Støtte Vector Machines Financial Applications. Listed i rekkefølge av citater per år, høyest på toppen. Last oppdatert september 2006.PANG, Bo, Lillian LEE og Shivakumar VAITHYANATHAN, 2002 Tommelen opp Sentiment Klassifisering ved hjelp av maskinlæringsteknikker i EMNLP 02 Proceedings of the ACL -02 Konferanse om empiriske metoder i naturlig språkbehandling - Volum 10 sider 79--86 Sitat av 154 36 66 år Sammendrag Vi vurderer problemet med å klassifisere dokumenter ikke etter emne, men ved overordnet følelse, for eksempel å avgjøre om en vurdering er positiv eller negativ Ved hjelp av filmanmeldelser som data, finner vi at standard maskininnlæringsteknikker definitivt overgår de menneskeskapte baselinjene. De tre maskinlæringsmetodene vi ansatte Naive Bayes, maksimal entropiklassifikasjon og støttevektormaskiner virker imidlertid ikke så godt på følelsesklassifisering som på tradisjonelle emnebasert kategorisering Vi konkluderer med å undersøke faktorer som gjør følelsesklassifiseringsproblemet mor e utfordrende. fastslått at ved hjelp av filmomtaler som data var standard maskininnlæringsteknikker definitivt bedre enn menneskeskapte baselinjer. De fant imidlertid også at de tre maskinlæringsmetodene de anvendte Naive Bayes, maksimal entropiklassifikasjon og støttevektormaskiner ikke utførte som bra på sentiment klassifisering som på tradisjonell emne-basert kategorisering. VAN GESTEL, Tony, et al 2001 Financial Time Series Prediction ved hjelp av minste kvadrater Støtte vektormaskiner innenfor bevisrammen IEEE-transaksjoner på nevrale nettverk Volum 12, nummer 4, juli 2001, sider 809 -821 Sitat av 77 14 82 år Sammendrag Den bayesiske bevisrammen brukes i dette papiret til minst kvadrater støttevektormaskin LS-SVM-regresjon for å utlede ikke-lineære modeller for å forutsi en finansiell tidsserie og den tilhørende volatiliteten På første nivå av innledning , er et statistisk ramme knyttet til LS-SVM formuleringen som tillater en å inkludere den tidsvarierende vol atilitet av markedet ved et passende valg av flere hyperparametere. Hyparametrene til modellen er utledet på andre nivå av inngripen. De utledte hyperparametrene, som er relatert til volatiliteten, brukes til å konstruere en volatilitetsmodell innenfor bevisrammen Modell sammenligning utføres på det tredje nivået av inngang for å automatisk justere parametrene til kjernen funksjonen og å velge de relevante inngangene LS-SVM formuleringen tillater en å utlede analytiske uttrykk i funksjonen plass og praktiske uttrykk er oppnådd i dual plass som erstatter det indre produktet ved den tilhørende kjernefunksjonen ved bruk av Mercer s teorem. Prinsippene for forutsigelse av den ukentlige 90-dagers regningstakten og de daglige DAX30-sluttkursene viser at betydelige utfall av prognoseforsøk kan være laget med respekt for pesaran-timmerman test statistikk. anvendt den bayesiske bevis rammeverket til minst kvadratene støtte vecto r maskinens LS-SVM regresjon for å forutsi den ukentlige 90-dagers regningsfrekvensen og de daglige DAX30 sluttkursene. TAY, Francis EH og Lijuan CAO, 2001 Søknad av støttevektormaskiner i økonomisk tidsserien prognoser Omega The International Journal of Management Science Volum 29, Utgave 4, august 2001, Sider 309-317 Sitat av 67 12 89 år Sammendrag Dette dokumentet omhandler bruk av en ny nevralnettteknikk, støttevektormaskin SVM, i prognoser for økonomisk tidsserier Formålet med dette papiret er å undersøke muligheten for SVM i finansiell tidsserieprognose ved å sammenligne den med et flerlags back-propagation BP-neuralt nettverk Fem real futures-kontrakter som samles fra Chicago Mercantile Market, brukes som datasettene. Eksperimentet viser at SVM overgår BP neuralt nettverk basert på kriteriene for normalisert gjennomsnittlig kvadratfeil NMSE, gjennomsnittlig absolutt feil MAE, retningssymmetri DS og vektet retningssymmetri WDS Siden det er n o strukturert måte å velge de frie parametrene til SVMs, er variabiliteten i ytelse med hensyn til de frie parametrene undersøkt i denne studien. Analysen av de eksperimentelle resultatene viste at det er fordelaktig å anvende SVMs til å prognostisere økonomiske tidsserier. funnet at en SVM overgikk et flerlags back-propagation BP-neuralt nettverk på fem reelle futureskontrakter fra Chicago Mercantile Market. TAY, Francis EH og LJ CAO, 2002 Modifiserte støttevektormaskiner i økonomisk tidsserieprognose Neurocomputing Volume 48, Issues 1-4, October 2002 , Sider 847-861 Sitat av 54 12 86 år Sammendrag I dette papiret foreslås en modifisert versjon av støttevektormaskiner, kalt C-encoding-støttevektormaskin, for å modellere ikke-stationære økonomiske tidsserier. enkel modifikasjon av den regulerte risikofunksjonen i støttevektormaskiner, hvorved de nylig-følsomme feilene blir straffet mer enn den fjerne - i usensitive feil Denne prosedyren er basert på forkunnskapen om at i de ikke-stasjonære økonomiske tidsseriene endres avhengigheten mellom inngangsvariabler og utgangsvariabler gradvis over tid, spesielt de siste siste dataene kunne gi mer viktig informasjon enn de fjerne tidligere dataene i eksperimentet, C-scanning-støttevektormaskiner blir testet ved hjelp av tre virkelige futures samlet fra Chicago Mercantile Market. Det er vist at C - ascending-støttevektormaskiner med de faktisk bestilte prøvedataene konsekvent prognostiserer bedre enn standard-støttevektormaskiner, med verste ytelse når de omvendt bestilte eksempeldataene brukes. Videre bruker C-skannende støttevektormaskiner mindre støttevektorer enn de av standard-støttevektormaskiner, noe som resulterer i en sparsomere representasjon av løsningen. utviklede C-skannende støttevektormaskiner, som straffer Nylig - sensitive feil er tyngre enn fjernt-sensensit Ive feil, og fant at de prognostiserer bedre enn standard SVM på tre reelle futures samlet fra Chicago Mercantile Market. HUANG, Zan, et al. 2004 Kredittvurdering analyse med støttevektorer og nevrale nettverk, en markedskomparativ studie Decision Support Systems Volume 37, Utgave 4 september 2004, Sider 543-558 Sitat 21 21 55 år Sammendrag Kvalitetsvurdering av kredittvurdering har tiltrukket seg mange forskningsinteresser i litteraturen Nyere studier har vist at Artificial Intelligence AI-metoder oppnådde bedre ytelse enn tradisjonelle statistiske metoder. Denne artikkelen introduserer en relativ ny maskininnlæringsteknikk, støtte vektormaskiner SVM, til problemet i forsøk på å gi en modell med bedre forklarende kraft Vi brukte backpropagation neuralt nettverk BNN som referanse og oppnådde prediksjonsnøyaktighet rundt 80 for både BNN og SVM-metoder for USA og Taiwan markeder Imidlertid ble det bare observert liten forbedring av SVM. En annen retning av forskningen er å forbedre tolkbarheten av de AI-baserte modellene. Vi brukte nyere forskningsresultater i tolkning av nettverksmodell og oppnådd relativ betydning av de inngående økonomiske variablene fra de neurale nettverksmodellene. På grunnlag av disse resultatene gjennomførte vi en markedskomparativ analyse på forskjellene i å bestemme faktorer i USA og Taiwan markets. applied backpropagation neurale nettverk og SVMs til bedrifts kredittvurdering prediksjon for USA og Taiwan markeder og fant ut at resultatene var sammenlignbare begge var overlegne for logistisk regresjon, med SVM litt bedre. CAO, Lijuan, 2003 Støttevektormaskiner eksperter til prognoser for tidsserier Neurokomputering Volum 51, april 2003, Sider 321-339 Sitat av 29 9 08 år Sammendrag I dette papiret legges det til grunn ved hjelp av støtteteknormaskiner SVMs eksperter for prognoser for tidsserier De generaliserte SVMs-eksperter har en to-trinns nevrale nettverksarkitektur I første etappe, selvorganisk Zing-funksjonskart SOM brukes som en klyngalgoritme for å partisjonere hele inngangsrommet til flere ubundne områder. En trestrukturert arkitektur er adoptert i partisjonen for å unngå problemet med å forhåndsdefinere antall partisjonerte områder. I andre trinn blir flere SVMs , også kalt SVM-eksperter, er de best passende partisjonerte områdene konstruert ved å finne den mest hensiktsmessige kjernefunksjonen og de optimale frie parametrene til SVMs. Sunspot-dataene, Santa Fe datasettene A, C og D, og de to bygningsdatasettene blir evaluert i eksperimentet Simuleringen viser at SVMs-eksperter oppnår betydelig forbedring i generaliseringsytelsen sammenlignet med de enkelte SVM-modellene. I tillegg samler SVMs-ekspertene seg raskere og bruker færre støttevektorer. viste at deres metode for SVM-eksperter oppnådde betydelig forbedring over single SVMs-modeller når det legges til Santa Fe datasett C høyfrekvente valutakurser mellom sveitsiske franc og t han amerikanske dollar. KIM, Kyoung-jae, 2003 Finansiell tidsserien prognoser ved hjelp av støttevektorgaskiner Neurokomputering Volum 55, Utgavene 1-2 september 2003, Sider 307-319 Sitat av 28 8 76 år Sammendrag Støttevektormaskiner SVMs er lovende metoder for forutsigelse av økonomiske tidsserier fordi de bruker en risikofunksjon bestående av empirisk feil og en regulert periode som