學(xué)院數(shù)據(jù)庫(kù)介紹實(shí)證研究經(jīng)驗(yàn)講座相關(guān)

關(guān)于實(shí)證研究的一些個(gè)人經(jīng)驗(yàn)?zāi)祥_大學(xué)金融學(xué)院趙博2018-10內(nèi)容1.計(jì)量研究的大致框架實(shí)證研究的相同性社會(huì)科學(xué)實(shí)證研究的特殊性2.數(shù)據(jù)的分類以及常見的數(shù)據(jù)庫(kù)個(gè)人行為方面商業(yè)行為方面“新”數(shù)據(jù)3.如何組織實(shí)證研究重復(fù)重復(fù)再重復(fù)常見的工具1.計(jì)量研究的大致框架實(shí)證研究的相同性社會(huì)科學(xué)實(shí)證研究的特殊性顯著？不顯著？實(shí)證研究的共同特征和數(shù)據(jù)打交道通用技能：數(shù)學(xué)、統(tǒng)計(jì)學(xué)、計(jì)算機(jī)特殊技能：各領(lǐng)域的專業(yè)知識(shí)（fieldknowledge）實(shí)證指導(dǎo)理論？理論指導(dǎo)實(shí)證？“InGodwetrust,allothersbringdata”社會(huì)科學(xué)實(shí)證研究的特殊性可預(yù)測(cè)性(可證偽性)+奧卡姆剃刀原則=科學(xué)的標(biāo)準(zhǔn)實(shí)驗(yàn)環(huán)境的缺乏=>可重復(fù)性的缺乏=>預(yù)測(cè)很難=>可證偽性難Ifyoutorturedataenough,finallyitwillconfessButatheoryisnotlikeanairlineorbustimetable.Weare

notinterestedsimplyintheaccuracyofitspredictions.Atheoryalsoservesasabaseforthinking.Ithelpsustounderstandwhatisgoingonbyenablingustoorganizeourthoughts.Facedwithachoicebetweenatheorywhichpredictswellbutgivesuslittleinsightintohowthesystemworksandonewhichgivesusthisinsightbutpredictsbadly,Iwouldchoosethelatter,andIaminclinedtothinkthatmosteconomistswoulddothesame.

—RonaldCoase判斷標(biāo)準(zhǔn)是？顯著性：“來制星星星星星星的你你你”Harvey(2017),PresidentialAddress:TheScienti?cOutlookinFinancialEconomicsunreportedtest,lackofadjustmentformultipletests,direct&indirectp-hacking…manyresultsbeingpublishedwillFAILtoHOLDUPinthefutureDewald,W.,Thursby,J.,&Anderson,R.(1986).ReplicationinEmpiricalEconomics:TheJournalofMoney,CreditandBankingProject.TheAmericanEconomicReview,76(4),587-603.inadvertenterrorsinpublishedempiricalarticlesareacommonplaceratherthanarareoccurrence一些期刊歡迎報(bào)告不顯著的結(jié)果,ReviewofFinance,JournalofMonetaryEconomics(?)總結(jié)：難怎么辦：看別人怎么做的數(shù)據(jù)的分類DatafromIndividualActivitySocialMediaNewsandReviewsWebSearches,PersonalDataDatafromBusinessProcessesTransactionDataCorporateDataAggregateDataDatafromDetectorsSatellitesGeolocationOtherSensors……DatafromIndividualActivitySocialmediaStockTwits,,twitterforstocksSumZero,,hedgefundprofessionalsiSentium,,timeseriesofsentiment雪球，股吧News,ReviewsYipitData,,publiclyavailableinformationoncompanyandgovernmentwebsitesAccern,,analyzesover300millionpublicnewswebsites,blogs,socialmedia,publicfinancialdocuments(SECfilings).RavenPack,CNRDS,2001年以來近240萬篇與A股上市公司相關(guān)的中文新聞Datago,1998年以來1154家媒體近1500萬篇財(cái)經(jīng)新聞Websearches,personaldataGoogletrends,BaidutrendsRakutenIntelligence,,customerpurchasingdata…DatafrombusinessprocessesTransactiondataOthercommercialtransactiondata:Quandl(CorporateAviationdata),FactsetRevere(supplychainrelationships,companies’interrelationships,etc.)Credit/debitcard,bank/mortgagedata:Superfly,Yodlee,SecondMeasureOrderbookandFlow:TickData,QuantConnect,WindCorporatedataBloomberg:Consumerdiscretionary:Crain(carinventory),SNLKagan(revenue,cashflowofCableandHD,broadcast,Telco,etc.)Bloomberg:Consumerstable;Energy,utilities;Financials;HealthCare;etc.Macroeconomic:OECDData,FRED,etc.Accountingdata:EDGAR,IRS990filingsonAWS,etc.Venturecapital:ThomsonONEBanker,清科GovernmentAgenciesDataGlobalOpenDataIndex,;China,;US,;HK,.hk;WorldBank,;IMFDatadatafromdetectorsorsensorsSatellitesSatellitesOrbitalInsights,;SkyWatch,;Rezatec;Genscape;etc.Free:NASAReverb,USGSEarthExplorer（美國(guó)內(nèi)政部）,ESA’sSentinelMission(歐洲空間局)GeolocationAirSage,Sense360(consumerlocationandmovementdatabasedonmobiledevice)DataAggregatorsWRDS,CSMAR,Bloomberg,Quandl,ThomsonReuters,Moody’sAnalytics,AWS高效、有序地組織實(shí)證研究做研究的過程：有想法=>收集數(shù)據(jù)=>分析=>得到結(jié)論=>寫作重復(fù)多次是概率為1的事件要做好documentation。(額外花費(fèi)1倍甚至更多的時(shí)間)不是給別人看的；是給你自己看的文件夾命名，多個(gè)版本，原始數(shù)據(jù)和清洗過的數(shù)據(jù)，程序的可移植性（也許僅僅是從一個(gè)文件夾移到另一個(gè)文件夾）“Manycouldnotlocatethedataforthearticle,whileothershadlosttheirprogramming.Evenwhentheprogramsanddatacouldbelocated,authorsoftenhadnotkeptacontemporaneousrecordoftheprogressoftheresearchandcouldnotreconstructtheirresult”

—Dewald,W.,Thursby,J.,&Anderson,R.(1986)版本控制：GitHub如果多人合作，并且工作之間的銜接性很強(qiáng)，應(yīng)該使用。文件夾命名：簡(jiǎn)單，英文命名用README.md描述工作內(nèi)容所有互相關(guān)聯(lián)的文件應(yīng)該放在同一個(gè)文件夾內(nèi)。文件運(yùn)行如有先后順序，可以在文件名之前加編號(hào)。例如，00-preproc.txt,10-panel.do,11-dynamic-panel.do,etc.原始數(shù)據(jù)應(yīng)原樣保存。處理過的數(shù)據(jù)分開放置。不要用絕對(duì)路徑或symlink。用相對(duì)路徑。這樣在轉(zhuǎn)移文件夾時(shí)，內(nèi)部的程序依然可以正常使用。例如：./data/，