CFA二级前导班培训项目：框架介绍_道德+数量+经济+固收+组合.docx

CFA二级培训项目讲师：Vincent竺5Mg*IX新虫理Vincent工作职称工金程教育密深培训师-教胃背景工英国埃塞克斯大学金融学硕士、通过CFA三锣、PMP（ProjectManageBentPTofessional项目管理专业认证）-工作IIfh曾任某外资银行总部项目经理，十二年的外企银行工作责历，积K了丰富的金触实战经验.现为金程敦曾CFA赛深培训讲I,担任CFAM目教学产品研发负责人.熟悉CFA考试重点，擅长授课包括职业伦理、经济学、固定收益、数St分折、组合管ML授课逻辑清断易If,结合实际案例深入浅出解择考点，备受学员欢迎.2-106服务客户t中国银行.中Si建设银行、肮州联分银行、机州银行、国泰君安证券、苏州元禾控股等有理囹RrlBllSTopicWeightingsinCFALevelIIISessionNO*ContentWeightingsIStudySession1-2Ethics&ProfessionalStandardS10-15StudySession3QuantitativeMethods5-10StudySesston4EconomicsS-IOStudySession5-6FinancialReportingandAnalysis10-15StudySession7-8CorporateFinance5-10StudySession9-11EquityValuationIO-15StudySession12-13FixedIncome10-15StudySession14Derivatives5-10StudySession15AlternativeInvestmentsS-10StudySession16-17PortfolioManagement10-153-106百亚医顺71B二级学习方法>一级二Si各科目权比较科目一级权重二级权重Ethics&ProfessionalStandards1510-15QuantitativeMethods105-10Economics105-lFinancialStatementAnalysis15IO-ISCorporateFinance10510EquityValuation610-15FixedIncome1110-15derivatives115-10AlternativeInvestments65-10PortfolioManagement610-154106互理囹新ItflEthics&ProfessionalStandards-tB三FrameworkASSlEthkalandProfessionalStandardsRlCcxleofEthicsandStandardsofProfessionalConduct/2primaryprinciples/Proceedings/Panel/AMCvsCodeandstandards/Codeofethics'SlWfttuFramework>SSlEthicalandProfessionalStandardsR2GuidanceforStandardsI-V11/1:Professionalism/11:Integrityofcapitalmarkets/III:Dutiestoclients/IV:Dutiestoemployers/V:Investmentanalysis,recommendations,andactions/VI:Conflictsofinterest7-106/V11:ResponsibilitiesasaCFAInstitutememberorCFAcandidate* Guidance for Standards IL ProfssionaHSm 卜吨宙IIKInKnowledgeofthlaw/了修与工作相关的注博，法ML准则和济会COdeandStanda通:/避守最严格法律:/不得被意参与il法Ir为：/育M向咨询上司/合燃.不需要向我附福门举H除参法律明文嵯定：/发现HiMl行为.自己不作为.初为忌嫌：Ind*p«nd«nceandobjectivity/直分礼物.基金经号的客户（/£）给的是小费.broker给基金经理或者上市公司给aru9tt都是要影响客脱独立性的不能收除等是modestgrfb/基金公司（买方）俺用分析师（卖方）I/防火应停RI离投行部与其他所有离门.q试特别修与研亢部的1«高：/Issuerpaidresearch11Ifj.也必须收fUfee.井进bit露：/差*费自己出.除葬IQ曼工具无法到达的情况.可接受目标公司的一般安K：Misrepresentation/没有及时改正的打字博误：/不能胡说八it（美于个人宠质、关于公司服务前困、模里结果没有交代清整、业缄度和日因PaCs,比如不可以ChEyPdmg、不色捐像不该拒保的收益、使用外品修命经理要披露，业HJbecchEark出界不备当/不能方4L必列恰当引用：Misconduct/欺在盛日等不信行为：/喝酒a箫酒：/任何伪者0如9信.信Ir工作展任能力的行为逢反m&:OndUul/与工作无美的个人信仰.政治修向的，议.不违反mcoCdUG/善欺作IMk的个人破产可以免贵.档案期内要披露.三W-»*IlflGuidanceforStandardsIIMaterialonpublicrnfbrmation大必须同时具3消息来淹可靠.以及对股价有重大影响（举例参见崖件）j2.Integrityofcapital/非公开，向:eZave对象披露不属于公开：marketsl'Mos理论*j/做市商掌握了MM不可以停止做市.应该做消极对手方：/从事无风险套利交易时.若获得了MNI,除非公司有能力证明1.流程和记录短范才可以继续交易.否则要停止,Marketmanipulatx>c/两片形式（关位看动机）：敢布Iy消息、基于交晶扭曲置价：/以避税为目的的交易.不违反市场爆ttU/基于特定的交易策略.不通反市场操纵,/如果为了堵加流动性，期货交易所会员，书面协议.并对外事先披露，则不住违埋.等业色*!