Upload
vuhuong
View
217
Download
1
Embed Size (px)
Citation preview
Creating a Repository of Economic Models
For Research and Education
An Interactive Qualifying Project (IQP) Submitted to the Faculty of
WORCESTER POLYTECHNIC INSTITUTE
In Partial Fulfillment of the Requirement for the
Degree of Bachelor of Science
Submitted by: Nicholas Fajardo Mitchell Farren
Synella Gonzales Mark Karpukhin
Artemii Maksimenko Yaofeng Wang
11 October, 2017
Submitted to:
Project Advisor
Svetlana Nikitina
Project Sponsor The Financial University under the Government of the Russian Federation
On-Site Liaison
Professor Anton Losev Department of Data Analysis, Decision-Making Theory and Financial Technology
i
Abstract TheFinancialUniversityundertheGovernmentoftheRussianFederationhas
expressedinterestindevelopinganeducationaltooltohelpstudentsandresearchersaccesseconomicmodels.Toreachthisgoal,weconductedextensiveresearchonavarietyofmodels,administeredasurveytoseepotentialinterestinaneconomicrepositoryanditscomponents,codedresultsintoarepository,anddevelopedaWikitofostereducationalapplicationsofarepository.Asaresult,wecompiledaversionofarepositorywhichcouldserveasaframeworkforfuturedevelopmentofeducationalandresearchapplications.
ii
Acknowledgements Wewouldliketothankthefollowingpeoplefortheirhelpandguidancethroughoutthisproject: ProfessorAntonLosevforalltheguidance,feedback,andhelphehasprovidedus.Hehasgivenusanexperiencethatweallwillremember.Betweenworkingwithusonourproject,providingresources,showinguslocalculture,andremainingonstandbytoansweranyquestionswemighthavehad,ProfessorAntonLosevgaveusmorethanwecouldhaveeverhoped.Withouthim,nothingwedidwouldhavebeenpossible. ProfessorSvetlanaNikitinaforbeingthebestpossibleprojectadvisorwecouldhavehopedfor.Shesharedhertime,dedication,valuableinput,andlanguageandwritingskillstous,whichhelpedourprojectreachitsmaximumpotential. SvetlanaChugunovaforherhospitalityattheFinancialUniversityandshowingusRussianculture,makingourstayinRussiaonetoremember. ZalyaNartokovaforheradministrativehelpgettingtheprojectofftheground,andforshowingusaroundthecityofMoscow. ProfessorOlegPavlov,ProfessorElkeRundensteiner,andProfessorKhalidSaeedfortakingtimetoletusinterviewthemandgiveusafoundationalunderstandingofourprojectbeforewearrivedinRussia.Becauseofthem,wewereabletohitthegroundrunningoncewearrived.
iii
Authorship
Section PrimaryAuthor Editor(s)
ExecutiveSummary NicholasFajardo All
Introduction YaofengWang SynellaGonzales
Background SynellaGonzales NicholasFajardo,MitchellFarren,YaofengWang
Methodology NicholasFajardo MitchellFarren,SynellaGonzales,YaofengWang,
ResultsandDiscussion
SynellaGonzales
NicholasFajardo,MitchellFarren,YaofengWang
ConclusionandRecommendations
MitchellFarren NicholasFajardo,SynellaGonzales,YaofengWang
References YaofengWang SynellaGonzales
AppendixA-SponsorDescription
NicholasFajardo MitchellFarren,SynellaGonzales,YaofengWang,
AppendixB-InterviewProtocol
NicholasFajardo,MitchellFarren,SynellaGonzales,YaofengWang
All
AppendixC-DashboardQuestionnaire
NicholasFajardo,MitchellFarren,SynellaGonzales,MarkKarpukhin,ArtemiiMaksimenko,YaofengWang
All
AppendixD-ModelDescriptions
NicholasFajardo,MitchellFarren,SynellaGonzales,MarkKarpukhin,ArtemiiMaksimenko,YaofengWang
All
iv
Table of Contents Abstract.............................................................................................................................................iAcknowledgements......................................................................................................................iiAuthorship......................................................................................................................................iiiTable of Figures.............................................................................................................................vTable of Tables..............................................................................................................................viExecutive Summary....................................................................................................................vii1. Frontiers in Russian Economics Studies..........................................................................12. Economic Models: Scope, Significance, and Applications.........................................3The Russian economy’s unique path and need for better modeling .................................... 3 How does education and research support economic growth? ............................................ 5 What are the current methods of economic study? ................................................................ 5 What are economic models? ..................................................................................................... 5 Why are models important? ...................................................................................................... 6 How will these models be accessed by university staff and students? ............................... 6 What are current methods used to study and analyze economic models? ......................... 7 3. Methodology: Program Fabrication....................................................................................8Economic Model Research ........................................................................................................ 8 Programming the models .......................................................................................................... 9 Organizing and storing the models ........................................................................................ 10 Front-end framework ................................................................................................................ 10 Survey Development ................................................................................................................ 11 Third Party Tools and Packages ............................................................................................. 11 4. Results and Discussion........................................................................................................13Survey Results ......................................................................................................................... 13 Creating the framework ........................................................................................................... 14 Study Limitations ..................................................................................................................... 15 5. Conclusion and Recommendations..................................................................................17Recommendations ................................................................................................................... 17 References....................................................................................................................................18Appendices...................................................................................................................................21Appendix A: Sponsor Information .......................................................................................... 21 Appendix B: Interview Protocol .............................................................................................. 23 Appendix C: Dashboard Questionnaire ................................................................................. 25 Appendix D: Model Descriptions ............................................................................................ 26
World Trade Model............................................................................................................................26Cobb-Douglas Production Function Model................................................................................29Portfolio Model..................................................................................................................................30Scoring Model....................................................................................................................................32
v
Table of Figures Figure1:TimespentbyEasternEuropeancountriesinatransitionalrecessionafterthefall
oftheSovietUnion(inyears). .............................................................................................. 3Figure2:Russia'scurrentGrossDomesticProduct(GDP) ........................................................ 4Figure3:Russia'scurrentGDPgrowth. ....................................................................................... 4Figure4:Mechanicsofarepository ............................................................................................. 7Figure5:Surveyquestions ......................................................................................................... 11Figure 6: Survey results – gauging the audience .......................................................................... 13Figure7:Surveyresults–decidingwhattoincludeintherepository ................................... 14Figure8:CurrentimplementationoftheWiki ......................................................................... 15Figure9:FinancialUniversityDepartmentStructure .............................................................. 21Figure10:CLIInterfaceforWorldTradeModel ...................................................................... 26Figure11:Calledfunctionscorrespondingtopossiblequeries .............................................. 27Figure12:Queryfortop5countriesin2003,accordingtoTradeIndex ............................... 27Figure13:Regressionresultsona3Dsurface .......................................................................... 29Figure14:Formulasforoptimizationcalculations .................................................................. 30Figure15:Efficientfrontierofrandomlygeneratedportfolios .............................................. 31Figure16:FormatforCvxOptPythonfunction ........................................................................ 31Figure17:MainfunctionofPortfolioOptimization ................................................................. 32Figure18:Finaloutputofoptimizedweights ........................................................................... 32Figure19:Graphofvariableimportance .................................................................................. 33
vi
Table of Tables Table1:Modelinputs,outputs,succinctdescriptionsandresearchsources .......................... 9Table2:ListofThirdPartytoolsandpackages ........................................................................ 12Table3:ListofCollaborationTools ........................................................................................... 12Table4:SnippetofOutputCSVforWorldTradeModel .......................................................... 28Table5:TheinputdatafortheCobb-DouglasProductionFunctionModel ........................... 29Table6:InputDatasetofAssets’ReturnValues ....................................................................... 30
vii
Executive Summary
1
1. Frontiers in Russian Economics Studies Accordingtoresearchdonebythe Center for Economic and Financial Research (a
Moscow-based independent think tank), enhancing human capital through education reform is one of the keys to unlocking the potential of the Russian economy. The Organization for Economic Co-Operation and Development (OECD) has even coined a term called “knowledge-based economy” based off of the idea that knowledge and technology foster more economic growth and expansion (OECD, 1996). Buildingaknowledgeandunderstandingofeconomicsrequireshugeamountofresearchinvariousformats--fromelectronicresourcestoconventionaltextsorresearchpapers(Acton,2013).Inthepast,researchers,students,andprofessorshadtoanalyzeeconomicdatabymanuallyapplyingeconomicmodelstoanticipateissuesandproblemsthatcanariseintheeconomy(Morey,2016).Theeconomistsmustperformresearchtofindtheapplicableeconomicmodelforthesituationtheyareanalyzing,andthenconvertthemodelintosomesortofprogram(orevencodeitthemselvesifaprogramdoesnotyetexist)inordertohandlebiggerdatasets.Thisprocessisextremelytediousand,ifdonemanually,virtuallyimpossibletoapplytolargedatasets.