er avledet fra strukturell risikominimering. Denne studien gjelder SVM for å forutsi aksjekursindeksen. I tillegg undersøker denne studien muligheten for å søke SVM i finansiell prognose ved å sammenligne det med tilbaketransplantasjonens neurale nettverk og saksbasert resonnement. De eksperimentelle resultatene viser at SVM gir et lovende alternativ til aksjemarkedsprognosering. Fastslår at SVM har overgått tilbakegående neurale nettverk og saksbasert resonnement når det brukes til prognose den daglige Korea sammensatte aksjekursindeksen KOSPI. SHIN Kyung-Shik, Taik Soo LEE og Hyun-jung KIM, 2005 En applikasjon på støttevektormaskiner i konkursprediksjonsmodell Ekspertsystemer med applikasjoner Volum 28, Utgave 1, Januar 2005, Sider 127-135 Sitat av 8 6 67 år Sammendrag Denne studien undersøker effekten av å bruke støttevektormaskiner SVM til konkursprosjektproblem Selv om det er et velkjent faktum at det propagasjonale nettverket BPN fungerer godt i mønstergenkjenningsoppgaver, har metoden noen begrensninger ved at det er en kunst for å finne en passende modellstruktur og optimal løsning. Dessuten legger du inn så mange av treningssettet Som mulig inn i nettverket er nødvendig for å søke i nettets vekt. På den annen side, siden SVM fanger geometriske egenskaper av funksjonsplass uten å utlede nettvekter fra treningsdataene, er det i stand til å utvinne den optimale løsningen med det lille treningssettet størrelse I denne studien viser vi at den foreslåtte klassifiseringen av SVM-tilnærmingen overgår BPN til problemet med selskaps konkursprediksjon Resultatene viser at SVM-nøyaktigheten og generaliseringsprestasjonen er bedre enn BPN, da treningssettets størrelse blir mindre. Vi undersøker også effekten av variabiliteten i ytelse med hensyn til ulike verdier av parametere i SVM. I tillegg undersøker og oppsummerer vi de flere overlegne punktene i SVM-algoritmen sammenlignet med BPN. demonstrated at SVMs utfører bedre enn tilbakeproduserende neurale nettverk når de brukes på selskaps konkursforutsigelse. CAO, LJ og Francis EH TAY, 2003 Støttevektormaskin med adaptive parametre i økonomisk tidsserieprognoser IEEE Transaksjoner på Neural Networks Volume 14, Issue 6, November 2003, Sider 1506-1518 Cited by 20 6 25 år Abstract En ny type læringsmaskin kalt supportvektormaskin SVM har mottatt økende interesse for områder som spenner fra sin opprinnelige applikasjon i mønster anerkjennelse til andre applikasjoner som regresjonsestimat på grunn av sin bemerkelsesverdige generalisering perfo rmance Dette papiret omhandler anvendelsen av SVM i økonomisk tidsserieprognoser. Muligheten til å anvende SVM i finansiell prognose blir først undersøkt ved å sammenligne den med flerlags-back-propagation BP-neuralt nettverk og den regulerte radiale basisfunksjonen RBF neurale nettverket. Variabiliteten i ytelse av SVM med hensyn til de frie parametrene blir undersøkt eksperimentelt Adaptive parametere blir da foreslått ved å inkorporere ikke-stationaritet i finansielle tidsserier i SVM. Fem virkelige futureskontrakter samlet fra Chicago Mercantile Market brukes som datasett. Simuleringen viser at blant de tre metodene SVM overpresterer BP-neurale nettverket i økonomisk prognoser og det er sammenlignbare generaliseringsprestasjoner mellom SVM og det regulerte RBF-neurale nettverket. Dessuten har de frie parametrene til SVM en stor effekt på generaliseringsytelsen SVM med adaptive parametre kan begge oppnå høyere generaliseringsytelseog bruker færre støttevektorer enn standard SVM i økonomisk prognose. brukte et SVM, et flerlags back-propagation BP-neuralt nettverk og et regulert radialbasert funksjon RBF neuralt nettverk for å forutsi fem virkelige futureskontrakter samlet fra Chicago Mercantile Market Results viste at SVM og det regulerte RBF neurale nettverket var sammenlignbare og begge overgikk det BP-neurale nettverket. CAO, Lijuan og Francis EH TAY, 2001 Finansiell prognose ved hjelp av vektorgravere for vektorgraver Neural Computing Applications Volum 10, nummer 2 mai 2001, sider 184-192 Sitat av 26 5 00 år Sammendrag Bruken av Support Vector Machines SVMs er studert i økonomisk prognose ved å sammenligne den med en flerlags perceptron opplært av Back Propagation BP algoritmen SVMs prognose bedre enn BP basert på kriteriene for Normalized Mean Square Error NMSE, Mean Absolute Feil MAE, Directional Symmetry DS, Korrigere CP-trend og korrigere ned-CD-trend SP 500 daglig prisindeks brukes som d ata sett Siden det ikke er strukturert måte å velge de frie parametrene til SVM, undersøkes generaliseringsfeilen med hensyn til de frie parametrene til SVMs i dette eksperimentet. Som vist i eksperimentet har de liten innvirkning på løsningen Analyse av eksperimentelle resultater viser at det er fordelaktig å bruke SVMs til å prognostisere den finansielle tidsserien. fastslår at SVMs anslår SP 500 daglig prisindeks bedre enn en flerlags perceptron utdannet av Back Propagation BP algoritmen. MIN, Jae H og Young-Chan LEE, 2005 Konkursforutsigelse ved hjelp av støttevektormaskin med optimal valg av kjernefunksjonsparametre Ekspertsystemer med applikasjoner Volum 28, Utgave 4, Mai 2005, Sider 603-614 Sitert av 6 5 00 år Sammendrag Konkursforutsigelse har trukket mange forskningsinteresser i tidligere litteratur , og nyere studier har vist at maskinlæringsteknikker oppnådde bedre ytelse enn tradisjonelle statistiske seg. Dette papiret gjelder støtte vektormaskiner SVMs til konkursprosjektproblemet i et forsøk på å foreslå en ny modell med bedre forklarende kraft og stabilitet For å betjene dette formål bruker vi en rutenettsøkteknikk ved hjelp av 5-talls kryss-validering for å finne ut de optimale parameterverdiene for kjernen funksjon av SVM I tillegg, for å evaluere prediksjonsnøyaktigheten til SVM, sammenligner vi ytelsen med de med multiple diskrimineringsanalyser MDA, logistisk regresjonsanalyse Logit og trelags fullt tilkoblet tilbakeproduserende nevrale nettverk BPNs Eksperimentresultatene viser at SVM overgår de andre metodene. Funnet at SVMs overpresterte flere diskrimineringsanalyser MDA, logistisk regresjonsanalyse Logit og trelags fullt tilkoblet back-propagation neurale nettverk BPNs. ABRAHAM, Ajith, Ninan Sajith PHILIP og P SARATCHANDRAN, 2003 Modeling kaotisk oppførsel av aksjeindekser ved hjelp av intelligente paradigmer Neural, Parallel Scientific Computations Volume 11, page s 143-160 Sitat av 10 4 55 år Sammendrag Bruken av intelligente systemer for aksjemarkedsprognoser har blitt etablert i stor grad I dette papiret undersøker vi hvordan aksjemarkedens tilsynelatende kaotiske oppførsel kunne være godt representert ved å bruke flere sammenhengende paradigmer og soft computing teknikker For å demonstrere ulike teknikker, vurderte vi Nasdaq-100-indeksen for Nasdaq Stock Market SM og S s Nasdaq 100 hovedindeksverdier og 4 år s NIFTY-indeksverdier. Dette papiret undersøker utviklingen av en pålitelig og effektiv teknikk for å modellere tilsynelatende kaotisk oppførsel av aksjemarkeder Vi vurderte et kunstig nevralt nettverk utdannet ved hjelp av Levenberg-Marquardt-algoritmen, Support Vector Machine SVM, Takagi-Sugeno neurofuzzy-modellen og et Difference Boosting Neural Network DBNN. Dette papiret forklarer kort hvordan de forskjellige forbindelsesparadigmaene kunne formuleres ved hjelp av ulike læringsmetoder og undersøker deretter om de kan gi det nødvendige nivået per formasjon som er tilstrekkelig bra og robust for å gi en pålitelig prognosemodell for aksjemarkedsindekser. Eksperimentresultater viser at alle sammenhengende paradigmer som vurderes kunne representere aksjeindeksens adferd svært nøyaktig. anvendt fire forskjellige teknikker, et kunstig nevralt nettverk utdannet ved hjelp av Levenberg-Marquardt-algoritmen, en støttevektormaskin, et differensforsterkende neuralt nettverk og et Takagi-Sugeno fuzzy inferencesystem lærte å bruke en neuronfuzzy modell for nevralnettverksalgoritmen til forutsigelsen av Nasdaq-100-indeksen på Nasdaq Stock Market og S - Nedadgående støttevektormaskiner for økonomisk tidsserieprognoser Neural Processing Letters 15 2 179-195 Sitat av 11 2 62 år Abstract Dette papiret foreslår en modifisert versjon av støttevektormaskiner SVMs, som kalles nedgående støttevektormaskiner-SDVM, for å modellere ikke-stillestående økonomiske tidsserier. SDVMene er oppnådd ved å inkorporere problemdomenekunnskapen ikke-stasjonær rity av finansielle tidsserier til SVMs I motsetning til standard SVMs som bruker et konstant rør i alle treningsdatapunktene, bruker SDVMene et adaptivt rør for å håndtere strukturendringene i eksperimentet viser at SDVMene generaliserer bedre enn standard SVMs i prognoser for ikke-stasjonære finansielle tidsserier En annen fordel ved denne modifikasjonen er at SDVMene samler seg til færre støttevektorer, noe som resulterer i en tynnere representasjon av løsningen. Inkorporert problemdomenekunnskapen om ikke-stationaritet av finansielle tidsserier til SVMer ved å bruke et tilpasningsrør i deres såkalte epsilon-nedadgående støttevektormaskiner