窗GuidanceforStandardsIII13.DutyToClnt-1.oyalty,prudncndcare/Fidgary需要履行extrjcare,higherstandard:/四美客户*individual,beneficiarymandate,investingPUbhe:/基于第合整体进行判Vh/Softdollar(softcommission)JK.brokerage是客户的M产必甯100%让X户直接受标/Directbrokerage的情况F仍然宵义务寻找bestexecution和bestprice:/VOGnggXy必务1进行.除尊批于性价比号虐.要向客户IfvotmgProxyPOhCles：FairdMing/Falr井TXqUah可'以有PfemlUmlev，lserv>c，但必渠械露.同时不仿害IC他客户利益：保客户有公允的机会时投责推荐出反应：/聚取round-IOtbass.避免odd-lotdtstnbutions：/family-memberaccounts井非beneficiary要与真ItX户一祝同仁：SuitebiHty/基于客户的IPS(RRTTLLU)，至少每年更新，投责重大交更必缪先修改IPS:/0燎分散化.基于组合整体角度，虐、/宫户囚执己见的文口.如果对整体彭看不大可以售闷.彭大0牙他改IPS.客户不同意修IMpSIt从管理联户中IM育贯金交让客户自己管理；PerformanceprMntation/过去业续不能"示在未来。I以达成I/业飨而报可以.但必争同以后续提供洋H信总：/IH合怛：E杈平均.终止的组合.所有相IU凤格怛令.相关支埼记承保存：Preservationofconfidentiality保定t对NFSIi去X户、旗在宫户、俞格滔在N户I是否保密先必须专虐法0缴定.10-106有曲倒Hft“GuidanceforStandardsIV4DutyTbemployers1.oyalty/IndePendemPQCtke是指与雇主业务在内容匕时间匕精力上相竟争的业务.必须告知雇主性质、expectedduration以及COmPenSatiOn并且得到许可：开雇主前不拿雇主一针一线正式离职前不可以先开始拉原客户:/仅仅知道几个客户是可以的.但不可以背诵客户名单：Additionalcompensationarrangements各方书面同意；，告知CoEpensatiOll的性质、大概金桢、duration：Responsibilityofsupervisors-下属违邂上Ret三反了主管的职责，除非表明已没充分尽责:/可以将工作指派其他人负贵，但最终后果自负：-在接受短导岗位之前.必须磷保公司有充分的合爆程序.如果不合!一定要提出改进措施.公可改正好接受岗位:/如果发现违规.必率立刻行动起来沏底调直，限制当事人工作或者加强监管.11-106aw*tnGuidanceforStandardsV5.InvestmentDiligenceandreasonablebasis-耳分析报告联君做投资推荐.必须分析宏费姓济.行业、公司基本面等全部因素：/保第三方信息勤勉尽贵(四个方面)：-选择外部*何要勤勉尽费(四个方面)：/如果是使用量化模型进行推介.必级MH.包括入变量.假设前提、局限性好：/崎发量化模M.需要了解偿型的方方面面，需要对模型进行焉试：集体报告如果不同意结论.但过程产i三仍然可以胃名：/HotiSSg没有尽IhCommunicationwithclients/区分事实与观点：/投资流程中的大变更要及时告知投资者.如模型.投资决策流程、投资范国.投贵限制.投费策略.关0人员改变等I/推吞可以是capsuleform,但只要投负者要.必须给出详细版*;-主要风险与局限性:Recordretention/城质版电子版保存皆口I:/IW公司的record未经批准不籍带走.如果没有带走支料数据.在新公司不能发布旧公司业绩或研报，除非通过公开信息重建:/遵守当地法律定，当地没有规定的，俗金建议保存7年.M业色斯【BQ、，GuidanceforStandardsVIDisclosureofconflicts/利益冲突.指潜在可能伤害客户哦者投资公众的.主要不叶对雇主1/必须事先平实的遏育告知8/个人交易持仓：投资标的公司担任童事：与投资标的公司有业务关系(做市商、企业融费等):分析川t与投行部之间，市场部与分析棘：重大个人关系：/如果奖金激励与客户利益有冲突必绩披露：-Priorityoftransaction/Client>employer>individual(beneficialowner).间RI时6.Conflictsofinterest;间不能太短.十天半月才合适:/区分fam"yaccounttjbeneficialOWner的区别；/个人交易要申报获得批准才可吸进行：Referralfees/事先披露.方使客户判断介绍是否客观.以及服务的真实成本*/最少每季度向雇主披露介绍费的性质及金H.13-106MIulaiKlGuidanceforStandardsV11Conductasmembersandcandidates，遵守考试纪律:，不得泄富考试内容和范围.但可以发表观点：/不得利用与孙会的联系为个人或公司牟取私利IReferencetoCFAinstitute,designation/正确使用CFA标志：/不可以表示因为CFA而比其他人更优秀，更胜任工作.投资业绩更好，7.Responsibilityasmembers14-106K出却曰QuantitativeMethodsFrameworkCalculateandinterpretSEEandR-squaredCalculatethepredictedVaIUeforthedependentvariableDescribeIimitatioinsofregressionanalysis161(½Framework202ICF/ERM持续更新+微信：xbajun888sDummyvariables17-106FrameworkSS2QuantitativeMethodsR6Time-seriesAnalysisTrendModelsQineartrendandlog-lineartrend)AutoregressiveMOdelS(AR)+Calculateforecasts+Noautocorrelation+RegressionWithmorethanonetimeseriesFrameworkASS3QuantitativeMethodsR7MachineLearning/OverviewofMachineLearning/SupervisedMachineLearningPenalizedRegressionSupportVectorMachineK-nearestNeighborClassificationandRegressionTreeRandomForest/UnsupervisedMachineLearningPrincipalComponentsAnalysisClustering/Neuralnetworks,deeplearningnets,andreinforcementlearning)19-106*Framework>SS3QuantitativeMethodsR8BigDataProjects/BigDataIntroduction/StructuredDataAnalysisCceptualizationofthemodelingtaskDatacollectionDatapreparationandwranglingDataexplorationModeltraining/UnstructuredDataAnalysisTextproblemformulationData(text)curationTextpreparationandwranglingTextexploration20-106Modeltraining三WQiff-IBfl*Framework>SS3QuantitativeMethodsR9ExcerptIromwProbabiIisticApproaches:ScenarioAnaIysisfDecisionTrees4andSimulations"/Simulation/ComparingteApproachesMW舞！*ReviewandExploringAReviewofLevelItopks1.HypothesistestingQ/Pvalue2.Type1error/TypeII(?rror>ExploringLevel11topics1.TypesofMachineLearning2.MLchallenge-Overfitting3.SupervisedML:K-NearestNeighbor4.BigData-StructuredDataAnalysis5.BigData-UnstructuredDataAnalysis号曲囹新t"HypothesisTestingabouttheRegressionCoefficient>RegressioncoefficientconfidenceintervalB1±(tfsiJClnotincludethehypothesizedvalue,rejectASignificancetestforaregressioncoefficientH0:b1=0.leststatistics:,=一hypOtheSEedVame办力=”_?Decisionrule:rejectH0if+tcritical<tort<-1criticalRejectionofthenullmeansthattheslopecoefficientisdifferentfromthehypothesizedvalueofb1>p-value:thesmallestsignificancelevelforwhichthenullhypothesiscanberejectedRejectH0ifp-value<FailtorejectH0ifp-value>23106aSffimuHypothesisTesting>TypeIerrorandType11errorTypeIerror:拒真.rejectthenullhypothesiswhenit'sactuallytrue/Significancelevel():theprobabilityofmakingaTypeIerror/Significancelevel=PfTypeIerror)TypeUerror:取伪，failtorejectthenullhypothesiswhenitsactuallyfalse/Powerofatest:theprobabilityofcorrectlyrejectingthenullhypothesiswhenitisfalse/Powerofatest=l-P(Type11error)M业SMrrlBIlHypothesisTestingH0isactuallytrueH0isactuallyfalseDonotrejectH0CorrectTypeHerrorRejectHP(TypeIerror)thesignificancelevelCorreetPoweroftest=1-P(TypeIIerror)>Withotherconditionsunchanged,eithererrorprobabilityarisesatthecostoftheothererrorprobabilitydecreasing.>Howtoreducebotherrors?IncreasetheSampleSize.25-106，1.TypesofMachineLearning>Machinelearningisbroadlydividedintothreedistinctclassesoftechniques:supervisedlearning,unsupervisedlearning,anddeeplearning.SupervisedlearninguseslabeledtrainingdatatoguidetheMLprogramtowardsuperiorforecastingaccuracy./LabeleddatasetonethatcontainsmatchedsetsofobservedinputsandtheassociatedoutputInunsupervisedlearning,theMLprogramisnotgivenlabeledtrainingdata;instead,inputs(i.evfeatures)areprovidedwithoutanycondusisaboutthoseinputs.26-106/Thealgorithmseekstodiscoverstructurewithinthedatathemselves.1W-»«Ilfl2.MLchallenge-Overfitting>OverfittingisanissuewithsupervisedMLthatresultswhenalargenumberoffeaturesareincludedinthedatasample,resultingthatthefittedalgorithmdoesfitWRlltoHaininQdatabutDOtQRneIaliZRWRllIonewdm.ItresultsininaccuracyforecastsonoutofSamPledata1randomnessismisperceivedtobeapattern/Whenamodelgeneralizeswell4itmeansthatthemodelretainsitsRXDlanatoryDOWerwhenitisappliedtonew(Le.,t-of-sample)data.3.SupervisedMLK-NearestNeighborAK-nearestneighbor(KNN).Moreccxnmonlyusedinclassification(butsometimesinregression),thistechniqueisusedtoclassifyanewobservationbyfindingsimilarities(*nearness,f)betweenthisnewobservationandtetrainingsample. KNN VWHt Ne Ohemtionf KalB. KNN WiUt New Obsmaiiour Kb5X28106w - - tra K-Nearest Neighbor> Two vital concerns The researcher specifies the value of k the hyper parameter; triggering the algorithm to lk for the k observations in te sample that are closest to the new observation that is being classified. If k is t small wil result in a high error rate,' if it is too large, it will dilute the result by averaging across too many outcomes./ H k is even, there Ciay be ties, with no clear winner. Analysts need to have d clear understanding of the data and underlying business to define ,similar" (Or near).29-106u< aw - tra 4. Big Data - Structured Data AnalysisA To illustrate the steps invclved in analyzing data for financial forecasting (traditional ML model), We will use an example of a consumer credit scoring model in the following five steps:L Conceptualization of the modeling task2. Data collection3 Data preparation and wrangling4. Data exploration5. Model training，StructuredDataAnalysisAStep3:DatapreparationandwranglingThisstepinvolvescleaningthedatasetandDreparinqitforthemodel.Datapreparation(Cleaning)istheprocessofexamining,identifying,andmitigatingeorsinrawdata,includesaddressinganymissingvaluesorverificationofanyout-of-rangevalues.31-106DataWrangling(Preprocessing)datamayperformstransformationsandcriticalprocessingstepsonthecleanseddatatomakethedatareadyforMLmodeltraining,involvingaggregating,filtering,orextractingrelevantvariables吨总却MH，StructuredDataPreparation(Cleansing)>Thepossibleerrorsinarawdatasetincludethefollowing:IncompletenesserroriswherethedataarenotpresentresultinginEiSEingdata./Themostcommonimputationsaremean,median,ormodeofthevariableorsimplyassumingzero.Invalidityerroriswherethedataare(XJtSideOfameaningfulrange.resultingininvaliddata./Thiscanbecorrectedbyverifyingotheradministrativedatarecords.InaccuracyerroriswherethedataarenotJmRasurcOftruevalue.32-106Thiscanberectifiedwiththehelpofbusinessrecordsandadministrators.1U!