Previousattemptsatmakingeconomicmodelsreadilyavailableforresearchers,professors,governments,andstudentsalikehavebeenmostlyimplementedusingmanualclassicalmethods.Thosemethodscanbedefinedastextbooks,researchpapers,articles,andlectures.However,associetyhasbecomemoredigitized,therearenowonlinetutorialsandforumsthatfosterdiscussiononeconomics.Theseattemptsarewidespreadanduncentralized,andtoutilizethesetoolsasalearningmodulewouldrequiresomesortoforganizationanddatastorage.Thereisnocentralrepositoryforthisspecifickindofinformation.Alibraryoronlinedatabasemayhavevarioustextsoneconomicmodels,butsimpleaccesstothemodelsthemselvesstillrequiressiftingthroughvarioussources,whichconsumesresearchtime.
TherearealsomanydifferentsoftwareandprogramminglanguagesthatalloweconomiststoperformthesameresearchandBigDataanalysis.Themaincaveatwiththesesoftwareandprogramminglanguagesisthatthesoftwareistypicallyunavailableoutsideofcorporateoruniversitysettings,andprogramminglanguagescantakeuptomonthstomaster.Whenthemainpointofresearchistoperformdataanalysis,codingthemodelsbecomesanecessaryevil.Simplyput,thereiscurrentlynoeasywaytocutoutthis“middleman”ofeconomicmodeling.
OurprojectwiththeFinancialUniversityundertheGovernmentoftheRussianFederation(FinancialUniversity)aimedtoaddressthisconcernandforgethisfrontier.TheFinancialUniversityisseekingtofinditsplaceinthedigitizedworldthatthisanalysisandresearchismovingtowards.Itsemphasisoneconomiceducationunderscoresthatthereneedstobeauniversity-widedigitalplatformtounifyavastnetworkofprofessors,researchers,andstudents,andfacilitateexpeditedanalysisoftrendsandbehaviorsintheRussianeconomy.Thiswillallowresearcherstocreatebettereconomicstrategiesthatwillfosterabroaderunderstandingoftheeconomytoallowforfutureimprovementsandcorrections.Furthermore,thefutureofthisplatformwillallowtheFinancialUniversitytoseewherestudents’mainareasofinterestamongeconomicresearchtopicslies,andscope
2
outtop-performingstudentsforpotentialscholarships,grantsorjobopportunities.Thegoalofthisprojectwastocreatethetooltoassistwitheconomicresearchandeducation. Toachieveourgoal,wesetouttocompletethreemainobjectives:
1. Identifymajoreconomicmodelsthatcoverawiderangeofeconomicfields;2. Programthemodelssuchthatstudentsandresearcherscaninputtheirowndata
andmanipulatethemodelsforanalysis;3. Createaframeworkforaninteractivewebsitethatimplementsthemodelingsystem
thisprojectseekstofacilitate.Throughthisimplementation,userswillbeabletomaximizetheirresearch
productivityandminimizetheamountoftimespentonmundane,repetitivetasksassociatedwithinputtingcompiledeconomicdataintoeconomicmodels.Inthenextchapter,wewillreviewsomebackgroundinformationonthecurrentstateoftheRussianeconomytofurtherclarifythesignificanceofthisproject.Wewillalsouncoverthepastandcurrenttoolseconomistsusetoperformtheirresearch,andpavethewayforthefinalplatformweaimtocreate.
3
2. Economic Models: Scope, Significance, and Applications
InthischapterwewillprovideabriefbackgroundtothepastandcurrentstatusoftheRussianeconomyandsummarizecurrenteconomictrends.WewillthenexplainwhytheRussianeconomyisinneedofbettereconomicanalysisforeducationalandresearchpurposes.Thiswillprovidearationaleforwhyoursponsor,theFinancialUniversity,hasrequestedthisresearchtobedone.Thiswillbefollowedbyexamplesandexplanationsofrepositoriesandtoolsforeconomicanalysis,bothpastandpresent. The Russian economy’s unique path and need for better modeling
Russia’seconomicprogresshasbeennotablyslowerthanmostcountriesexperiencingatransitionofpower(Berglöfetal.,2003),anditspathhasbeeninmanywaysuniqueanddifficulttopredict.AfterthefalloftheSovietUnionin1991,Russiastruggledtoregainitsfootingandwasforcedtobeginitsrebuildingduringaperiodofeconomicrecession.
Whilethisisnottransitioncountries,Russia’srecessionlastedmuchlongerandwasworsethanothertransitioningeconomies(Kolodko,2000).InFigure1,itisclearthatRussiaspentmoretimeinarecessionthanmostothercountriesexperiencingatransitionofpower.Furthermore,anotherindicatorofRussia’seconomicperformancecanbeseenwithitslowGrossDomesticProduct(GDP).ByconventionGDPistypicallyusedtodeterminethehealthofaneconomy(Berglöfetal.,2003).
Figure1:TimespentbyEasternEuropeancountriesinatransitionalrecessionafterthe
falloftheSovietUnion(inyears).
4
Figure2belowdisplaystheinstabilityofRussia’sGDP.Asacountrysittingonapproximately30%oftheplanet’snaturalresources,theRussianeconomyshouldbeoutperformingmostcountriesduetothisgeographicadvantage(Korabik,1997).However,becauseRussia’sGDPreliessoheavilyonoilprices,theeconomyremainsextremelysensitivetothechangingvalueofoil.
Figure2:Russia'scurrentGrossDomesticProduct(GDP)
TheRussianeconomyhasseenslightimprovementtowards2016,albeitatastrikinglylowrate.Historicallylowoilpricesandinternationalsanctionshavepreventedtheeconomyfrommakingastrongrecovery.AccordingtodatacollectedbyBloomberg,theGDPdeclined0.225%in2016,sinceitpulledoutoftherecessionin2016(Andrianova,2017).ThiscanbeseenbelowinFigure3.
Figure3:Russia'scurrentGDPgrowth.
Thisslowgrowthiscauseforconcern,especiallyifthepriceofoilcontinuestodropbelowasustainablevalue.DuetothenatureoftheRussianeconomy,thereneedstobemoreanalysisandresearchdoneinordertounderstandthemechanicsoftheeconomyandbuildnewtechniquestoimprovegrowthandsustainprosperity.
5
How does education and research support economic growth? WilliamWalstad,theDirectoroftheNationalCenterforResearchinEconomic
Education,statedthat“nationsbenefitfromhavinganeconomicallyliteratepopulationbecauseitimprovesthepublic'sabilitytocomprehendandevaluatecriticalissues”(Walstad,1998).Insummary,increasedeconomicliteracyhasthepotentialtoimproveeconomicperformancebecauseitenablescitizenstomakesmarterdecisionsregardinginvestments.Byunderstandinghowtheeconomyworks,acitizencanmakeeducateddecisions,whichleadstoanincreasedreturnoninvestments.Furthermore,themoreresearchthatisdoneontheRussianeconomy,themorepossiblesolutionsandpreventativemethodscanbecreatedtomitigateeconomiccrisissituations.Forexample,ifthepriceofoilsuddenlydrops,aninformedcitizenwouldbeabletounderstandthattheRussianGDPwouldbenegativelyaffectedasaresult,andwouldmanagehisorherassetsaccordingly. What are the current methods of economic study?