epsilon-SDVMs Experiment viste at epsilon-SDVM generaliserer bedre enn standard SVMer ved prognostisering av ikke-stationære økonomiske tidsserier, og konvergerer også til færre støttevektorer, noe som resulterer i en tynnere representasjon av solution. DEBNATH, Sandip og C Lee GILES, 2005 En læringsbasert modell for overskriftens utvinning av nyhetsartikler t o Finn forklarende setninger for hendelser i K-CAP 05 Utførelser av den tredje internasjonale konferansen om kunnskapsopptak Sider 189-190 Citert av 2 1 67 år Sammendrag Metadatainformasjon spiller en avgjørende rolle for å øke dokumentorganiseringseffektiviteten og arkiverbarheten. Nyhets-metadata inneholder DateLine ByLine HeadLine og mange andre Vi fant ut at HeadLine-informasjon er nyttig for å gjette temaet for nyhetsartikkel. Spesielt for finansielle nyhetsartikler, fant vi at HeadLine kan være spesielt nyttig for å finne forklarende setninger for større hendelser som for eksempel betydelige endringer i aksjekursene i dette papiret undersøker vi en støttevektorbasert læringsmetode for å automatisk trekke ut HeadLine-metadataene. Vi finner at klassifikasjonsnøyaktigheten ved å finne HeadLine s forbedres dersom DateLine s blir identifisert først. Vi brukte deretter den ekstraherte HeadLine s for å starte et mønster samsvar med søkeord for å finne setningene som er ansvarlige for historietema Bruke dette temaet og en enkel språkmodell er det mulig å finne noen forklarende setninger for noen vesentlig prisendring. vis en ny tilnærming til å utvinne nyhetsmetadata HeadLines ved hjelp av SVMs og bruke dem til å finne historietemaer for å få en setningsbasert forklaring på en aksjekursendring. Fra GESTEL, Tony et al. 2003 En støttevektormaskin tilnærming til kredittvurdering Bank en Financiewezen Volum 2, mars, sider 73-82 Sitert av 5 1 56 år Sammendrag Drevet av behovet for å tildele kapital på en lønnsom måte og ved den nylig foreslåtte Basel II forskrifter, finansinstitusjoner blir mer og mer forpliktet til å bygge kredittvurderingsmodeller som vurderer risikoen for mislighold av sine kunder. Mange teknikker har blitt foreslått for å takle dette problemet. Support Vector Machines SVMs er en lovende ny teknikk som nylig har stått fra forskjellige domener som anvendt statistikk, nevrale nettverk og maskinlæring I dette papiret eksperimenterer vi med minst kvadrater støttevektormaskiner LS-SVMs, en nylig modifisert versjon av SVMs, og rapportere signifikant bedre resultater i motsetning til de klassiske teknologiske parat metodene, Ordinary Least Squares OLS, Ordinal Logistic Regression OLR, Multilayer Perceptron MLP og minst kvadrater støttevektormaskiner LS-SVMs når de brukes til kreditt scoring. SVM-metoden ga betydelige og konsekvent bedre resultater enn de klassiske lineære klassifiseringsmetodene. FAN, Alan og Marimuthu PALANISWAMI, 2000 Valg av konkursprediktorer ved hjelp av en støttevektormaskinens tilnærming IJCNN 2000 Forhandlinger av IEEE-INNS-ENNS internasjonale felleskonferanse om neurale nettverk, volum 6 redigerte av Shun-Ichi Amari et al side 6354 Sitat av 9 1 45 år Sammendrag Konvensjonell Neural Network tilnærming har blitt funnet nyttig for å forutsi bedriftsforstyrrelser fra regnskapet. I dette papiret har vi vedtatt en Support Vector Machine tilnærming til problemet. En ny måte å Valg av konkurspredikatorer vises ved hjelp av euklidisk avstand basert kriterium beregnet i SVM-kjernen En komparativ studie er gitt ved hjelp av tre klassiske bedriftsmodeller og en alternativ modell basert på SVM-tilnærmingen. Bruk SVMs til å velge konkursprotokoller og gi en komparativ studie. TID, Francis Eng Hock og Li Juan CAO , 2001 Forbedret økonomisk tidsserier prognoser ved å kombinere Support Vector Machines med selvorganiserende funksjonskart Intelligent Data Analyse Volum 5, Nummer 4, Sider 339-354 Sitat av 7 1 35 år Sammendrag En to-trins neural nettverksarkitektur konstruert ved å kombinere Support Vector Maskiner SVMer med selvorganiserende funksjonskart SOM er foreslått for prognostisering av økonomiske tidsserier I første trinn brukes SOM som en klyngalgoritme for å partisjonere hele inngangsrommet til flere ujevne områder. En trestrukturert arkitektur er adoptert i partisjonen for å unngå Problemet med å forhåndsbestille antall partisjonerte områder I andre fase, såkalte flere SVMs, også kalt SVM-eksperter ts som passer best for hver partisjonert region, er konstruert ved å finne den mest hensiktsmessige kjernefunksjonen og de optimale læringsparametrene til SVMs. Santa Fe-valutakursen og fem virkelige futureskontrakter brukes i forsøket. Det vises at den foreslåtte metoden oppnår begge betydelig høyere prediksjonsytelse og raskere konvergenshastighet sammenlignet med en enkelt SVM-modellbasert SVM med et selvorganiserende funksjonskart SOM og testet modellen på Santa Fe-valutakursen og fem virkelige futureskontrakter. De viste at deres foreslåtte metode oppnår både betydelig høyere prediksjonsytelse og raskere konvergenshastighet sammenlignet med en enkelt SVM-modell. SANSOM, DCT DOWNS og TK SAHA, 2003 Evaluering av støttevektormaskinbasert prognostiseringsverktøy i elektrisitetsprognoser for australske nasjonale elektrisitetsmarkedsaktører. Journal of Electrical P CNX NIFTY aksjeindeks De intelligente paradigmene som ble vurdert var et kunstig nevralt nettverk t regnet med Levenberg-Marquardt-algoritmen, støttevektormaskinen, Takagi-Sugeno-neurofluktmodellen og en forskjell som øker neuralt nettverk. De forskjellige paradigmene ble kombinert ved hjelp av to forskjellige ensemble-tilnærminger for å optimalisere ytelsen ved å redusere de ulike feiltiltakene. Den første tilnærmingen er basert på en direkte feilmåling og den andre metoden er basert på en evolusjonær algoritme for å søke den optimale lineære kombinasjonen av de ulike intelligente paradigmene. Eksperimentelle resultater viser at ensembleteknikkene har vært bedre enn de enkelte metodene og den direkte ensemble-tilnærmingen ser ut til å fungere godt for problemet vurderes. vurdert et kunstig nevralt nettverk utdannet ved hjelp av Levenberg-Marquardt-algoritmen, en støttevektormaskin, en Takagi-Sugeno-neurofluktmodell og et forskjell som øker nevrale nettverk for å forutsi NASDAQ-100 Index of The Nasdaq Stock Market og S REZ-CRUZ, Fernando, Julio En AFONSO-RODR GUEZ og Javier GINER, 2003 Esti paring GARCH-modeller ved hjelp av støttevektormaskiner Quantitative Finance Volume 3, Nummer 3, juni 2003, Sider 163-172 Sitert av 2 0 63 år Sammendrag Støttevektormaskiner SVM er et nytt, ikke-parametrisk verktøy for regresjonsestimering. Vi vil bruke dette verktøyet til å estimere parametrene for en GARCH-modell for å forutse den betingede volatiliteten i børsavkastningen GARCH-modeller er vanligvis estimert ved bruk av maksimal sannsynlighet for ML-prosedyrer, forutsatt at dataene blir normalt fordelt. I dette papiret vil vi vise at GARCH-modeller kan estimeres ved bruk av SVMs, og at slike estimater har en høyere forutsigbar evne enn de som er oppnådd via vanlige ML-metoder. brukte SVM til regresjon for å estimere parametrene for en GARCH-modell for å forutsi den betingede volatiliteten i børsreturneringer og viste at slike estimater har en høyere forutsigbar evne enn de som er oppnådd via vanlig maksimal sannsynlighet ML metoder. Fra GESTEL, T et al 2003 Konkurs forutsigelse med minst firkanter støtte vektor m achine classifiers I 2003 IEEE International Conference on Computational Intelligence for Financial Engineering Proceedings sider 1-8 Cited by 2 0 63 år Abstrakte klassifikasjonsalgoritmer som lineær diskriminantanalyse og logistisk regresjon er populære lineære teknikker for modellering og forutsigelse av bedriftens nød. Disse teknikkene tar sikte på å finne en optimal lineær kombinasjon av forklarende inputvariabler, for eksempel solvens - og likviditetsforhold, for å analysere, modellere og forutsi selskapsrisiko. Nylig har kjernebaserte, ikke-lineære klassifiseringsteknikker, som støttevektorer, minst kvadrater vektorer og kjernefisker diskriminantanalyse, har blitt utviklet. I utgangspunktet kartlegger disse metodene inngangene først på en ikke-lineær måte til et høydimensjonalt kjernevirksomhetsrom, hvor en lineær klassifikator er konstruert i det andre trinnet. Praktiske uttrykk oppnås i det såkalte dobbeltrom ved bruk av Mercer s orem I dette papiret forklarer vi relasjonene mellom lineær og ikke-lineær kjernebasert klassifisering og illustrerer deres ytelse ved å forutse konkurs av mid-cap firmaer i Belgia og Nederland. Brukte minst kvadratene støtter vektormaskin klassifiseringer for å forutsi konkurs av mid cap bedrifter i Belgia og Nederland. CAO, LJ og WK CHONG, 2002 Funksjon utvinning i støtte vektor maskin en sammenligning av PCA, XPCA og ICA ICONIP 02 Proceedings av den 9. internasjonale konferansen om Neural Information Processing, volum 2 redigert av Lipo Wang, et al sider 1001-1005 Sitert av 2 0 48 år Abstract Nylig støttet vektormaskin SVM har blitt et populært verktøy i prognoser for tidsserier. Ved utvikling av en vellykket