Support vector machine as an efficient framework for stock market volatility forecasting

《Support vector machine as an efficient framework for stock market volatility forecasting 》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、CMS(2006)3:147160DOI10.1007/s10287-005-0005-5ORIGINALPAPERSupportvectormachineasanefficientframeworkforstockmarketvolatilityforecastingValeriyV.Gavrishchaka·SupriyaBanerjeePublishedonline:16February2006©Springer2006AbstractAdvantagesandlimitationsoftheexistingmodelsfor

2、practicalforecastingofstockmarketvolatilityhavebeenidentified.Supportvectormachine(SVM)havebeenpro-posedasacomplimentaryvolatilitymodelthatiscapabletoextractinformationfrommulti-scaleandhigh-dimensionalmarketdata.PresentedresultsforSP500indexsuggestthatSVMcanefficientl

3、yworkwithhigh-dimensionalinputstoaccountforvolatilitylong-memoryandmultiscaleeffectsandisoftensuperiortothemain-streamvolatilitymodels.SVM-basedframeworkforvolatilityforecastingisexpectedtobeimportantinthedevelopmentofthenovelstrategiesforvolatilitytrading,advancedris

4、kmanagementsystems,andotherappli-cationsdealingwithmulti-scaleandhigh-dimensionalmarketdata.1.IntroductionAvailabilityofhigh-resolutionandmulti-sourcedataincreasesinmanyfieldsofpracticalinterestincludingfinancialindustry.However,itiswell-knownthatthemajorityofadvancedst

5、atisticalandmachinelearningalgorithms,includingneuralnetworks(NN),canencoun-terasetofproblemscalleddimensionalitycursewhenappliedtohigh-dimensionaldata(Bishop1995).Nonstationarityofthetimeseriescanalsoimposesignificantlimitationsondataavailablefortrainingthatoftenleads

6、topoorgeneralizationabilityofthemodel.Thelatterfeatureisespeciallyrelevantforfinancialapplications.Apromisingalgorithmthatcantoleratehigh-dimensionalandincompletedataissupportvectormachine(SVM)(Vapnik1995,1998).SVMshaverecentlybeenreceivingsignificantinterestduetoexcell

7、entresultsinvariousapplications(CristianiniandShawe-Taylor2000).SVMcombinesthetrainingefficiencyandsimplicityoflinearalgorithmswiththeaccuracyofthebestnonlineartechniquesaswellassystematicapproachforoptimalgeneraliza-tion.InmanypracticalapplicationsSVMscantoleratehigh-

8、dimensionaland/orincompletedataandoftendemonstrateperformancessuperiortothebestavailabletechniquesinclud-ingclassicalNNs(Cri