-QllffItflStructuredDataPreparation(Cleansing)>Thepossibleerrorsinarawdatasetincludethefollowing:Inconsistencyerroriswherethedataconflictwiththecorrespondingdatapointsorreality./Thiscontradictionshouldbeeliminatedbyclarifyingwrthathersource.Non-uniformityerroriswherethedataarenotPleSentinUnidenticalformat/Thiscanberesolvedbyconvertingthedatapointsintoapreferablestandardformat.DuplicationerroriswheredDicateObSeC加ionsareDreSenl/Thiscanbecorrectedbyremovingtheduplicateentries.DataPreparation(Cleansing)Invalictity errorInconsistency errorDnfLr of Rirth1 5/1 力 975切 1 M 942SalaryInCC me方bStateCredi t Card,V41 9RD$60 50 0nd'p Cteness er rDp,'Ton error34-i062021GEA&ERM持续更新A new variable can be created from the current variableFiltration: The data rows that are not needed for the projectm ustbeidentifiedandfiltefed357DataWranglinlJ:TransformationADatabeforetransfermaltionDataWrangling:Scaling>ScalingisaprocessOfadjustingtherangeofafeaturebyshiftingandchangingthescaleofdata.>Herearetwoofthemostcommonwaysofscaling:Normalizationistheprocessofrescalingnumericvariablesintherangeof0.1,i(normAlixed)二3AnliHC一in/sensitivetooutliers,sotreatmentofoutliersisnecessarybeforenormalizationisperformed./ used when the distribution of the data is not known.lesssensitivetooutliersasitdependsonthemeanandstandarddeviationofthedata.37-106/Thedatamustbenormallydistributedtousestandardization.aw-aw!B«5.UnstructuredDataAnalysis>Unstructured,texted-baseddataismoresuitableforhumanuse.Thefivestepsinvolvedneedtobemodified(thefirstfour)inordertoanalyzeunstructured,text-baseddata:1.Textproblemformulation.Theanalystwilldeterminetheproblemandidentifytheexactinputsandoutputofthemodel.2.Datacollection(curation).Thisisdeterminingthesrcesofdatatobeused(e.g.,webscouring,specificsocialmediasites).3.Textpreparationandwrangling.Thisrequirespreprocessingtbestream(s)ofunstructureddatatomakeitusablebytraditionalstructuredmodelingmethods./UnStRKtWeddatacanbeintheformolteximages1videos,andaudiofiles.4.TextexplorationThisinvolvestestvisualizationaswellastextfeatureselectionandengineering.5.Modeltraining38-106aw-st-TextPreparation(Cleaning)>Unstructureddatacanbeintheformoftext,images,videos,andaudiofiles.ForanalysisandusetotraintheMLmodel,theunstructureddatamustbetransformedintostructureddata.ForExampleRoht Ajv 5SampleTextfromRobotsAreUSHomePageUVeryhomeandbu>inegSbaUldhavearb<M.RawTextfromtheSource<hlclss=*text-leftmb3*,>RobotsAreUa<hl><h2>Everyhucneandbuinesshoudlhavearobot<h2>UnstructuredTextPreparation(Cleaning)ATextcleansinginvolvesthefollowingsteps:1.RemoveHTMLtags.TextcollectedfromwebpageshasembeddedHTMLtags,whichmayneedtoberemovedbeforeprocessing.2.Removepunctuations.Textanalysisusuallydoesnotneedpunctuations,sotheseneedtoberemovedaswell.Somepunctuations(e.g.,%,$)maybeneededforanalysis,andifso,theyarereplacedwithannotations(i,