SVM-forecaster er funksjonens ekstraksjon det første viktige trinnet. Dette papiret foreslår applikasjoner av hovedkomponentanalyse PCA, kjerne hovedkomponentanalyse KPCA og uavhengig komponentanalyse ICA til SVM for funksjonell ekstraksjon PCA linearly tran sforms de originale inngangene til ukorrelerte funksjoner KPCA er en ikke-lineær PCA utviklet ved å bruke kjernemetoden I ICA blir de originale inngangene lineært omdannet til statistisk uavhengige funksjoner Ved å undersøke solflekdataene og en sann futureskontrakt viser eksperimentet at SVM ved ekstraksjon av funksjon ved bruk av PCA, KPCA eller ICA kan det fungere bedre enn det uten ekstraksjon. Videre er det bedre generaliseringsytelse i KPCA og ICA-funksjonens ekstraksjon enn PCA-funksjonens ekstraksjon. Vurdert anvendelsen av hovedkomponentanalyse PCA, kjernekomponentanalyse KPCA og uavhengig komponentanalyse ICA til SVMs for funksjonsuttak Ved å undersøke solfleksdataene og en reell futureskontrakt viste de at SVM ved ekstraksjon med PCA, KPCA eller ICA kan fungere bedre enn det uten ekstraksjon. Videre fant de at det er bedre generaliseringsytelse i KPCA og ICA har ekstraksjon enn PCA featur e-utvinning. CAO, LJ og Francis EH TAY, 2000 Feature Selection for Support Vector Machines i Financial Time Series Forecasting i Intelligent Data Engineering og Automated Learning - IDEAL 2000 Data Mining, Financial Engineering og Intelligent Agents redigert av Kwong Sak Leung, Lai - Wan Chan og Helen Meng, sider 268-273 Sitat av 3 0 48 år Sammendrag Dette papiret omhandler bruk av saliencyanalyse for å støtte vektormaskiner SVMs for funksjonsvalg Viktigheten av funksjonen er rangert ved å evaluere sensitiviteten til nettverksutgangen til funksjonsinngang i form av partiell derivat En systematisk tilnærming til å fjerne irrelevante funksjoner basert på sensitiviteten er utviklet. Fem futureskontrakter blir undersøkt i forsøket. Basert på simuleringsresultatene, er det vist at denne analyseanalysen er effektiv i SVMs for å identifisere viktige funksjoner. dealt med anvendelse av saliency analyse til funksjonsvalg for SVMs Fem futures kontrakter ble undersøkt ed og de konkluderte med at saliency analyse er effektiv i SVMs for å identifisere viktige funksjoner. Zhou, Dianmin, Feng GAO og Xiaohong GUAN, 2004 Bruk av nøyaktig online support vektorregresjon i energiprisprognose WCICA 2004 Femte verdens kongress om intelligent kontroll og automatisering, volum 2 sider 1838-1842 Sitert av 1 0 45 år Sammendrag Energipris er den viktigste indikatoren i elektrisitetsmarkedene og dens egenskaper er relatert til markedsmekanismen og endringen i forhold til markedsdeltakernes adferd. Det er nødvendig å bygge en realtidspris forecasting model with adaptive capability In this paper, an accurate online support vector regression AOSVR method is applied to update the price forecasting model Numerical testing results show that the method is effective in forecasting the prices of the electric-power markets. applied an accurate online support vector regression AOSVR to forecasting the prices of the electric-power markets, results showed t hat it was effective. FAN, A and M PALANISWAMI, 2001 Stock selection using support vector machines IJCNN 01 International Joint Conference on Neural Networks, Volume 3 Pages 1793-1798 Cited by 2 0 38 year Abstract We used the support vector machines SVM in a classification approach to beat the market Given the fundamental accounting and price information of stocks trading on the Australian Stock Exchange, we attempt to use SVM to identify stocks that are likely to outperform the market by having exceptional returns The equally weighted portfolio formed by the stocks selected by SVM has a total return of 208 over a five years period, significantly outperformed the benchmark of 71 We also give a new perspective with a class sensitivity tradeoff, whereby the output of SVM is interpreted as a probability measure and ranked, such that the stocks selected can be fixed to the top 25.used SVMs for classification for stock selection on the Australian Stock Exchange and significantly outperformed the benchmark. Van GESTEL, Tony, et al 2000 Volatility Tube Support Vector Machines Neural Network World vol 10, number 1, pp 287-297 Cited by 2 0 32 year Abstract In Support Vector Machines SVM s , a non-linear model is estimated based on solving a Quadratic Programming QP problem The quadratic cost function consists of a maximum likelihood cost term with constant variance and a regularization term By specifying a difference inclusion on the noise variance model, the maximum likelihood term is adopted for the case of heteroskedastic noise, which arises in financial time series The resulting Volatility Tube SVM s are applied on the 1-day ahead prediction of the DAX30 stock index The influence of today s closing prices of the New York Stock Exchange on the prediction of tomorrow s DAX30 closing price is analyzed. developed the Volatility Tube SVM and applied it to 1-day ahead prediction of the DAX30 stock index, and significant positive out-of-sample results were obtained. CAO, Li Juan, K ok Seng CHUA and Lim Kian GUAN, 2003 Combining KPCA with support vector machine for time series forecasting In 2003 IEEE International Conference on Computational Intelligence for Financial Engineering pages 325-329 Cited by 1 0 31 year Abstract Recently, support vector machine SVM has become a popular tool in time series forecasting In developing a successful SVM forecaster, the first important step is feature extraction This paper applies kernel principal component analysis KPCA to SVM for feature extraction KPCA is a nonlinear PCA developed by using the kernel method It firstly transforms the original inputs into a high dimensional feature space and then calculates PCA in the high dimensional feature space By examining the sunspot data and one real futures contract, the experiment shows that SVM by feature forms much better than that extraction using KPCA per without feature extraction In comparison with PCA, there is also superior performance in KPCA. applied kernel principal compon ent analysis KPCA to SVM for feature extraction The authors examined sunspot data and one real futures contract, and found such feature extraction enhanced performance and also that KPCA was superior to PCA. YANG, Haiqin, 2003 Margin Variations in Support Vector Regression for the Stock Market Prediction Degree of Master of Philosophy Thesis, Department of Computer Science - insensitive loss function is usually used to measure the empirical risk The margin in this loss function is fixed and symmetrical Typically, researchers have used methods such as crossvalidation or random selection to select a suitable for that particular data set In addition, financial time series are usually embedded with noise and the associated risk varies with time Using a fixed and symmetrical margin may have more risk inducing bad results and may lack the ability to capture the information of stock market promptly In order to improve the prediction accuracy and to consider reducing the downside risk, we extend the standard SVR by varying the margin By varying the width of the margin, we can reflect the change of volatility in the financial data by controlling the symmetry of margins, we are able to reduce the downside risk Therefore, we focus on the study of setting the width of the margin and also the study of its symmetry property For setting the width of margin, the Momentum also including asymmetrical margin control and Generalized Autoregressive Conditional Heteroskedasticity GARCH models are considered Experiments are performed on two indices Hang Seng Index HSI and Dow Jones Industrial Average DJIA for the Momentum method and three indices Nikkei225, DJIA and FTSE100, for GARCH models, respectively The experimental results indicate that these methods improve the predictive performance comparing with the standard SVR and benchmark model On the study of the symmetry property, we give a sufficient condition to prove that the predicted value is monotone decreasing to the increase of the up margin Therefore, we can reduce the predictive downside risk, or keep it zero, by increasing the up margin An algorithm is also proposed to test the validity of this condition, such that we may know the changing trend of predictive downside risk by only running this algorithm on the training data set without performing actual prediction procedure Experimental results also validate our analysis. employs SVMs for regression and varys the width of the margin to reflect the change of volatility and controls the symmetry of margins to reduce the downside risk Results were positive. CALVO, Rafael A and Ken WILLIAMS, 2002 Automatic Categorization of Announcements on the Australian Stock Exchange Cited by 1 0 24 year Abstract This paper compares the performance of several machine learning algorithms for the automatic categorization of corporate announcements in the Australian Stock Exchange ASX Signal G data