Economistscurrentlyusedirectresearchandhardcodingtoconducteconomicanalyses(Pavlov,4April,2017).Economistsmanuallysearchthrougheconomicliterature,notes,journals,and/oronlinesourcesinordertofindtheapplicablemodelsandformulas.Thentheyhavetomanuallyhardcodesaidmodelsandformulasintotheprogramoftheirchoice.Economistswastevaluabletimeperformingthisresearchastheyaretryingtosolveaproblemoranalyzesomeeconomicphenomenon.Forhigher-leveleconomicmodels,thereisaplethoraoffactorsthatrequiresomeoneproficientinaprogramminglanguageorproprietarysoftwaretofullycapture.Economistsarefacedwithadecision:eitherlearnhowtodoitthemselves,whichtakesasignificantamountoftime,orcontractanexternaldevelopertotacklethetechnicalaspectsofthemodel. What are economic models? Economicmodelsareoneofthebestwaystolearnabouteconomicbehavior.Simplyput,theyaretheoreticalconstructsusedtoexplaincurrentbehaviorandpredictfuturebehaviorofeconomicsystems(Morey,2016;Mankiw,2001).Theyareessentialtoeconomictheory.Theyallowyoutounderstandhowandwhyspecificeconomicbehaviororphenomenaoccur(SamuelsonandNordhouse,1998).Thesemodelscantakemanyshapes--frombasicgraphstocomplexequations.Theyarecreatedbycalculatingtherelationshipbetweenvariablesinasystem--aswellastakingintoaccountthepastandpresentbehaviorofthesystemundervaryingcircumstances--tocreateatheoreticalgeneralizationabouthowthesystembehaves.
Throughtheuseofeconomicmodels,economistsgaininsightintotheinnerworkingsofasystem.Theyallowtheusertopeelbackhiddenlayersandgraspthemorefine-tuneddetailsthatmaynotbeexplicitlyvisible,andpavethewayforamorewell-roundedknowledgeofthesystem.Themoreknowledgeonehasofasystem,themoreroomthereisforunderstandinghowtoimproveit.Onalargerscale,onecanuseeconomicmodelstostudyactualeconomiesofentirecountries.Theycandeterminehowvaluablearesourcemaybe,ortoanalyzeriskfactorsofpoliticalmoves.Thisallowsonetoestimatetheoutcomeofpotentialdecisionswithouttheriskofactuallymakingthem.Indoingso,
6
onecanseethecollateraleffectsofadecision,maximizingtheconservationoftime,money,andevenlives. Why are models important?
Toemphasizetheimportanceandimpactofeconomicmodeling,letuslookatanexampleofastudydonebyresearchersattheUniversityofSouthernCalifornia’sKeckSchoolofMedicineandTruvenHealthAnalyticsinCambridge,Massachusetts(Johnson,2015).Aneconomicmodelwascreatedtodeterminethebenefitsofco-testingforcervicalcancerandhumanpapillomavirus(HPV).Themodelconcludedthatbycombiningthetwotests,doctorswouldbeabletodetectupto150,000additionalcasesofcervicalcancer,prevent100,000deaths,andsave$4.4billionovera40-yeartimeperiod.
Economicmodelsvaryfromcasetocase,buttheirapplicationsmayalsobeusefulinotherscenarios.InastudydonebyGibbardandVarian(1978),theresearchersconcludedthattheamountofvarianceamongeconomicmodelscanbeverysmall,anditisdifficulttodistinguishbetweenthedifferenttypesofmodels.Thisisduetothefactthateconomicmodelscanrepresentmanyphenomenaandsituations,withverysmallvariationamongthem.Thismakesitdifficultforeconomiststofindthemodelmostapplicabletotheproblemathand.TheFinancialUniversityislookingforawaytoorganizeandstoreimportanteconomicmodelsinacentralizedlocation--arepository--toexpeditetimespentresearchingforitsstudentsandforothereconomicresearchers.Bydoingso,thestudentsandresearchersoftheFinancialUniversitywillbeabletogetstraighttotheirresearch,ratherthanwastingtimesearchingandcodingthemodelstheyneedtoperformtheiranalysis. How will these models be accessed by university staff and students? Arepositoryisacentralizedlocationinwhichdataarestoredandfromwhichtheycanbeaccessed(OberkamfandRoy,2010).Therefore,themodelsshouldbehousedinasinglerepositorytocreatea“one-stop-shop”foreconomiststousetocarryouttheirresearch. Thegoalofarepositoryistocutdownthetimewastedontrivialsteps,suchascoding,inordertoalloweconomiststogaintheinformationtheyneedquickly.
Witharepositoryofeconomicmodels,economistscouldefficientlysearchthroughthedifferentmodelstofindtheonemostapplicabletosolvetheproblem.Previously,GibbardandVarian(1978)categorizedeconomicmodelstheystudiedinaresearchpaperdependingonthetypeofcalculationperformed.Thisallowedthereadertofullyunderstandhowacertaintypeofmodelwithinaspecificcategorycanbeappliedtoreal-worldsituations.Thesameresultscouldbeachievedbyimplementingarepositoryoffullyfunctioningeconomicmodelsthatarereadyforuse.Forexample,ifeconomistswerelookingforaneconomicmodelforportfoliooptimization,theycouldsimplysearchforoptimizationmodelsandeasilylocatethedesiredone. Thevalueofarepositoryofeconomicmodelscanbebetterunderstoodbystudyingapreviousimplementation.In2012,astudywasdonebytheUniversityofWarwickinwhichtheyanalyzedbusinessmodelsandbusinesscasesappropriatetosupplychaindata(Cave,2012).Theystoredthisdatainarepository,whichallowedstakeholderstoaccesstheinformationeasily,andassistedthemindesigningsuitablebusinessmodelstoenhancethevalueoftheirexistingrelationships.Forexample,ifastakeholderwantedtocreatea
7
cost-benefitanalysisofpurchasinganewmachinedesignedtomaximizeproductionwhileminimizingtime,theycouldsimplygointothisdatabaseandsearchforthebusinessmodelforthissituation.Then,theycouldinputtheircompany’sdataandstatisticstoseehowbeneficialthisbusinessmovecouldbeinthelongrun.Figure4belowprovidesavisualizationofthemechanicsofsucharepository.
Figure4:Mechanicsofarepository
What are current methods used to study and analyze economic models? ThereareseveraltoolseconomistsinRussiacancurrentlyutilizetointeractwith
theseeconomicmodels.SoftwarelikeMatLab,Maple,andStataorprogramminglanguageslikeR,Python,andC/C++canallbeusedtocodeeconomicmodelsusingvariablesthattheusercanchangeaccordingtodatatheyneedtoanalyze(Dr.OlegPavlov,personalcommunication,April19,2017).Thereareeventutorialsonlineonhowtousethesesoftwareprogramsandprogramminglanguagestocodeeconomicmodels(SargentandStachurski,2014).However,whilethesetoolsareallextremelyusefulandapplicabletostudyingeconomicmodels,theyarenotthetoolsofaneconomist,butratherofacomputerscientist.Theissueisthatonemustbeproficientinusingthesetoolsinordertoapplythemtoeconomicmodelingandanalysis.Aresearchermustgothroughsomesortof“middle-man”beforeperformingtheiranalysisusingthesetools.Inmorecomplexcases,aresearchermayneedtoevenhireaprofessionalsoftwareengineer(sometimesevenateamofengineers),ormaybeattendaclassonhowtousecertainsoftware.Economistswouldprefertonotdealwiththistypeofwork,buttodosotheywillneedotherstohelpthemtobeabletofocussimplyonapplyingeconomicmodelsusingdatathattheyorothershavecollectedtoanalyzetheeconomictrendsandmakepredictions.
Inthenextchapter,wewillexplainthemethodswewillapplytocarryoutourresearchinordertoachieveourgoalofcreatingaplatformforFinancialUniversitystudentstohaveeasyaccesstoeconomicmodelstoenhancetheireducationandfostertheirgrowthinthefield.
8
3. Methodology: Program Fabrication Thescopeofthisprojectwastocreateasystemthatwouldactasarepositoryforeconomicmodels.Afterconsultationwithouron-siteliaisonProfessorAntonLosev,wedeterminedwhatwasfeasibletoaccomplishwithinourtimehere.Weassessedourgoalsinrelationtothetimeframeweweregiventoachievethecreationoftheplatform,andconcludedthatwewouldtackleonlyseveralessentialaspectsoftheoveralltasktobeabletodeliverqualitywork. Theseareourobjectives:
1. Identifymajoreconomicmodelsthatcoverawiderangeofeconomicfields;2. Programthemodelssuchthatstudentsandresearcherscaninputtheirowndata
andmanipulatethemodelsforanalysis;3. Createaframeworkforaninteractivewebsitethatimplementstheeducational
modelingsystemthisprojectseekstofacilitate.
Bydoingso,wehopetoprovidetheFinancialUniversitywithatooltoallowforeasieraccesstoeconomicdataandfacilitateresearchontheRussianeconomy.Thistoolcanbefurtheradaptedforuseinthebusiness/consultingfield. Economic Model Research Afterweassessedourgoals,weconcludedthatthereshouldbefourstartingmodels:WorldTrade,Cobb-Douglas,LoanScoring,andPortfolioOptimizationmodels.Thesemodels,proposedbyoursponsor,exemplifythediversityofanalysismethodsthereareintherealworld.Byensuringthisdiversityexistsintherepository,studentswhousetherepositorywillbemorewell-roundedwhenproblemsolving.Themodelscoverthemainaspectsofeconomicmodeling:optimization,scoring,ranking,anddataassessment.Additionally,eachmodelhasthreemaincomponentsthatareessentialtotheeducationalvalueofthisrepository:inputdata,modelalgorithm,andmodeldescription. Thefirsttaskwetackledwasdatacollection.Wedecidedthatthedataforeachmodelshouldcomefrompublicsources.Thiswasdeterminedbecausepublicdataisthemostaccessiblesourceforthetargetusersofthissystem,whoarestudentsandresearchers.Whilediscussingthiswithoursponsor,welearnedthatuserscaneasilypulldatafromawebsitecalledKaggle.ThiswebsiteisessentiallyahubforBigData--ithostscompetitionsondataanalysisandprovidesfreedownloadsoflargedatasetsasCommaSeparatedValuefiles,or‘.csv’files,whichisthestandardfiletypeforinputsinthisrepository.Furthermore,thedatasetsareoftenextremelyspecialized.Youcanfindanythingfrommovielists,stockexchangehistory,andeventrendingYouTubevideos.Thisallowsuserstoapplythemodelstoanytopictheymayhaveinterestin.
Next,wehadtoactuallycodetheeconomicmodels.Intermsofthealgorithmconstruction,wedeterminedthattheeasiestlanguagetousewasPython.Thiswasdeterminedforavarietyofreasons.Pythonfollowsastrictpactof“Thereshouldbeone--andpreferablyonlyone--obviouswaytodoit.”(Peters,2004).WefeltthatthisaspectofPythonisessentialforcollaborationwithourRussianteammates,whowerefamiliarwiththisprogramminglanguage.Pythonalsohasthelargestthirdpartyselectionfordata
9
sciencepackages.Thiswillgreatlyimprovetheperformanceofthealgorithms,asthesethirdpartypackagesareextremelypowerfulandwillensureefficientandaccurateresultswhilecuttingdownonprogrammingtime.CompetitionforPythonisnotevidentfordatascience.Infact,onlytheRprogramminglanguageiscomparabletoPythonforstatisticalanddataanalysisusage.WechosePythonoverRsincethelatterlanguagecannotbeeasilyintegratedintoawebsite'sframework,whilePythoncan.Duetoitswidespreadinternationaluse--thereisanabundanceofframeworksspecificallydesignedforwebdevelopment.
ThelastmaincomponentofourplatformisthedescriptionWiki.Thisisthedocumentationthatuserscanuseasguidelinesforthesoftwareandmodelsthatwedesign.Theyexplainthesignificanceofthedata,thealgorithmstructure,theoutputstructureandexplanationastowhatausercandowiththeresultsofthealgorithm.Togainanunderstandingofthemodelsourselves,wehadtoconductextensiveresearchonthemodelstounderstandtheparameters,calculationsandexpectedresults.Thisresearchwasthenputintotextalongsidethecorrespondingcodesothatuserscanseetheconnectionbetweenthedatamanipulationandthemodelalgorithm.Tofurtherenhancetheeducationalvalue,weensuredthatthecodecontainedmanyvisualizationsandgraphicalrepresentationsofthedatamanipulationtoprovideforamorewell-roundeduserexperience.Thiscomponentisessentialtotheeducationalaspectoftherepository–byreadingthemodeldescriptions,userscanactuallygainanunderstandingofhowthemodelworksbeforeusingthealgorithmstoapplytotheirowndataforanalysis. Programming the models Tobeginthecodingprocessforeachmodel,wefirsthadtodeterminetheinputandoutputparameters,aswellasthedescriptionofwhatthemodelwillactuallycompute.Thiswillshedlightontheeducationalvalueofstudyingthemodel,asseeninTable1below.
Table1:Modelinputs,outputs,succinctdescriptionsandresearchsources
Model Input Output Description Source
World Trade Trade and GDP
Datasets
CSV file with ranked countries & command
line query interface
Model shows what inputs need to be altered for a desired output. This
also graphically shows the relationship between parameters
Mikic and Gilbert (2009)
Cobb- Douglas
Dynamic dataset
Graphic representation of relationships
Model shows what inputs need to be altered for a desired output. This
also graphically shows the relationship between parameters
Stewart (2008)
Portfolio Optimization
CSV file of stock
portfolio returns
Optimized weights for each asset for
maximum return
Model shows the relationship between investments.
Sharpe, Alexander and Bailey (1999)
Scoring Credit history dataset
Binary predictions of whether customers
should receive a loan.
Model allows the user to gain a better understanding of machine
learning techniques.
Gadil (2016)
10
AmorerobustexplanationoftheinnerworkingsofeachmodelcanbeseeninAppendixD. Organizing and storing the models
Westoredthemodelsinanetworkdatabasewhereallthealgorithmsforthevariousmodelswouldbeaccessible.Weorganizedtheselectedeconomicmodelsintofolderscorrespondingtotheirpracticaluse.Forexample,producingcalculationsorperformingdataanalysis.Withinthosefolders,theywillbesubdividedintothecategoriesinwhichtheyoperate--forexample,optimizationorrankingparameterswithinlargedatasets.Bydoingso,weexpeditetheeconomist’sresearchandallowthemtogetstraighttotheanalysis. Front-end framework Thefront-endofthisprojectwouldideallybeaccessibletoalargeaudience.Sincethemainideaofthisplatformistoprovideatoolforresearchandeducationlimitingtheaccessibilityoftheplatformwouldessentiallynegatetheimpactwehopetoachieve.Thisinitselflimitsthedevelopmentprocessofthedeliverabletospecificmethods.ThefirstistomakeanexecutablepackageforWindowsbasedsystems.However,hislimitstheaudiencesincethiswouldnotbecompatibleforMacintoshcomputers.Thesecondavenuewouldbetomakeaweb-basedtool.Thiswouldbetheidealsolutionsinceweb-basedtoolsareaccessiblefromallmoderndevices,andcanshiftthebalanceoftheprocessingfromtheuserclienttotheserver.
Wecannowdeterminethestructureofthewebsitebasedontherequirementsforthesystemasawhole.Thetwopossibilitiesthatcanbeexploredinthistimeframeiseithermakingalocalwebsitethatthefinancialuniversitywillhost,ormakingapagethatwouldbehostedbytheFinancialUniversityfittedwithdescriptions,Pythoncode,andalgorithms. Thefirstoptionthatcouldbeexploredwouldallowanyuser(whetheritbestudent,teacher,orcompany)toruncodewithoutdownloadinganyfileslocally.Thiswouldalleviatetheprocessingrequirementsontheclientendofthesystem,andmakeitaccessiblefromanydevicethatcanconnecttotheuniversityoranyotherhostinglocation.Accessthroughtheuniversitywillrequirestudentsandteacherstologinwiththeiruniversitycredentialstoaccesstheplatform.Tobroadentheplatform’soutreach,itcouldbehostedonapublicwebsite.Thiswouldallowpeoplewithoutcredentialstoaccesstheplatform.Thesecondoptionistohaveafrontendstructurethathasadetaileddescriptionofthemodel,howitworks,howtorunit,andwhatdatacanbefedintoit.Thispagewouldalsoentailthedata,code,andnecessarythirdpartypackages.Thispossibilitycomeswiththeassumptionthattheuserhaspreviousknowledgeorguidanceavailableaboutrunningtheprogramlocally. Sincewehonedourfocustodevelopingthemodelscorrectly,wedecidedtousethesecondoptionforthefrontendsinceitislesstimeconsuming.Thefirstoptioncouldstillbeachievableinthefuture,andwouldallowtheuserstohaveanenvironmentthatwouldencapsulatetheprocessofdevelopingamodel,testingamodel,andsharingamodelwithemployersorteachers.
11
Survey Development Allofourobjectivesliewithintheunderstandingthatthistoolwillbecomeanintegralpartofeconomicseducation.Todefendourassumption,wesentoutasurvey,throughatoolcalledMicrosoftForms,inordertogatherresultsrapidlyinanorganizedfashion.Sincewewantedthesurveytobefast,easy,andaccessible,wefabricatedsevenquestionsthattouchonbackground,appealofthetool,andwaysthetoolshouldbeused.Figure5showsthelistofquestions,alongwiththesectionsthatcorrespondtoitssignificance.
