基于人工神经网络预测尼日利亚股市(IJISA-V4-N11-8)

I.J.IntelligentSystemsandApplications,2012,11,68-74PublishedOnlineOctober2012inMECS()DOI:10.5815/ijisa.2012.11.08Copyright©2012MECSI.J.IntelligentSystemsandApplications,2012,11,68-74PredictionofStockMarketinNigeriaUsingArtificialNeuralNetworkPeterAdebayoIdowu,ChrisOsakwe,AderonkeAnthoniaKayode,EmmanuelRotimiAdagunodoDept.ofComputerScienceandEngineering,ObafemiAwolowoUniversity,Ile-Ife,[e-mail];[e-mail];[e-mail];[e-mail]—PredictionofNigerianstockmarketisalmostnotdonebyanyresearcherandisanimportantfactorwhichcanbeusedtodeterminetheviabilityofNigerianstockmarket.Inthispaper,thepredictionmodelsweredevelopedusingArtificialNeuralNetwork.TheresultofthepredictionofNigerianStockExchange(NSE)marketindexvalueofselectedbanksusingArtificialNeuralNetworkwaspresented.Themulti-layerfeedforwardneuralnetworkwasused,sothateachoutputunitistoldwhatitsdesiredresponsetoinputsignalsoughttobe.Thisworkhasconfirmedthefactthatartificialneuralnetworkcanbeusedtopredictfuturestockprices.Thedatacollectionperiodisfrom2003to2006.IndexTerms—ArtificialNeuralNetwork,Prediction,NigerianStockExchange,InputSignalI.IntroductionParticlePredictioninfinanceespeciallyinstockmarketisanimportantissue.ArtificialNeuralNetworkthesedaysarebeingusedtopredictinfinancialsector.Artificialneuralnetworksareparalleldistributedinformationprocessingmodelsthatcanrecognizehighlycomplexpatternswithinavailabledata..Itisaninformationprocessingsystemthathascertainperformancecharacteristicsincommonwithbiologicalneuralnetworksandtherefore,eachnetworkisacollectionofneuronsthatarearrangedinspecificformations.Thebasicelementsofneuralnetworkcompriseneuronsandtheirconnectionstrengths(weights).ANNshavebeenfoundtohaveasoundtheoreticalbasisfromtheperspectiveofstatisticallearningtheoryaswellasmachinelearning,andusuallyyieldgoodperformancewhenusedforreal-worlddataanalysis[1].Apartfromthat,ANNshavegoodgeneralizationcapabilitiesandareusuallyrobustagainstnoisyormissingdata.AlltheseandmoremakeANNhighlydesirableinthepredictionoftheStockMarket.AStockMarketmaybedefinedasamarketwhereshares,stocks,governmentbonds,debentures,andotherapprovedsecuritiesaresoldandbought[2].Itis,thus,amarketwherelargeandsmallinvestorsalikebuyandsellthroughstockbrokers,thestocks(orshares)ofcompaniesandgovernmentagencies.Theideaofstockmarketpredictionisnotnew;ofcourse,businesspeopleoftenattempttoanticipatethemarketbyinterpretingexternalparameters,suchaseconomicindicators,publicopinion,andcurrentpoliticalclimate.Beforetheageofcomputers,peopletradedstocksandcommoditiesprimarilyonintuition.Asthelevelofinvestingandtradinggrew,peoplesearchedfortoolsandmethodsthatwouldincreasetheirgainswhileminimizingtheirrisk.TheNigerianstockmarketisbylawprincipallymanagedandcontrolledbytheNigerianStockExchangewhichwasestablishedin1960throughtheActsofParliamentandstartedoperationsin1961with19securitieslistedfortrading.Today,therearemorethan260securitieslistedontheExchange.Mostofthelistedcompanieshavemultinationalaffiliationsandrepresentacross-sectionoftheeconomy,rangingfromagriculturethroughmanufacturingtoservices.ThepublictrustintheNSEhasgrowntremendously,withabout3millionindividualinvestorsand100sofinstitutionalinvestors,usingthefacilitiesoftheExchange.Amajorchallengeposedatanystockinvestoristheabilitytopredictstockprices;thisisagreatconcerntobothinstitutionalandindividualinvestors.ModelingtechniquessuchasANNisusedtobuildpredictivemodels,theadvantageofartificialneuralnetworkoverconventionalprogrammingliesintheirabilitytosolveproblemsthatdonothaveanalgorithmicsolutionortheavailablesolutionistoocomplextobefound.Neuralnetworksarewellsuitedtotackleproblemsthatpeoplearegoodatsolving,likepredictionandpatternrecognition.Hence,thegrowingneedandlackofaveritablecumreliabletoolforpredictionofthestockmarketserveasimpetusforthisthesis.Therefore,thiscouldprovetobeavitaltoolforinvestmentdecisionmakingsinceitisexpectedtoassistinvestorsinmakingbetterandqualitydecisionsandmanagingtheirstocksmuchmoreprofitably.However,thedifficultyinidentifyinggoodrawdata,pre-processingthisdata,traininganartificialneuralnetworkandrepeatingthisprocessuntilagoodmodelisdevelopedshouldnotbediscounted.PredictingstockmarketdatahasproventobedifficultduetoerraticandunpredictablemarketPredictionofStockMarketinNigeriaUsingArtificialNeuralNetwork69Copyright©2012MECSI.J.IntelligentSystemsandApplications,2012,11,68-74behavior.Thestockmarketischaracterizedbysomanyfluctuationssuchaspolitical(orindustrial)climate,governmentpolicies,economicindicators,etcandmainlyasaconsequenceoftheclosetorandom–walkbehaviorofastocktimeseries.AllthesehavemadeitprettyarduoustopredictaccuratefuturetrendsintheStockMarketspecificallypriceofstocks.InNigeria,therehasbeenalmostnoresearchthatdealswithpredictionofNigerianstockmarketvalue.Inthispaper,wepresenttheresultofNigerianStockExchange(NSE)marketindexvaluespredictionusingANN.II.ArtificialNeuralNetwork:AnOverviewArtificialNeuralNetwork(ANN)grewoutofresearchinArtificialInt