroughoutthesuptechlifecycle,includinggovernance,identificationofusecases,deploymenttosupervisorsandmeasurementofeffectiveness.Section4examineshowsuptechtoolsarebeingusedduringthepandemicanddescribesareaswheretheyprovedtobeparticularlyuseful.Section5discussespracticalconsiderationsinusingsuptechtools,includinglessonslearnedduringthepandemic.Section6concludes.二、審慎監(jiān)管的監(jiān)管科技數(shù)據(jù)分析工具類(lèi)型本文考察了71種用于審慎監(jiān)管的suptech數(shù)據(jù)分析工具。對(duì)用于報(bào)告、數(shù)據(jù)管理(即驗(yàn)證、可視化、存儲(chǔ)、聚合等)、進(jìn)行監(jiān)督和反洗錢(qián)(AML)監(jiān)督的用例超出了本文的范圍。接下來(lái)的討論僅與Thepaperexamines71suptechdataanalyticstoolsforprudentialsupervision.Authoritiesthatrespondedtothesurveyreported130suptechusecases.Outofthese,weconsideredusecasesthatareforreporting,datamanagement(ievalidation,visualisation,storage,aggregationetc),conductsupervisionandanti-moneylaundering(AML)oversightasout-of-scopeforthispaper.4Thediscussionsthatfollowpertainonlytotheremaining71suptechdataanalyticstools.分為子類(lèi)(圖3)。廣泛的分類(lèi)基于工具主要查看的數(shù)據(jù)類(lèi)型——定查自動(dòng)化。Suptechtoolsforprudentialsupervisionaregroupedintothreebroadclassifications,eachofwhichcanbefurtherclassifiedintosubcategories(Chart3).Thebroadclassificationsarebasedonwhattypesofdatathetoolsmainlylookat–qualitative,quantitativeorboth.Thesubcategoriesarebasedonhowthetoolsareused,withsometoolsclassifiedinmorethanonesubcategory.Toolsthatmainlyfocusonqualitativedatamaybeusedfortextanalysis,textsummarisation,informationclassificationorsentimentanalysis.ToolsthatmainlylookatquantitativedataareusedcationToolsthatlookrelativelyequallyatbothqualitativeandquantitativedatamaybeusedfornetworkanalysis,peergroupidentificationorautomationofinspections.定量數(shù)據(jù)的工具和同時(shí)使用定性和定量數(shù)據(jù)的工具各占四分之一左析定性和定量數(shù)據(jù)的工具是suptech顯示出巨大潛力的另一個(gè)領(lǐng)是如何改進(jìn)它們。Toolsthatmainlyusequalitativedatamakeupslightlymorethanhalfofthoseexamined.Toolsthatmainlyusequantitativedataandthosethatusebothqualitativeandquantitativedataeachaccountforaboutaquarter.Intermsofsubcategories,toolsfortextanalysisarethemostcommon,followedbytoolsforriskidentification,informationclassificationandautomationofinspections.Theprevalenceoftoolsforqualitativedatareflectstheimportanceofaidingsupervisorsinreviewingdocumentsintextformat,whichuntilnowtheystillhavetodomanually.Toolsthatanalysebothqualitativeandquantitativedataisanotherareawheresuptechshowsgreatpotential,sincetheseenabletheintegrationofbothtypesofdatafordeeperinsights.Meanwhile,supervisorsalreadyhaveexistingtoolstoanalysequantitativedata,sothefocusofsuptechissimplyonhowtoimprovethem.許多工具已經(jīng)投入使用，幾乎所有工具都已經(jīng)或正在內(nèi)部開(kāi)發(fā)(圖4)。操作工具占所檢查工具的48%，而開(kāi)發(fā)中和實(shí)驗(yàn)工具分別占22%和30%。