Figure5:Surveyquestions
Third Party Tools and Packages Toaccomplishourgoalsofcreatingefficientandaccurateprogramsforeachmodel,weneededtousefourspecificthirdpartytools.Thesetoolsarepackagesthataredesignedtoeasilyparse,manipulate,andanalyzeBigData,andareexplainedinTable2below.Thesepackagesareextremelypowerfulandallowustoperformhigh-levelcalculationsonlargedatasets.
12
Table2:ListofThirdPartytoolsandpackages
Package/Tool Name
Application Description
Pandas Big Data analysis in Python
Allows data to be easily manipulated
NumPy Big Data analysis in Python
Allows mathematical operations to be easily run
Spyder Python IDE Streamlines programming process by catching bugs quicker
CvxOpt Data optimization in Python
Allows convex optimization application straightforward
Aspreviouslymentioned,allaspectsofthisprojectuseopensourceresourcesin
ordertoachievethegoalssetout.Thesethirdpartypackagesstreamlinedtheprocessesweneededtousetoachieveourtechnicalgoals.Thiscomeswiththerequirementthattheenduserwouldalsohavetodownloadthesepackageswiththecodeattachedtoeachmodel.Wewillbebundlingthecodewiththethirdpartypackagesinordertomaketheprocessfortheusertousethealgorithmseasier.Itisimportanttonotethatwearenotviolatinganytrademarks,copyrights,orpatentssinceeverythingisopensource.
Intermsofcollaboration,weuseddiverseoftoolstosharecode,workondocumentstogether,communicate,sendoutsurveys,andtrackourprogress.Table3explainseachtool'ssignificance.
Table3:ListofCollaborationTools
Package/Tool Name
Application Description
GitHub Version Control Allowed the team to effectively share code Google Drive Document Storage and
Fabrication Allowed the team to work on document and
presentations simultaneously. Telegram Communication Enabled the team to effectively communicate over
laptops and mobile devices Trello Organization Allowed the team to effectively track progress, set
tasks, and stay on top of deadlines Thegoalsetbyoursponsorwasextremelychallenging,butwiththesemethodswe
wereabletofostereasycollaborationbetweenusandtheFinancialUniversitystudentstocompleteourobjectives.Inthenextchapter,wewilldiscusstheresultsofourprojectandourgoalsforitsfuture.
13
4. Results and Discussion TheresultofourprojectwasthedevelopmentoftheFinancialUniversity’sWikifor
economicmodels.ThewebsiteislocatedontheFinancialUniversity’swebsite.Aftercompletingourobjectives,wecreatedaframeworkthatsetsthefoundationforaplatformthathasthepotentialtobecomeavitalresearchandanalysistoolnotonlyfortheFinancialUniversity,butforeconomistsandeconomicstudentsacrosstheglobe.Ourprojectutilizedourstrengthsindataanalysisandmanipulation.ThisintroducesaFinancialUniversity-specificwayforresearchersandstudentswhoarespecializedinareasotherthandatasciencetoleveragetheirexpertisewithBigData,andproduceapreviouslylaborintensiveresultveryquickly. Survey Results
Theresultsofoursurvey(shownbelow)paralleledourdevelopmentoftheplatform.
Figure 6: Survey results – gauging the audience
ItisimportanttonotethatwedidnotreceiveanyresponsesfromFinancialUniversityfaculty.Becauseofthis,wecannottellifourplatformwillbeofuseasateaching
14
toolforprofessorstouseintheclassroom.Wedecidedthatwewouldmoveforwardundertheassumptionthatour30respondentswillbecomeourtargetaudience,asweneededtobegindevelopingourplatform.Thatbeingsaid,wecateredourplatformforamorepersonal(outsideoftheclassroom)educationaluse.Asourimplementationismodular,thistoolcanbefurtherdevelopedtobecomeateachingtoolforprofessorstouse,ifneeded.
OurrespondentsweremostlyundergraduatestudentsoftheFinancialUniversity,whosemainresearchisconductedthroughcurrentsoftwarelikeMicrosoftExcelandAnyLogic.However,themajorityofthesestudentsstatethattheywouldlikeamoreinteractiveeducationalsoftware.InFigure7below,itisclearthatasignificantportionofthestudentssurveyedwouldlikemodeldescriptionsofthemodelstounderstandthetechnicalequationsandhigh-leveltheorybehindthem.
Figure7:Surveyresults–decidingwhattoincludeintherepository
TheseresultssupportedtheformatoftheWikiandgaveusabetterunderstandingofourpotentialaudience.Bycollectingthisdata,wewereabletodeveloptheplatformwiththedesiredmodulesandmodelsforend-users. Creating the framework
Withoursurveyresultsinmind,wecustomizedourplatformtoourtargetdemographic-FinancialUniversityundergraduatestudents.Wedecidedthatwithourtimeconstraints,creatingasimple“Wiki”pageofthemodeldescriptionsandlinkingdownloadablecodeofthemodels(shownbelowinFigure8)wouldbetheeasiestimplementationwecouldsuccessfullyimplement.
15
Figure8:CurrentimplementationoftheWiki
ThisWikipageisontheFinancialUniversity’swebsitesoitiseasilyaccessibletostudents.TheWikipageincludesathoroughdescriptionofthemodel,theformulasneededformanualcalculation,andthedownloadlinksforthefilesholdingthemodelalgorithms.Thisallowsstudentstogainanunderstandingofthetheorybehindthemodels(throughthedescriptions),andanalyzethelogicofthemodels(throughtheheavilydocumentedalgorithms).Furthermore,asthestudentscandownloadthealgorithmsfortheirownuse,theyareabletoinputandviewresultsontheirowndata.
ThemodularframeworkoftheWikialsoallowsforfuturedevelopmentofthisprojectthroughaweb-basedinterface.Notonlydoestheframeworkmakeourcodeeasiertoimplementinthefuture,italsoprovidesatemplateforaddingnewmodels.Thistemplatewilloffertheabilitytoseamlesslyintegratemoremodelsinconsistentform,providingascalabilityfactortoourproject.However,thedevelopmentofanywebapplicationisboundtohavelimitations. Study Limitations
ThefirsttwoweeksofourtimeattheFinancialUniversitywasessentiallyspentdiscussingwithoursponsor,andourRussianteammates,whattheend-goalofourprojectwas.Theideaoftherepositoryhadbeensetinstonesincethebeginningofthisproject,butwehadtodeterminehowtoshapethefinalplatformandWiki.Wealsohadtodiscusswhatwasfeasibletocompleteintheremainingweeks.Onlyafterwesuccessfullyidentifiedthesetwogoals,wecouldbeginourwork.
Furthermore,asweareateamofcomputerscience,roboticsengineering,andindustrialengineeringmajors,ourknowledgeofeconomicsandeconomicmodelswasextremelylimited.Foralltheeconomicmodels,wehadtospendweeksperformingresearchwithourRussiancounterpartsinordertoensurethatouralgorithmswere
16
accurate,andthateachstepwasheavilydocumentedtoensuretheeducationalvalueoftherepositoryremainedintact.
17
5. Conclusion and Recommendations
Throughthedevelopmentandimplementationofoureconomicmodelrepository,webelievethatbotheconomicsstudentsandresearchersalikewillbeabletoexperiencethesignificantsimplificationofthedatascienceaspectofeconomicsresearch.Wehavefoundthattheincludedmodelswillaideawiderangeofproblemsthatstudentsandresearchersface.Whiletherearealternativestoourplatform,suchastheRprogramminglanguageandproprietarysoftwaresuites,theyrequiretheusertobecomeproficientineitheraprogramminglanguageoracommercialpieceofsoftware,whichcanbedifficulttolearnfromscratch.