stream The article also describes some of the applications that the categorization of corporate announcements may enable We have performed tests on two categorization tasks market sensitivity, which indicates whether an announcement will have an impact on the market, and report type, which classifies each announcement into one of the report categories defined by the ASX We have tried Neural Networks, a Na ve Bayes classifier, and Support Vector Machines and achieved good resultspared the performance of neural networks, a na ve bayes classifier, and SVMs for the automatic categorization of corporate announcements in the Australian Stock Exchange ASX Signal G data stream The results were all good, but with the SVM underperforming the other two models. AHMED, A H M T 2000 Forecasting of foreign exchange rate time series using support vector regression 3rd year project Computer Science Department, University of Manchester Cited by 1 0 16 year. used support vector regression for forecasting a foreign exchange rate time series. GUESDE, Bazile, 2000 Predicting foreign exchange r ates with support vector regression machines MSc thesis Computer Science Department, University of Manchester Cited by 1 0 16 year Abstract This thesis investigates how Support Vector Regression can be applied to forecasting foreign exchange rates At first we introduce the reader to this non linear kernel based regression and demonstrate how it can be used for time series prediction Then we define a predictive framework and apply it to the Canadian exchange rates But the non-stationarity in the data, which we here define as a drift in the map of the dynamics, forces us to present and use the typical learning processes for catching different dynamics Our implementation of these solutions include Clusters of Volatility and competing experts Finally those experts are used in a financial vote trading system and substantial profits are achieved Through out the thesis we hope the reader will be intrigued by the results of our analysis and be encouraged in other dircetions for further researc h. used SVMs for regression to predict the Canadian exchange rate, wisely recognised the problem of nonstationarity, dealt with it using experts and claimed that substantial profits were achieved. BAO, Yu-Kun, et al 2005 Forecasting Stock Composite Index by Fuzzy Support Vector Machines Regression Proceedings of 2005 International Conference on Machine Learning and Cybernetics, Volume 6 pages 3535-3540 not cited 0 year Abstract Financial time series forecasting methods such as exponential smoothing are commonly used for prediction on stock composition index SCI and have made great contribution in practice, but efforts on looking for superior forecasting method are still made by practitioners and academia This paper deals with the application of a novel neural network technique, fuzzy support vector machines regression FSVMR , in SCI forecasting The objective of this paper is not only to examine the feasibility of FSVMR in SCI forecasting but presents our efforts on improving the accuracy of FSVMR in terms of data pre-processing, kernel function selection and parameters selection A data set from Shanghai Stock Exchange is used for the experiment to test the validity of FSVMR The experiment shows FSVMR a better method in SCI forecasting. used fuzzy support vector machines regression FSVMR to forecast a data set from the Shanghai Stock Exchange with positive results. CHEN, Kuan-Yu and Chia-Hui HO, 2005 An Improved Support Vector Regression Modeling for Taiwan Stock Exchange Market Weighted Index Forecasting ICNN s issuer credit rating systems using support vector machines Expert Systems with Applications Volume 30, Issue 3, April 2006, Pages 427-435 not cited 0 year By providing credit risk information, credit rating systems benefit most participants in financial markets, including issuers, investors, market regulators and intermediaries In this paper, we propose an automatic classification model for issuer credit ratings, a type of fundamental credit rating information, b y applying the support vector machine SVM method This is a novel classification algorithm that is famous for dealing with high dimension classifications We also use three new variables stock market information, financial support by the government, and financial support by major shareholders to enhance the effectiveness of the classification Previous research has seldom considered these variables The data period of the input variables used in this study covers three years, while most previous research has only considered one year We compare our SVM model with the back propagation neural network BP , a well-known credit rating classification method Our experiment results show that the SVM classification model performs better than the BP model The accuracy rate 84 62 is also higher than previous research. used an SVM to classify Taiwan s issuer credit ratings and found that it performed better than the back propagation neural network BP model. CHEN, Wun-Hua, Jen-Ying SHIH and Soushan WU, 20 06 Comparison of support-vector machines and back propagation neural networks in forecasting the six major Asian stock markets International Journal of Electronic Finance Volume, Issue 1, pages 49-67 not cited 0 year Abstract Recently, applying the novel data mining techniques for financial time-series forecasting has received much research attention However, most researches are for the US and European markets, with only a few for Asian markets This research applies Support-Vector Machines SVMs and Back Propagation BP neural networks for six Asian stock markets and our experimental results showed the superiority of both models, compared to the early researchespared SVMs and back propagation BP neural networks when forecasting the six major Asian stock markets Both models perform better than the benchmark AR 1 model in the deviation measurement criteria, whilst SVMs performed better than the BP model in four out of six markets. GAVRISHCHAKA, Valeriy V and Supriya BANERJEE, 2006 Support V ector Machine as an Efficient Framework for Stock Market Volatility Forecasting Computational Management Science Volume 3, Number 2 April 2006 , Pages 147-160 not cited 0 year Abstract Advantages and limitations of the existing models for practical forecasting of stock market volatility have been identified Support vector machine SVM have been proposed as a complimentary volatility model that is capable to extract information from multiscale and high-dimensional market data Presented results for SP500 index suggest that SVM can efficiently work with high-dimensional inputs to account for volatility long-memory and multiscale effects and is often superior to the main-stream volatility models SVM-based framework for volatility forecasting is expected to be important in the development of the novel strategies for volatility trading, advanced risk management systems, and other applications dealing with multi-scale and high-dimensional market data. used SVMs for forecasting stock market vola tility with positive results. HOVSEPIAN, K and P ANSELMO, 2005 Heuristic Solutions to Technical Issues Associated with Clustered Volatility Prediction using Support Vector Machines ICNN B 05 International Conference on Neural Networks and Brain, 2005, Volume 3 Pages 1656-1660 not cited 0 year Abstract We outline technological issues and our fimdings for the problem of prediction of relative volatility bursts in dynamic time-series utilizing support vector classifiers SVC The core approach used for prediction has been applied successfully to detection of relative volatility clusters In applying it to prediction, the main issue is the selection of the SVC training testing set We describe three selection schemes and experimentally compare their performances in order to propose a method for training the SVC for the prediction problem In addition to performing cross-validation experiments, we propose an improved variation to sliding window experiments utilizing the output from SVC s decision function Together with these experiments, we show that accurate and robust prediction of volatile bursts can be achieved with our approach. used SVMs for classification to predict relative volatility clusters and achieved accurate and robust results. INCE, H and T B TRAFALIS, 2004 Kernel principal component analysis and support vector machines for stock price prediction Proceedings of the 2004 IEEE International Joint Conference on Neural Networks, Volume 3 pages 2053-2058 not cited 0 year Abstract Financial time series are complex, non-stationary and deterministically chaotic Technical indicators are used with principal component analysis PCA in order to identify the most influential inputs in the context of the forecasting model Neural networks NN and support vector regression SVR are used with different inputs Our assumption is that the future value of a stock price depends on the financial indicators although there is no parametric