同時(shí)，只有三個(gè)工具是由外部獨(dú)家開(kāi)發(fā)的，六個(gè)是Manyofthetoolsarealreadyoperational,andalmostallofthemwereorarebeingdevelopedinternally(Chart4).Operationaltoolsmakeup48%ofthetoolsexamined,whilein-developmentandexperimentaltoolsmakeup22%and30%,respectively.Meanwhile,onlythreeofthetoolsweredevelopedexclusivelybyexternalparties,andsixwerejointcollaborationsbyinternalandexternalparties.Therestwereorarebeingdevelopedinternally.Quiteafewoftheagencieshavedatascientiststhatdevelopthetools,oftenwithinputfromlinesupervisors.Fortoolsrelyingonsomeexternalassistancetobolsterinternaldevelopment,advicecancomefromuniversitiesorrelatedresearchbodies.Theextentofcollaborationwithlinesupervisorsduringthedevelopmentphaseappearstobecorrelatedwiththedegreetowhichthetooliseventuallyintendedtobeusedmorewidelybysupervisors.Thewiderthesupervisoryusethatisenvisioned,themoreinitialinputissoughtfromlinesupervisors.定性數(shù)據(jù)分析工具析涵蓋了自然語(yǔ)言處理(NLP)中的一系列用例。文本通常是機(jī)密或oT會(huì)議記錄分析器和意大利銀行(BdI)公司治理分析.Textanalysisusesmachinelearning(ML)toobtainspecificationfromadocumentTextanalysiscoversarangeofusecasesinnaturallanguageprocessing(NLP).Textisoftenunstructureddatathatiseitherconfidentialornon-confidential,butwhichservesasupervisorypurpose.Documentsthatcanbereviewedinanautomatedwayrangefromcontractstoauditors’statements,frompressarticlestooperationalriskreporting,frommeetingminutesatafirmtobankriskprofiles.Thegoalistoautomatethesearchingofinformationinordertosavesupervisors’timeandenergy.ExamplesofsuchtoolsincludetheBankofSpain’s(BdE)textminingforwisersampling,theEuropeanCentralBank’s(ECB)AutomatedTopicModeling,theBankofThailand’s(BoT)boardminuteanalyserandtheBankofItaly’s(BdI)corporategovernanceanalysis.更好地識(shí)別(與手動(dòng)流程相關(guān))銀行監(jiān)管審查和評(píng)估敘述中的一般主以幫助深化對(duì)銀行治理的分析(有關(guān)BoT和BdI工具的更多詳細(xì)信Whilesuchtoolsareusedforvariouspurposes,allcanidentifycommonlyusedwordsinacertaincontextandcananalyseawidesetofdocumentsacrossarangeofsupervisoryusecases.TheBdE’stoolanalysesunstructureddatafromaninstitution’screditfilestoobtainasampleofcreditexposuresthatmayhavebeenwronglyidentifiedas“performing”.AttheECB,AutomatedTopicModellinganalysestextualdatatobetteridentify–inrelationtomanualprocesses–generaltopicswritteninbanks’SupervisoryReviewandEvaluationnarratives.6TheBoTtoolanalysesboardminutestoidentifyrisksthatarebeingdiscussedandtoassessthedegreeofboardengagement.Similarly,theBdItoolaimstoapplytextminingtoboardofdirectors’meetingminutestohelpdeepentheanalysisofbankgovernance(seeBox1inAnnex2formoredetailsontheBoTandBdItools).閱讀，例如從多個(gè)頁(yè)面創(chuàng)建概述段落。7巴西中央銀行(BCB)的語(yǔ)言提取(LEX)工具包括開(kāi)發(fā)摘要工具以及其他15個(gè)用例(有關(guān)Textsummarisationistheprocessofhighlightingkeypointsinlargedocumentsforquickersupervisoryconsumption.Textsummarisationiscloselyrelatedtotextanalysis,butthedifferenceisthattheformerfocusesonsummarisingtextwhilethelatterfocusesonfindinginformation.Summarisationtoolscondensetheamountoftextintoamanageableportiontoreadquickly,suchascreatinganoverviewparagraphfrommultiplepages.7TheCentralBankofBrazil’s(BCB)MARIAsummarisationtoolusesanunsupervisedMLalgorithmtosummariselongtexts,allowingsupervisorsandmanagementtoscreenandevaluatecontentbeforehand.MARIAiscurrentlybeingimprovedwithstate-of-the-artalgorithms,whichareinthefinalstagesoftraining.serveBankofNewYorksFRBNYLanguageExtraction(LEX)toolincludesdevelopmentofasummarisationtoolalongside15otherusecasesseeBoxinAnnex2formoredetailsonLEX).類(lèi)。