Bycreatingcodetomanipulatelargeamountsofspecificdatatoobtainclearlydefinedresults,wehavearmedeconomistsandstudentswithapowerfultoolthatcantaketheirresearchtothenextlevel.Thedatawecollectedthroughoursurveyhelpedusdevisetheoptimalwaytodeliverthisplatformtotheclientsthatwillhaveaccesstoit.Additionally,theworkwehavedonehasbeenperformedwithfuturedevelopmentoftheplatforminmind.Futuredevelopmentswillneedtoincludetheadditionofothermodels,thedevelopmentofapracticalfront-enddashboard,andtheabilitytoinputthird-partydatasetsthatcaninteractwiththemodels.ByloweringthebarriertoentryforstudentsandresearcherstouseBigDatathroughcomprehensivemodeldescriptionsandfunctions,ourgoalistobolsterfinancialliteracyandpromoteeconomicresearchandeducation. Recommendations
Afterweresearched,programmed,andimplementedourrepository,weanticipateandencouragethefollowingfuturedevelopmentsandimprovementsforboththeendusersofourplatformaswellasforsoftwaredevelopers. Forusers: • Werecommendreadingthedocumentation(developedbyus)providedwiththe
models.Whilelearningtheactualcodedefeatsthepurposeofourproject,havingageneralideaofwhatisgoingonwillhelpwiththeoptimalutilizationofthemodels.
Fordevelopers: • Back-testingonrealmarketdataisrecommended,assomemodelswereonlytestedon
ourowncollecteddata.• Whenaddingadditionalmodels,werecommendtocreateseparatefilesforeachmodel,
andutilizepre-existingdatamanipulationpackages,suchaspandas,NumPy,andsklearn,thatweincorporatedintoourmodels.Thiswillnotonlymakemanycomplicateddatamanagementtaskseasier,butalsoallowformodelstosharerobust,provencode,minimizingtheneedforrepetitivecode.
• Whendealingwithlargequantitiesofdata,theamountoftimeittakestoactuallyruntheprogramcanbesubstantial.Keepthetimeittakestocalculatetheresultsinmindwhenincorporatingnewfeaturesormodels.Ifthecalculationsaredoneonacentralizedserverandtheresultsaredeliveredthroughawebapporothermedium,runningcalculationsformultipleusersmaybogdownperformanceandmakethe
18
platformunusable.Thiscanberectifiedbydevelopingback-end,loadmanagementcode.
• WhileusingthePythonconsoleisbetterthaneditingthecodeyourself,amoreuser-friendlyuserinterfacewouldfurtherbroadentheuserbaseofthisplatformaswellasmaketheutilizationofsuchaplatformlessintimidatingtoanewuser.Thiswouldalsominimizetheusererrorbyprovidinglimitedvalueinputs.
• Ifthecodeiscentralizedtooneserverandaccessedthroughawebapp,werecommendtheuserinputbesanitized.Thiswillpreventmaliciouscodefrombeingintroducedintothemodelcode,preventingintentionalorunintentionalcorruptionofthefiles.Inthecaseofmodelsthatusesuchmethodsasmachinelearning,thecodeneedstolearnfrommultipleiterationswhatisgood,andifthecodewasresettoitsinitialstateaftereveryiteration,thisnewlearningwouldbelost.
Duringthenextstepsoftheproject,werecommendthatthefocusbeonintegrating
therepositoryintoacentralizedserverthatcanbeaccessedthroughawebapp.Thisisthenextlargeportionofthefinalplatform,andbecauseofthis,itneedsasignificantinvestmentoftimeandresourcestocomplete.TheadditionofmoremodelstotheWiki,whilenotbeingeasy,canbeaccomplishedincrementallynowthattheplatformstructureissetup.
Tosummarize,werecommendthenextstepindevelopmentofthisprojectshouldbetofocusondevelopingawebappthatallowsuserstonotbedependentonhavingtodownloadanycodeordataset,butratherinteractwithauserinterfacethatallowssimilardatainputandoutput.Thisrepositorywascreatedwiththeintentionofalleviatingtheextraeffortstudentsandresearchersmustputinwhenperformingeconomicsresearchandanalysis.WehopefuturedevelopmentofthisrepositoryleadstotheFinancialUniversitygainingauniqueandspecifictooltohelpboosttheirstudentswork.Bydoingso,wehopethatourplatformachievesouroverallgoalofprovidinguserswithatooltotrackandanalyzetheRussianeconomytogainabetterunderstandingofit.
19
References
• Acton,Q.(2013).IssuesinGeneralEconomicResearchandApplication.Georgia,Atlanta:ScholarlyEditions
• Andrianova,A.(2017,May17).RussianRecoverySputtersasEconomyContinuesSlogAfterCrisis.RetrievedSeptember15,2017,from
• BalatskiiE.V.andKonyshevV.A.(2006),Russianmodeloftheeconomy'spublicsector.London,England:Springer.
• Berglöf,E.,Kunov,A.,Shvets,J.,&Yudaeva,K.(2003).TheNewPoliticalEconomyofRussia.Cambridge,MA:MITPress.
• Cave,Jonathan,BusinessandEconomicModelsforSupplyChainDataRepositoryServices(July30,2012).RetrievedApril18,2017,fromSSRN:https://ssrn.com/abstract=2141913orhttp://dx.doi.org/10.2139/ssrn.2141913
• Few,S.(2005).DashboardDesign:BeyondMeters,Gauges,andTrafficLights.BusinessIntelligenceJournal,10(1),18-24.
• FinancialUniversityundertheGovernmentoftheRussianFederation(2001).WelcometoFinancialUniversity.Retrievedfromhttp://international.fa.ru/Pages/Home.aspx.
• Gibbard,A.,&Varian,H.(1978).EconomicModels.TheJournalofPhilosophy,75(11),664-677.Retrievedfromhttp://www.jstor.org/stable/2025484
• Jacobs,J.R.B.(2014).SecurityUsingDataAnalysis,Visualization,andDashboards.Somerset:JohnWiley&Sons,Incorporated.Retrievedfromhttp://ebookcentral.proquest.com.ezproxy.wpi.edu/lib/wpi/detail.action?docID=1599321
• Johnson,M.(2015,November18).EconomicModelSaysCervicalCancerCo-TestingwithmRNA-BasedMDxCouldSaveLivesandMoney.RetrievedApril15,2017,fromhttps://www.genomeweb.com/pcr/economic-model-says-cervical-cancer-co-testing-mrna-based-mdx-could-save-lives-and-money
• Kołodko,G.W.(2002).Globalizationandcatching-upintransitioneconomies.Rochester,NY:Univ.ofRochesterPress.
• Korabik,K.M.(1997,December1).Russia'sNaturalResourcesandtheirEconomicEffects[Scholarlyproject].InPennsylvaniaStateUniversityCollegeofEarthandMineralSciences.RetrievedApril21,2017,fromhttp://www.ems.psu.edu/~williams/russia.htm
• Kudrin,A.,&Gurvich,E.(2015).AnewgrowthmodelfortheRussianeconomy.RussianJournalofEconomics,1(1),30-54.
• Lee,J.C.(2016,April12).WhytheRussianEconomyIsTumbling.TheNewYorkTimes.RetrievedApril21,2017,fromhttps://www.nytimes.com/interactive/2016/04/12/world/europe/russian-economy-tumbling.html?_r=0
• Mankiw,N.G.(2001).PrinciplesofEconomics(2nded.),Boston,MA:HarcourtCollegePublishers.
• Monroe,J.(2016,March29).DesktopEconomicDashboardUsingR.RetrievedApril21,2017,fromhttps://monroestatistics.wordpress.com/2016/03/18/desktop-economic-dashboard-using-r/
20
• Morey,E.R.(2010).AnIntroductiontoEconomicModels.Boulder,CO:UniversityofColorado,Boulder.
• Oberkampf,W.L.,Roy,C.J.,&SafariBooksOnline.(2010).VerificationandvalidationinScientificComputing(1sted.).NewYork:CambridgeUniversityPress.
• Ouliaris,S.(2012).Finance&Development.RetrievedApril18,2017,fromhttp://www.imf.org/external/pubs/ft/fandd/basics/models.htm
• Rundensteiner,E.(2017,April14).Personalinterview.• Saeed,K.(2017,April14).Personalinterview.• Saeed,K.(2016).DevelopmentPlanningandPolicyDesign(2edition).England:
AveburyBooks.• Samuelson,PaulA.,andNordhaus,WilliamD..(1998).Economics.Boston,The
McGraw-HillCompanies,Inc.,Chapter1,(pages3-7).• Sargent,T.J.,&Stachurski,J.(2014,June10).QuantitativeEconomics.Retrieved
April21,2017,fromhttps://lectures.quantecon.org/about_lectures.html• Sharpe,W.F.,Bailey,J.V.,&Alexander,G.J.(1995).Investments.EnglewoodCliffs,
NJ:PrenticeHallInternational• Stewart,J.,Calculus:EarlyTranscendentals.ThomsonBrooks/Cole,6thEdition,
2008,(pages857-887).• TheKnowledge-BasedEconomy.(1996).RetrievedSeptember15,2017,from
https://www.oecd.org/sti/sci-tech/1913021.pdf• Varian,H.(2014).MachineLearningandEconometrics[PDFDocument].Retrieved
fromLectureNotesOnlineWebsite:https://pdfs.semanticscholar.org/fcb1/10653256f8e535bf074403c8847a6a3be76b.pdf.