model to explain this relationship This relationshi p comes from technical analysis Comparison shows that SVR and MLP networks require different inputs The MLP networks outperform the SVR technique. found that MLP neural networks outperform support vector regression when applied to stock price prediction. KAMRUZZAMAN, Joarder, Ruhul A SARKER and Iftekhar AHMAD, 2003 SVM Based Models for Predicting Foreign Currency Exchange Rates Proceedings of the Third IEEE International Conference on Data Mining ICDM 03 Pages 557-560 not cited 0 year Abstract Support vector machine SVM has appeared as a powerful tool for forecasting forex market and demonstrated better performance over other methods, e g neural network or ARIMA based model SVM-based forecasting model necessitates the selection of appropriate kernel function and values of free parameters regularization parameter and varepsilon - insensitive loss function In this paper, we investigate the effect of different kernel functions, namely, linear, polynomial, radial basis and spline on predictio n error measured by several widely used performance metrics The effect of regularization parameter is also studied The prediction of six different foreign currency exchange rates against Australian dollar has been performed and analyzed Some interesting results are presented. investigated the effect of different kernel functions and the regularization parameter when using SVMs to predict six different foreign currency exchange rates against the Australian dollar. investigated comprehensible credit scoring models using rule extraction from SVMs. NALBANTOV, Georgi, Rob BAUER and Ida SPRINKHUIZEN-KUYPER, 2006 Equity Style Timing Using Support Vector Regressions to appear in Applied Financial Economics not cited 0 year Abstract The disappointing performance of value and small cap strategies shows that style consistency may not provide the long-term benefits often assumed in the literature In this study we examine whether the short-term variation in the U S size and value premium is predictabl e We document style-timing strategies based on technical and macro - economic predictors using a recently developed artificial intelligence tool called Support Vector Regressions SVR SVR are known for their ability to tackle the standard problem of overfitting, especially in multivariate settings Our findings indicate that both premiums are predictable under fair levels of transaction costs and various forecasting horizons. used SVMs for regression for equity style timing with positive results. ONGSRITRAKUL, P and N SOONTHORNPHISAJ, 2003 Apply decision tree and support vector regression to predict the gold price Proceedings of the International Joint Conference on Neural Networks, 2003, Volume 4 Pages 2488-2492 not cited 0 year Abstract Recently, support vector regression SVR was proposed to resolve time series prediction and regression problems In this paper, we demonstrate the use of SVR techniques for predicting the cost of gold by using factors that have an effect on gold to estimate its price We apply a decision tree algorithm for the feature selection task and then perform the regression process using forecasted indexes Our experimental results show that the combination of the decision tree and SVR leads to a better performance. applied a decision tree algorithm for feature selection and then performed support vector regression to predict the gold price, their results were positive. Van GESTEL, Tony, et al 2005 Linear and non-linear credit scoring by combining logistic regression and support vector machines, Journal of Credit Risk Vol 1, No 4, Fall 2005, Pages 31-60 not cited 0 year Abstract The Basel II capital accord encourages banks to develop internal rating models that are financially intuitive, easily interpretable and optimally predictive for default Standard linear logistic models are very easily readable but have limited model flexibility Advanced neural network and support vector machine models SVMs are less straightforward to interpret but can capture mo re complex multivariate non-linear relations A gradual approach that balances the interpretability and predictability requirements is applied here to rate banks First, a linear model is estimated it is then improved by identifying univariate non-linear ratio transformations that emphasize distressed conditions and finally SVMs are added to capture remaining multivariate non-linear relations. apply linear and non-linear credit scoring by combining logistic regression and SVMs. YANG, Haiqin, et al 2004 Outliers Treatment in Support Vector Regression for Financial Time Series Prediction Neural Information Processing 11th International Conference, ICONIP 2004, Calcutta, India, November 2004, Proceedings not cited 0 year Abstract Recently, the Support Vector Regression SVR has been applied in the financial time series prediction The financial data are usually highly noisy and contain outliers Detecting outliers and deflating their influence are important but hard problems In this paper, we pr opose a novel two-phase SVR training algorithm to detect outliers and reduce their negative impact Our experimental results on three indices Hang Seng Index, NASDAQ, and FSTE 100 index show that the proposed two-phase algorithm has improvement on the prediction. proposed a novel two-phase SVR training procedure to detect and deflate the influence of outliers The method was tested on the Hang Seng Index, NASDAQ and FSTE 100 index and results were positive However, it s not clear why the significance of outliers such as market crashes should be understated. YU, Lean, Shouyang WANG and Kin Keung LAI, 2005 Mining Stock Market Tendency Using GA-Based Support Vector Machines Internet and Network Economics First International Workshop, WINE 2005, Hong Kong, China, December 15-17, 2005, Proceedings Lecture Notes in Computer Science edited by Xiaotie Deng and Yinyu Ye, pages 336-345 not cited 0 year Abstract In this study, a hybrid intelligent data mining methodology, genetic algorithm based supp ort vector machine GASVM model, is proposed to explore stock market tendency In this hybrid data mining approach, GA is used for variable selection in order to reduce the model complexity of SVM and improve the speed of SVM, and then the SVM is used to identify stock market movement direction based on the historical data To evaluate the forecasting ability of GASVM, we compare its performance with that of conventional methods e g statistical models and time series models and neural network models The empirical results reveal that GASVM outperforms other forecasting models, implying that the proposed approach is a promising alternative to stock market tendency exploration. applied a random walk RW model, an autoregressive integrated moving average ARIMA model, an individual back-propagation neural network BPNN model, an individual SVM model and a genetic algorithm-based SVM GASVM to the task of predicting the direction of change in the daily S P500 stock price index and found that their proposed GASVM model performed the best. HARLAND, Zac, 2002 Using Support Vector Machines to Trade Aluminium on the LME Proceedings of the Ninth International Conference, Forecasting Financial Markets Advances For Exchange Rates, Interest Rates and Asset Management edited by C Dunis and M Dempster not listed Abstract This paper describes and evaluates the use of support vector regression to trade the three month Aluminium futures contract on the London Metal Exchange, over the period June 1987 to November 1999 The Support Vector Machine is a machine learning method for classification and regression and is fast replacing neural networks as the tool of choice for prediction and pattern recognition tasks, primarily due to their ability to generalise well on unseen data The algorithm is founded on ideas derived from statistical learning theory and can be understood intuitively within a geometric framework In this paper we use support vector regression to develop a number of trading submodel s that when combined, result in a final model that exhibits above-average returns on out of sample data, thus providing some evidence that the aluminium futures price is less than efficient Whether these inefficiencies will continue into the future is unknown. used an ensemble of SVMs for regression to trade the three month Aluminium futures contract on the London Metal Exchange with positive results. Van GESTEL, T et al 2005 Credit rating systems by combining linear ordinal logistic regression and fixed-size least squares support vector machines, Workshop on Machine Learning in Finance, NIPS 2005 Conference, Whistler British Columbia, Canada , Dec 9 not listed. developed credit rating systems by combining linear ordinal logistic regression and fixed-size least squares SVMs. Machine Learning How Support Vector Machines can be used in Trading. What is a Support Vector Machine. A support vector machine is a method of machine learning that attempts to take