該工具使用NLP，預(yù)計(jì)將審查BdE員工如何對(duì)他們上傳到系統(tǒng)Informationclassificationseekstounderstandpatternsfromlargeamountsofunstructureddata,withtheintenttoclassifyandstructureinformationinamoreorganisedway.Authoritiesareclassifyingandorganisingawiderangeoftext,includingregulatorysubmissions,newsarticlesandotherdocuments.TheGuernseyFinancialServicesCommission(GFSC)hasanexperimentaltoolthatclassifiesdocumentswithmaterialsupervisoryconcerns,andaimstoreducesupervisoryeffortby“flagging”onlythosedocumentswithmaterialissuesformanualreview(seeBox3inAnnex2formoredetailsonthistool).TheBdEhasanexperimentaltoolthatdeterminesautomaticallywhetherasupervisorydocumenthasbeencorrectlyclassifiedormisclassified.Thetool,whichusesNLP,isexpectedtoreviewhowBdEstaffclassifydocumentstheyuploadintothesystemandidentifyanypotentialmisclassification.8ThiswillhelpimprovethequalityofunstructureddataintheBdE’ssystem.例如BoT用于衡量機(jī)構(gòu)對(duì)其Covid相關(guān)政策和救濟(jì)措施的敏感性和意見(jiàn)的工具?？ㄋ柦鹑谥行谋O(jiān)管局(QFCRA)正在開(kāi)發(fā)一種推文構(gòu)正計(jì)劃開(kāi)發(fā)情感分析工具或?qū)⑶楦蟹治黾{入其文本分析工具的未SentimentanalysisusesNLPtodeterminewhetherdataarepositive,negativeorneutral.Amongthesuptechtoolsexamined,relativelyfewfocusprimarilyonsentimentanalysisdespitetherelativelyhighlevelofinterestsuchtoolsattractedfromauthoritiesinpreviousyears.Afewauthoritieshavestandalonesentimentanalysistools,suchastheBoT’stooltomeasureinstitutions’sensitivityandopiniontowardsitsCovid-relatedpoliciesandreliefmeasures.TheQatarFinancialCentreRegulatoryAuthority(QFCRA)isdevelopingatweetsentimenttool,whichwillallowsupervisorstobettergaugepublicsentimentsurroundingtheirfirmsonadailybasis(seeBox4inAnnex2formoredetailsonthistool.Inaddition,afewauthoritieshaveincorporatedsentimentanalysisforotheruses.TheBdI’sexperimentalcorporategovernanceanalysistoolappliessentimentanalysistogaugethetoneofinterventionbyindividualboarddirectors.TheMonetaryAuthorityofSingapore(MAS)isdevelopinganintegratedsurveillanceplatformthatcollatesdatafromvarioussources(egnews,financialstatements,macroeconomicindicators,regulatoryreports)andappliesvariousMLtechniques,includingsentimentanalysis,tofacilitatein-depthanalysisandriskidentification.Severalotherauthoritiesareplanningtodevelopsentimentanalysistoolsorincorporatesentimentanalysisintofutureationsoftheirtextanalysistools定量數(shù)據(jù)分析工具風(fēng)險(xiǎn)識(shí)別工具主要通過(guò)使用定量和/或結(jié)構(gòu)化數(shù)據(jù)來(lái)幫助發(fā)現(xiàn)金B(yǎng)預(yù)期損失可能無(wú)法被監(jiān)管實(shí)體充分識(shí)別的客戶(hù)(有關(guān)此工具的更多詳何演變的預(yù)測(cè)來(lái)預(yù)測(cè)銀行的監(jiān)管類(lèi)別，并估計(jì)(使用ML)小銀行的風(fēng)險(xiǎn)加權(quán)資產(chǎn)不再需要提交此類(lèi)報(bào)告。上述MAS綜合監(jiān)控平臺(tái)還旨在根據(jù)各種數(shù)據(jù)(包括定量和定性數(shù)據(jù))識(shí)別金融機(jī)構(gòu)的風(fēng)險(xiǎn)。奧地利共和國(guó)中央銀行(OeNB)擁有通過(guò)考慮盈利能力、資本充足率和各種風(fēng)險(xiǎn)(包括信用、市場(chǎng)、運(yùn)營(yíng)、流動(dòng)性和資金風(fēng)險(xiǎn))來(lái)識(shí)別高風(fēng)BRiskidentificationtoolshelptospotrisksatfinancialinstitutionsbyusingmainlyquantitativeand/orstructureddata.Capital,creditandliquidityaresomeoftheriskareastargeted.TheBCB’sADAMtoolisanML-basedapplicationlookingforcustomerswithhighprobabilityofdefaultandwhoseexpectedlossmaynotbeadequatelyrecognisedbysupervisedentities(seeBox5inAnnex2formoredetailsonthistool).TheSwissFinancialMarketSupervisoryAuthority(FINMA)hasdevelopedorisdevelopingseveraltoolsfordifferentuses,suchasforecastingsupervisorycategoriesofbanksbasedonpredictionsofhowtheirriskswouldevolveandestimating(usingML)risk-weightedassetsofsmallbanksthatarenolongerrequiredtosubmitsuchreports.9TheMASintegratedsurveillanceplatformmentionedabovealsoaimstoidentifyrisksatfinancialinstitutionsbasedonvariousdata,includingquantitativeaswellasqualitativedata.10TheCentralBankoftheRepublicofAustria(OeNB)hasatoolthatidentifieshigh-riskbanksbyconsideringprofitability,capitaladequacyandvariousrisks,includingcredit,market,operational,liquidityandfundingrisks.TheNetherlandsBank(DNB)hasanexperimentaltoolthatcombinesmonthlyregulatoryreportingdatawithdailypaymentsystemsdatatoestimateadailyproxyoftheliquidityriskratioofsupervisedinstitutions.TheQFCRAisdevelopingariskscorertoolthatwillprovideanindependentviewofbanks’risksandchallengessupervisoryteamstoarriveatconsistentscoresintheirinternalsupervisoryratingprocess.定性和定量數(shù)據(jù)工具一E驗(yàn)工具可自動(dòng)識(shí)別來(lái)自各種結(jié)構(gòu)化和非結(jié)構(gòu)化數(shù)據(jù)源的人員之間的絡(luò)分析如何評(píng)估當(dāng)前關(guān)系不僅可以預(yù)見(jiàn)當(dāng)前而且還可以預(yù)見(jiàn)新風(fēng)險(xiǎn)Networkanalysistoolslookatrelationshipsbetweenentitiestobetterunderstandhowriskscascadefromoneentitytotheother.Networkanalysistoolsdrawonarangeofquantitativeandqualitativedataandmethods,fromneuralanalysistopatternrecognition.TheBdE’stoolanalysestherelationshipsbetweenentities,identifyingnotonlyformalrelationshipsbutalsolessformalconnectionsthatwouldbedifficultorimpossibleforsupervisorstofindmanually.ThisallowstheBdEtoevaluatetheimpactofagivenriskacrossthewholenetwork(seeBox6inAnnex2formoredetailsonthistool).FINMA’sexperimentaltoolautomatestheidentificationoflinksbetweenpersonsfromvariousstructuredandunstructureddatasources,withthepotentialtobeusedinarangeofprudentialusecases.Oneareaofworkgoingforwardmaybehownetworkanalysisassessesthelikelihoodthatcurrentrelationshipsmayforeseenotjustcurrentbutalsoemergingriskconcerns.于識(shí)別可能未被注意的公司之間的相似性(有關(guān)FINMA和OeNB工Toolsthatenhancepeergroupidentification–whichsetsthefoundationforoff-sitemonitoring–aimtorecognisesimilaritiesbetweensupervisedentities.Themethodsmightbesimilartonetworkanalysisbuttheobjectivesaredifferent;hence,theyareseparatelyclassified.Thesetoolsallowforaconsistentsupervisoryapproachtobankswithsimilarriskprofiles.FINMA’sexperimentalpeergroupdeterminationtoolchallengestheexistingrules-basedpeergroupsystembyapplyingunsupervisedclusteringalgorithmstokeyvariablesuponwhichfirmsarecategorised.Similarly,theQFCRAisdevelopingatoolthatwillseparatelyanalysefirms’assetsandliabilitiesinsteadoflookingatthewholebalancesheetinidentifyingpeergroups.Indoingso,itidentifiescloselyrelatedinvestmentandfundingstrategies.TheOeNBisalsodevelopingatoolthatwillidentifysimilaritiesbetweenfirmsthatmayhavegoneunnoticed(seeBox7inAnnex2formoredetailsontheFINMAandOeNBtools).素，期間自動(dòng)進(jìn)行數(shù)據(jù)分析(例如，用于信用風(fēng)險(xiǎn)管理審查)。Inspectionautomationtoolsareintendedtomakethemulti-steprocesssimplerandmoreefficientThetoolsinthiscategoryareanespeciallyvariedgroup,withsomecoveringonestepoftheexaminationandother