• Vladimir,C.,&Galina,A.(2016).ProblemsoftheRussianEconomy.MediterraneanJournalofSocialSciences.
• Walstad,W.(1998,December1).Whyit'sImportanttoUnderstandEconomics.RetrievedApril21,2017,fromhttps://www.minneapolisfed.org/publications/the-region/why-its-important-to-understand-economics
• WorldBankGroup.(2017,January1).RussianFederation.RetrievedApril22,2017,fromhttp://data.worldbank.org/country/russian-federation
21
Appendices Appendix A: Sponsor Information TheFinancialUniversityundertheGovernmentoftheRussianFederation,openedonMarch2nd,1919,isoneofthetop15Russianuniversities(ExpertRussianRatingsAgency,2016).TheFinancialUniversityisstate-fundedandhoststhirtyeightsubdivisionsthatalllendthemselvestowardsthesuccessoftheirstudents--mainlyineconomicfields.ThemottooftheFinancialUniversityis“Wevaluethepast.Webuiltthefuture.”(FinancialUniversity).Theypridethemselvesontheadvancedlearningtechnologiesandstateoftheartprogramsthathelpmoldhigh-performing,expertgraduates. Initiallystartingwithanenrollmentofaround2,000students,theFinancialUniversityhasgrowntomorethan20timesthatsize.Overthespanofits10campusesand25universitybrancheslocatedthroughoutthecountry,theschoolhasanenrollmentofaround50,000studentsasof2016(FinancialUniversity).HighereducationopportunitiesrangefromBachelorDegreeprogramstoPhDprograms.Underthe21scientificschools,studentscanchoosetostudywithinthefieldsofeconomics,financeandlending,management,law,politicalscience,appliedmathematics,computerscience,andmore.Additionally,theFinancialUniversityalsooffersarangeofresearchopportunities.Theuniversityaimsitseffortstowardsthe“developmentofnewprojects”and“evolvingcreativethinkingoftheteachingstaff”toassemblehigh-levelresearchteamsinordertofurtherdevelopmentofknowledgeabouttheeconomy,finance,security,andcorporategovernment/businessstrategydevelopment.
Figure9:FinancialUniversityDepartmentStructure
InternationalcollaborationisanotherskilltheFinancialUniversityfostersinitsstudents;thefacultyattheuniversityteachtheirstudentshowtoinitiateandpromotedialoguebetweendifferentcountriesandculturesbyofferingarangeofexchange
22
programs.Bydoingso,theyhopetoshapewell-roundedandopen-mindedindividuals,whocanthenusethecommunicationandcooperationskillsgainedfromworkinginmulticulturalenvironmentsandapplythemtotheprofessionalworld.
ThisprojectspecificallyhastiestotheDepartmentofDataAnalysis,DecisionMakingTheoryandFinancialTechnology,whichisoneofthe14maineconomicdepartmentsintheuniversity.Thisdivisionspecificallydealswiththedecisionsinbusinessthatusedataanalytics,forecasting,andotherBusinessIntelligencetools.Thebranchleaderanddepartmentheads,Dr.VladimirSolovyev,Dr.IrinaDenezhkina,andDr.VadimFeklin,allhaveexperienceinbusiness,mathematics,andinformationtechnology.Theon-siteliaisonforthisIQP,ProfessorAntonLosev,isanassistantprofessorwithinthisdivision.
23
Appendix B: Interview Protocol Before the day of the actual interview:
Ø Email/contacttheintervieweeifwehavepermissiontousetheirname,title,andquotesfromtheinterviewinourprojectproposal.Furthermore,wewillaskifwehaveconsenttouseanaudiorecorderduringtheinterviewtoensurewecanquoteasaccuratelyaspossible.
• Ifwedonothaveclearancetouseanaudiorecorder,oneofourteammemberswillbetakingnotesduringtheinterview.Thesenoteswillbesenttotheintervieweeaftertheinterviewsotheyarefullyawareofthepointsofinformationgathered,andtoallowthemtoconfirmordenyanydirectquotes.
Attheactualinterview: Ø Introduceteammembersandproject/projectgoals.Ø Thanktheintervieweefortakingthetimetomeetwithus.Ø Clarify:Canweusethisinformationasacitationinourdocumentation?Wewillcitethe
dateandtimeofthisinterview,aswellastheyournameandtitle.Wemayalsousedirectquotes,aswewillbeusinganaudiorecorderforthisinterviewtoensurewedonotmisquoteanythingsaidtous,ormisinterpretanyinformation.
o Iftheanswerisyes,telltheintervieweethatwewillcontactthembeforewereleaseanydocumentationandaskifthecontentwrittenwithregardstotheirinterviewisanaccuratereflectionoftheirintentions/opinions.
Ø Askquestions1DataScience/ComputerScienceRelatedInterviewee
1. Haveyouhadanyprojectsthathavehadtoimplementthecreationofananalyticsdashboard/visualinterfacefromarepository?i. Ifso,pleaseelaborateontheprocessesthattheproject(s)underwent,the
challengesthatthespecificproject(s)encountered,andanythingelseimperativethatyoumightfeelisusefultous.
ii. Ifyouhadtochoosewhichstepsoftheprocesswerethemostimportant,whichstepswouldyouchooseandwhy?
2. Whattimeframesdotheseprojectsusuallytake?i. Iftheyvary,whattrendsdoyouseeinregardstothequalityoftheproject
versusthetimespentonit?3. Whatsetbacks,ifany,haveyourprojectsusuallyencountered,andaretherewaysto
preemptivelycounterthem?4. WhatexperiencehaveyouhadwithEconomics?Doyouhaveanyfamiliaritywith
economicmodeling?Ifso,pleaseexplainwhateconomicmodelingyouarefamiliarwith.
5. WhatdashboardshaveyouseenthatlookthebestforBigData(orspecificallyeconomicdata)?Pleaseexplaintheirmostimportantfeatures.
6. WhatwouldyousayaretheProsandConsofusingawebbaseddashboardversusanexecutablepackagepreinstalledonaPC?
7. Doyouknowofanyresourcesthatyouthinkwouldhelpusout?Ifso,whatarethey?
1Notallquestionswillbeasked.Dependingontheexpertise/experienceofthepersonandtime,questionsmaybeomittedduringtheinterviewastherelevanceoftheintervieweevariesandastimebecomesalimitingfactor.
24
EconomyModelingrelatedInterviewee1. Howusefulwouldyouforeseeanaccessiblerepositoryofeconomicmodels?2. Whateconomicmodelswouldyouseeusefultoputintothisrepository?3. Wouldtheimplementationofsucharepositorybeusefulinteaching/research
environments?Why?4. Haveyoueverworkedwithateamtocreatearepositoryorapplicationliketheone
wearebeingaskedtodevelop?i. Ifso,canyouelaborateonthatprocess?
5. Whatarethecurrentalternativestocreatingeconomicprojections/forecastsotherthanusingsoftware?
6. Whatmetrics/algorithms/methodshaveyoufoundhelpfulforcreatingaccurateeconomicprojections/forecasts?
7. Whattypeofvisualaidsdoyoufindtheeasiesttounderstandwhenusingcomputersoftware?
8. AsComputerSciencemajors,wedonothaveastrongbackgroundineconomics.Canyoudirectustoanyresources(papers,journals,books,etc.)thatmayhelpusdevelopabasicunderstandingofeconomicmodelingandforecasting?
WrapUpQuestions(appliestoallinterviewees)1. Haveyouhadanylanguagebarrierissuesinanyprojectsyouhaveparticipated/supervised
in?i. Ifso,whatwerethetacticsusedtogetaroundanycommunicationissues?
2. Doyouhaveanyadvicethatyoufeelwouldapplytoourproject?TechnicalQuestions
1. Reiterate:canweusethisinformationasacitationinourdocumentation?Wewillcitethedateandtimeofthisinterview,aswellastheinterviewee’snameandtitle.i. Iftheanswerisyes,telltheintervieweethatwewillcontactthembeforewe
releaseanydocumentationandaskifthecontentwrittenwithregardstotheirinterviewisanaccuratereflectionoftheirintentions/opinions.