input data and classify into one of tw o categories In order for a support vector machine to be effective, it is necessary to first use a set of training input and output data to build the support vector machine model that can be used for classifying new data. A support vector machine develops this model by taking the training inputs, mapping them into multidimensional space, then using regression to find a hyperplane a hyperplane is a surface in n-dimensional space that it separates the space into two half spaces that best separates the two classes of inputs Once the support vector machine has been trained, it is able to assess new inputs with respect to the separating hyperplane and classify it into one of the two categories. A support vector machine is essentially an input output machine A user is able to put in an input, and based on the model developed through training, it will return an output The number of inputs for any given support vector machine theoretically ranges from one to infinity, however in practical terms computing power does limit how many inputs can be used If for example, N inputs are used for a particular support vector machine the integer value of N can range from one to infinity , the support vector machine must map each set of inputs into N-dimensional space and find a N-1 - dimensional hyperplane that best separates the training data. Figure 1 Support Vector Machines are input output machines. The best way to conceptualize how a support vector machine works is by considering the two dimensional case Assume we want to create a support vector machine that has two inputs and returns a single output that classifies the data point as belonging to one of two categories We can visualize this by plotting it on a 2-dimensional chart such as the chart below. Figure 2 Left Support vector machine inputs mapped to a 2D chart The red circles and blue crosses are used to denote the two classes of inputs. Figure 3 Right Support vector machine inputs mapped to a 2D chart The red circles and blue cros ses are used to denote the two classes of inputs with a black line indicating the separating hyperplane. In this example, the blue crosses indicate data points that belong to category 1 and the red circles that represent data points that belong to category 2 Each of the individual data points has unique input 1 value represented by their position on the x-axis and a unique input 2 value represented by their position on the y-axis and all of these points have been mapped to the 2-dimensional space. A support vector machine is able to classify data by creating a model of these points in 2 dimensional space The support vector machine observes the data in 2 dimensional space, and uses a regression algorithm to find a 1 dimensional hyperplane aka line that most accurately separate the data into its two categories This separating line is then used by the support vector machine to classify new data points into either category 1 or category 2.The animation below illustrates the process of traini ng a new support vector machine The algorithm will start by making a random guess finding a separating hyperplane, then iteratively improve the accuracy of the hyperplane As you can see the algorithm starts quite aggressively, but then slows down as it starts to approach the desires solution. Figure 4 An animation showing a support vector machine training The hyperplane progressively converges on the ideal geometry to separate the two classes of data. The 2-dimensional scenario above presented allows us to visualize the the process of a support vector machine, however it is only able to classify a data point using two inputs What if we want to use more inputs Thankfully, the support vector machine algorithm allows us to do the same in higher dimensions, though it does become much harder to conceptualize. Consider this, you wish to create support vector machine that takes 20 inputs and can classify any data point using these inputs into either category 1 or category 2 In order to do this, the support vector machine needs to model the data in 20 dimensional space and use a regression algorithm to find a 19 dimensional hyperplane that separates the data points into two categories This gets exceedingly difficult to visualize as it is hard for us to comprehend anything above 3-dimensions, however all that you need to know is that is works in exactly the same way as it does for the 2-dimensional case. How do Support Vector Machines Work Example Is It A Schnick. Imagine this hypothetical scenario, you are a researcher investigating a rare animal only found in the depths of the Arctic called Shnicks Given the remoteness of these animals, only a small handful have ever been found let s say around 5000 As a researcher, you are stuck with the question how can I identify a Schnick. All you have at your disposal are the research papers previously published by the handful of researchers that have seen one In these research papers, the authors describe certain characteristics about the Schnicks they found, i e height, weight, number of legs, etc But all of these characteristics vary between the research papers with no discernible pattern. How can we use this data to identify a new animal as a schnick. One possible solution to our problem is to use a support vector machine to identify the patterns in the data and create a framework that can be used to classify animals as either a schnick or not a schnick The first step is to create a set of data that can be used to train your support vector machine to identify schnicks The training data is a set of inputs and matching outputs for the support vector machine to analyze and extract a pattern from. Therefore, we must decide what inputs will be used and how many Theoretically, we can have as many inputs as we want, however this can often lead to slow training the more inputs you have the more time it takes the support vector machine to extract patterns Also, you want to choose inputs values that will tend to be relatively con sistent across all schnicks For example, height or weight of the animal would be a good example of an input because you would expect that this would be relatively consistent across all schnicks However, the average age of an animal would be a poor choice of input because you would expect the age of animals identified would all vary considerably. For this reason, the following inputs were chosen. The number of legs. The number of eyes. The length of the animal s arms. The animals average speed. The frequency of the animals mating call. With the inputs chosen, we can start to compile our training data Effective training data for a support vector machine must meet certain requirements. The data must have examples of animals that are schnicks. The data must have examples of animals that are not schnicks. In this case we have the research papers of scientist that have successfully identified a schnick and listed their properties Therefore we can read these research papers and extract the data under e ach of the inputs and allocate an output of either true or false to each of the examples The training data in this case may look similar to the table below. Table 1 Example table of schnick observations. Once we have gathered the data for all of our training inputs and outputs, we can use it to train our support vector machine During the training process, the support vector machine will create a model in seven dimensional space that can be used to sort each of the training examples into either true or false The support vector machine will continue to do this until it has a model that accurately represents the training data within the specified error tolerance Once training is complete, this model can be used to classify new data points as either true or false. Does the Support Vector Machine Actually Work. Using the Schnick scenario, I have written a script that tests how well a support vector machine can actually identify new schnicks To do this, I have used the Support Vector Machine Lea rning Tool function Library that can be downloaded from the Market. To model this scenario effectively, we need to first decide what are the actual properties of a Schnick The properties I have assumed in this case have been listed in the table below If an animal satisfies all of the criteria below, then it is a Schnick. Table 2 Summary of parameters that define a schnick. Now that we have defined our Schnick, we can use this definition to experiment with support vector machines The first step is to create a function that is able to take the seven inputs for any given animal and return the actual classification of the animal as a schnick or not This function will be used to generate training data for the support vector machine as well as assess the performance of it at the end This can be done using the function below. The next step in the process is to create a function that can generate the training inputs and outputs Inputs in this case will be generated