資源描述：
《stock prediction-a neural network approach》由會(huì)員上傳分享，免費(fèi)在線閱讀，更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫。
1、StockPrediction–ANeuralNetworkApproachKarlNygrenkarlnyg@kth.se28thMarch2004MasterThesisRoyalInstituteofTechnology,KTHSupervisor:Prof.KennethHolmstr¨omExaminer:Dr.TorkelErhardssonAbstractPredictingstockdatawithtraditionaltimeseriesanalysishasproventobedif-?cult.Ana
2、rti?cialneuralnetworkmaybemoresuitableforthetask.Primarilybecausenoassumptionaboutasuitablemathematicalmodelhastobemadepriortoforecasting.Furthermore,aneuralnetworkhastheabilitytoextractusefulinformationfromlargesetsofdata,whichoftenisrequiredforasatisfyingdescrip
3、tionofa?nancialtimeseries.Thisthesisbeginswithareviewofthetheoreticalbackgroundofneuralnet-works.SubsequentlyanErrorCorrectionNeuralNetwork(ECNN)isde?nedandimplementedforanempiricalstudy.Technicalaswellasfundamentaldataareusedasinputtothenetwork.One-stepreturnsoft
4、heSwedishstockindexandtwomajorstocksoftheSwedishstockexchangearepredictedusingtwoseparatenetworkstructures.DailypredictionsareperformedonastandardECNNwhereasanextensionoftheECNNisusedforweeklypredictions.Inbenchmarkcomparisons,theindexpredictionprovestobesuccessfu
5、l.Theresultsonthestocksarelessconvincing,neverthelessthenetworkoutperformsthenaivestrategy.SammanfattningAttpredikterab¨orsdatamedtraditionelltidsserieanalysharvisatsigvarasv?art.Ettarti?cielltneuraltn¨atverkkanvaramerpassandef¨oruppgiften.Fr¨amstd¨arf¨orattingaan
6、tagandenomenpassandematematiskmodellm?asteg¨orasinnanprediktering.Vidareharettneuraltn¨atverkf¨orm?aganattextraheraanv¨andbarinformationfr?anstoradatam¨angder,vilketofta¨arn¨odv¨andigtf¨orentillfredsst¨allandebeskrivningaven?nansielltidsserie.Deth¨arexamensarbetet
7、b¨orjarmedengenomg?angavteorinbakomneuralan¨atverk.D¨arefterde?nierasochimplementerasettfelkorrigerandeneuraltn¨atverk(ECNN)f¨orenempiriskstudie.B?adetekniska-ochfundamentaladataanv¨andssomindatatilln¨atverket.Enstegsavkastningarf¨orGeneralindexsamttv?astoraaktier
8、p?aStockholmsb¨orsenpredikterasmedtv?aseparatan¨atverks-strukturer.Dagligaprediktionerutf¨orsp?aenstandardECNNmedanenut¨okadvariantavECNNanv¨andsf¨orvec