2. Wouldyouliketobenotifiedaboutourprogress?i. *Ifallowedbysponsortoshareresearch:Wearedevelopingawebsitethat
willdocumentourjourneyinMoscowwithandwithoutrelationtothisproject.Ifyouwishtobenotified,wecanaddyouremailtothelistwhenthereisupdatestothewebsite.
ii. *Ifnotallowedbysponsortoshareresearch:Wecanshareourfinishedprojectproposalattheconclusionoftheterm.Afterwards,oursponsorwishestokeepourresearchprivate.
Ø Thanktheintervieweeagainfortakingthetimetomeetwithus,andexchangecontactinformationsowecancontinuecorrespondenceifnecessaryandsendthemtheinterviewnotesifneeded.
25
Appendix C: Dashboard Questionnaire
• GeneralQuestions:o Ageo Field/backgroundo Areyouaninteractive/hands-onlearner?Ordoyouprefertolearnthrough
textbooksorpapers?o Wouldyoubeinterestedinaplatformthatallowsyoutointeractand
experimentwithpre-codedeconomicmodels?o Wouldyoubeinterestedinaplatforminwhichproblemsareposted,and
usersareabletosubmittheirownsolutionsusingtheeconomicmodels?Thisplatformwillallowemployersandprofessorstoseestudent’sworkcapabilities,andwouldpotentiallyleadtoscholarshipgrantsandevenjoboffers.
• TechnicalQuestions:o Areyoucomfortableusingthecommandlinetointeractwithcode?o Wouldyoupreferacolorfulandcompletedashboardwithvisualizations?Or
asimpleinput/outputofcalculations?
26
Appendix D: Model Descriptions World Trade Model TheWorldTradeModelevaluatesindicatorsforcountriesoveraspecifiedtimeperiod,andoutputsafilethatgivesdetailedanalysisoneachpossiblecountrypossible.Theworldtrademodeltakesintwodatasetsasaninput.Thedatathatisanalyzedwasgatheredfromtwomainsources,UNComtradeandWorldBankOpenData.
Thealgorithmitselfusesthedatamentionedabove,andthencomputesindicatorsforeachcountryincludedinthedata.Keepinmindthatthetwodatasetshaveadifferentamountofcountries,soweonlyusecommoncountriesbetweenthem.Eachcountryhasthefollowingindicators:Alloftheindicatorsarecompiledintoacsvfile.
Figure10:CLIInterfaceforWorldTradeModel
TheCLI(commandlineinterface)takesinadesirednumber,andthencallscertainfunctionsdependingonthequery.ThesefunctionsuseaDataFrameconstructedfromthedata,andthencalculatestherankstoreturntotheuser.
27
Figure11:Calledfunctionscorrespondingtopossiblequeries
Thesefunctionswillreturnthedesiredresults,andtheCLIwilldisplaythecorrectdata,andthenremainonstandbytoreceivefurtherqueries.
Figure12:Queryfortop5countriesin2003,accordingtoTradeIndex
OnceauserisdonewiththeCLI,theycanenter0toterminatethecode,andalocalcopyoftheindexcalculationswillbesavedinthesamedirectoryasthepythoncode.This
28
fileiscalledindexData(ordered),andhasthecountrieslistedtopdownaccordingtotheirrankthroughtheyearsoverall.
Table4:SnippetofOutputCSVforWorldTradeModel
Country 2000 2001 2002 2003
China 0.03506 0.038745 0.044938 0.051997
EU-28 0.109885 0.115268 0.118873 0.119441
USA 0.109784 0.106438 0.095677 0.085858
Germany 0.077324 0.083203 0.085019 0.088815
Japan 0.067429 0.058729 0.057516 0.056005
Rep. of Korea 0.024236 0.021904 0.022423 0.022997
29
Cobb-Douglas Production Function Model TheCobb–Douglasmodelisaproductionfunction,widelyusedtorepresentthetechnologicalrelationshipbetweentheamountsoftwoormoreinputs,particularlyphysicalcapitalandlabor,andtheamountofoutputthatcanbeproducedbythoseinputs.Thismodelwaschosenbecauseitisthemostversatilemodeltorepresentwhatacompanywouldneedtodoinordertosignificantlyaffecttheirprofits,marketingstrategiesforcertainproducts,orotherservicestheyseektoanalyzeTheinputdataofthemodelcanbeseeninTable5below.
Table5:TheinputdatafortheCobb-DouglasProductionFunctionModel
Withthisdata,youcancalculatetheregressionmodel(Figure13below).
Figure13:Regressionresultsona3Dsurface
30
Portfolio Model Theportfoliomodeltakesadatasetofportfoliosasaninputandgeneratesthe
optimalweightsforeachassetthatproducethemostprofitableportfolio(basedonhistoricalreturnoverrisk).
Table6:InputDatasetofAssets’ReturnValues
ASSET RETURN VALUE
Gazprom 0.0114 -0.0037 -0.0335 0.0164 -0.0168 …
Aeroflot -0.0311 -0.0073 -0.0368 0.0229 0.0329 …
Sberbank 0.0090 -0.0206 -0.0601 0.0075 -0.0023 …
Nornickel 0.0183 -0.0105 -0.0434 -0.0055 0.0071 …
Mechel 0.0024 -0.0007 -0.0421 0.0022 0.0017 …
Thismodelfirsttakesina.csvfile(exampleaboveinTable6)consistingofreturn
valuesoffivedifferentassetsbaseondifferentobservationtimes.Theuseristhenpromptedtochoosebetweentwocalculationmethods–MarkowitzOptimizationandTangencyPortfolioOptimization.ThecalculationsforthesetwooptimizationmethodscanbeseeninFigure14below.
Figure14:Formulasforoptimizationcalculations
Themodelthenassignseachassetarandomweightandcalculatesthemeanandrisk(whichisdeterminedbythestandarddeviation.Forallnormalrandomizedportfolio,theyhavesimilarmeanandstandarddeviationdistribution,asaresulttheyhavethesameefficientfrontier(alineofhighestreturnforeachrisk).InFigure15,theyellowlineshows
31
theefficientfrontieroflargerandomportfolios;becausetheexampledatasetonlyconsistsoffiveassetsandtenobservationsforeachasset,thescatterplotonlyhaveasmallpartoverlappingtheefficientfrontier.
Figure15:Efficientfrontierofrandomlygeneratedportfolios
Thefinalresultofthemodelistogetthemostoptimalweightthatcangeneratethemostreturnandleastrisk.Theseoptimizedportfoliosarerepresentedbytheyellowpointsonthescatterplot.TheoptimizationtoolthismodelusesisConvexOptimization(CvxOpt).ThiscanbeseenbelowinFigure16.
Figure16:FormatforCvxOptPythonfunction
Finally,thisfunctioniscalledinthemainfunction(Figure17)toproducetheoptimalweights(Figure18).
32
Figure17:MainfunctionofPortfolioOptimization
Figure18:Finaloutputofoptimizedweights
33
Scoring Model Thescoringmodelingeneralneedsdatathatcangivethemostdetailsonacustomer.Thiscanincludedetailslikeage,creditscore,currentloans,andassets.Thisdatawillbeusedinanalgorithmthatwilldecideifthecustomerisvalidforarequestedloanthroughamachinelearningprocess.Thismachinelearningprocesswillusetrainingdatatotunethemachinecorrectly.Thiswillinturnmakeaalgorithmthatcandecidetheloanapprovalstatusbasedoffofhistoricaldata,andcurrentdata.
ThespecificmachinelearningprocesswewillbeusingiscalledRandomForest.Thisalgorithmusesmultipledecisiontreesinordertocomeupwithaverdictforadecision.
Afterthemodelistrainedfromtestdata,itreturnsthevariablesfromthedataandhowmuchofanimpacttheyhaveondeterminingwhetherornotapotentialcustomerwouldpaybacktheirloan.Thisiscalledvariableimportance,andtheresultsfromtheinitialtrainingdatasetcanbeseeninFigure19.
Figure19:Graphofvariableimportance
Thismodel,throughtheuseofauserinterface,cantakeindataofapotentialloancustomeranddeterminewitharound77.7%accuracywhetherornottheywillpayofftheirloan.Thismodelisdesignedtohelpbanksmakemoreeducateddecisionsonwhotheyshouldgivealoanto.