资源描述:
《Applying Dynamic Fuzzy Model in Combination》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、ApplyingDynamicFuzzyModelinCombinationwithSupportVectorMachinetoExploreStockMarketDynamismDeng-YivChiuandPing-JieChenDepartmentofInformationManagement,ChungHuaUniversityHsin-chuCity,Taiwan300,R.O.C.{chiuden,m09310006}@chu.edu.twAbstract.Inthestudy,anewdynamicfuzz
2、ymodelisproposedincombinationwithsupportvectormachine(SVM)toexplorestockmarketdynamism.Thefuzzymodelintegratesvariousfactorswithinfluentialdegreeastheinputvariables,andthegeneticalgorithm(GA)adjuststheinfluentialdegreeofeachinputvariabledynamically.SVMthenservestop
3、redictstockmarketdynamisminthenextphase.Inthemeanwhile,themultiperiodexperimentmethodisdesignedtosimulatethevolatilityofstockmarket.Then,wecompareitwithothermethods.Themodelfromthestudydoesgeneratebetterresultsthanothers.1IntroductionStockmarketisacomplicatedandv
4、olatilesystemduetotoomanypossibleinfluentialfactors.Inthepaststudies,asaresult,dynamisminthestockmar-ketwasoftenconsideredasrandommovement.Nevertheless,accordingtotheresearchesintherecentyears,itisnotentirelyrandom.Instead,itishighlycomplicatedandvolatile[1].Manyf
5、actors,includingmacroeconomicvariablesandstockmarkettechnicalindicators,havebeenproventohaveacertainlevelofforecastcapabilityonstockmarketduringacertainperiodoftime[2].Inthepastdecade,variousmethodshavebeenwidelyappliedinthestockmarketforecastsuchaslinearandnonli
6、nearmathematicalmodelsormulti-agentmech-anism[3]tosimulatethepotentialstockmarkettransactionmechanism,suchasartificialneuralnetwork(ANN)ofmultiplelayersofthresholdnonlinearfunction.Becauseoftheadvantagesofarbitraryfunctionapproximationandneedlessofstatisticsassump
7、tion,ANNiswidelyappliedinthesimulationofpotentialmar-kettransactionmechanism[4].Also,toimprovetheforecastperformance,somemachinelearningmethodsareapplied.Forexample,geneticalgorithm(GA)isusedtoreduceinputfeaturedimensionandselectbettermodelparameters[5]toincrease
8、theforecastaccuracyrate.Supportvectormachine(SVM)isanewlydevelopedmathematicalmodelwithoutstandingperformancesinhandlinghighdimensionentryspaceproblems.Suchafe