by creating random numbers withi n a set range for each of the seven input values Then for each of the sets of random inputs generated, the isItASchnick function above will be used to generate the corresponding desired output This is done in the function below. We now have a set of training inputs and outputs, it is now time to create our support vector machines using the Support Vector Machine Learning Tool available in the Market Once a new support vector machine is created, it is necessary to pass the training inputs and outputs to it and execute the training. We now have a support vector machine that has been successfully trained in identifying Scnhicks To verify this, we can test the final support vector machine by asking it to classify new data points This is done by first generating random inputs, then using the isItASchnick function to determine whether these inputs correspond to an actual Schnick, then use the support vector machine to classify the inputs and determine whether the predicted outcome matches the actual outcome This is done in the function below. I recommend playing with the values within the above functions to see how the support vector machine performs under different conditions. Why is Support Vector Machine So Useful. The benefit of using a support vector machine to extract complex pattern from the data is that it is not necessary a prior understanding of the behavior of the data A support vector machine is able to analyze the data and extract its only insights and relationships In this way, it functions similar to a black box receiving an inputs and generating an output which can prove to be very useful in finding patterns in the data that are too complex and not obvious. One of the best features of support vector machines is that they are able to deal with errors and noise in the data very well They are often able to see the underlying pattern within the data and filter out data outliers and other complexities Consider the following scenario, in performing your research on Sc hnicks, you come across multiple research papers that describe Schnicks with massively different characteristics such as a schnick that is 200kg and is 15000mm tall. Errors like this can lead to distortions your model of what a Schnick is, which could potentially cause you to make an error when classifying new Schnick discoveries The benefit of the support vector machine is that it will develop a model that agrees with the underlying pattern opposed to a model that fits all of the training data points This is done by allowing a certain level of error in the model to enable the support vector machine to overlook any errors in the data. In the case of the Schnick support vector machine, if we allow an error tolerance of 5 , then training will only try to develop a model that agrees with 95 of the training data This can be useful because it allows training to ignore the small percentage of outliers. We can investigate this property of the support vector machine further by modifying our Schni ck script The function below has been added to introduce deliberate random errors in our training data set This function will select training points at random and replace the inputs and corresponding output with random variables. This function allows us to introduce deliberate errors into our training data Using this error filled data, we can create and train a new support vector machine and compare its performance with the original one. When the script is run, it produces the following results in the Expert Log Within a training data set with 5000 training points, we were able to introduce 500 random errors When comparing the performance of this error filled support vector machine with the original one, the performance is only reduced by 1 This is because the support vector machine is able to overlook the outliers in the data set when training and is still capable of producing an impressively accurate model of the true data This suggests that support vector machines could potentially be a more useful tool in extracting complex patterns and insights from noisy data sets. Figure 5 The resulting expert log following the running of the Schnick script in the MetaTrader 5.Demo Versions. A full version of the above code can be downloaded from Code Base, however this script can only be run in your terminal if you have purchased a full version of the Support Vector Machine Learning tool from the Market If you only have a demo version of this tool downloaded, you will be limited to using the tool via the strategy tester To allow testing of the Schnick code using the demo version of the tool, I have rewritten a copy of the script into an Expert Advisor that can be deployed using the strategy tester Both of these code versions can be downloaded by following the links below. Full Version - Using a Script that is deployed in the MetaTrader 5 terminal requires a purchased version of the Support Vector Machine Learning Tool. Demo Version - Using an Expert Advisor that is deployed in the MetaTrader 5 strategy tester requires only a demo version of the Support Vector Machine Learning Tool. How Can Support Vector Machines be used in the Market. Admittedly, the Schnick example discussed above is quite simple, however there are quite a few similarities that can be drawn between this example and using the support vector machines for technical market analysis. Technical analysis is fundamentally about using historical market data to predict future price movements In the same way within the schnick example, we were using the observations made by past scientists to predict whether a new animal is a schnick or not Further, the market is plagued with noise, errors and statistical outliers that make the use of a support vector machine an interesting concept. The basis for a significant number of technical analysis trading approaches involve the following steps. Monitoring several indicators. Identifying what conditions for each indicator correlates with a potentially successful trade. Watch each of the indicators and assess when they all or most are signalling a trade. It is possible to adopt a similar approach to use support vector machines to signal new trades in a similar way The support vector machine learning tool was developed with this in mind A full description of how to use this tool can be found in the Market, so I will only give a quick overview The process for using this tool is as follows. Figure 6 The block diagram showing the process for implementing the support vector machine tool in an Expert Advisor. Before you can use the Support Vector Machine Learning Tool, it is important to first understand how the training inputs and outputs are generated. How are Training Inputs Generated. So, the indicators you want to use as inputs have been already been initialized as well as your new support vector machine The next step is to pass the indicator handles to your new support vector machine and instruct it on how to generate the training data This is done by call ing the setIndicatorHandles function This function allows you to pass the handles of initialized indicators into the support vector machine This is done by passing and integer array containing the handles The two other inputs for this function is the offset value and the number of data points. The offset value denotes the offset between the current bar and the starting bar to be used in generating the training inputs and the number of training points denoted by N sets the size your training data The diagram below illustrates how to use these values An offset value of 4 and an N value of 6 will tell the support vector machine to only use the bars captured in the white square to generate training inputs and outputs Similarly, an offset value of 8 and an N value of 8 will tell the support vector machine to only use the bars captured in the blue square to generate training inputs and outputs. Once the setIndicatorHandles function has been called, it is possible to call the genInputs function This function will use the indicator handles to passed to generate an array of input data to be used for training. Figure 7 Candle chart illustrating the values of Offset and N. How are Training Outputs Generated. Training outputs are generated by simulating hypothetical trades based on historical price data and determining whether such a trade would have been successful or unsuccessful In order to do this, there are a few parameters that are used to instruct the support vector machine learning tool how to assess a hypothetical trade as either successful or unsuccessful. The first variable is OPTRADE The value of this can either be BUY or SELL and will correspond to either hypothetical buy or sell trades If the value of this is BUY, then when generating the outputs it will only look at the potential success of hypothetical buy trades Alternatively, if the value of this is SELL, then when generating the outputs it will only look at the potential success of hypothetical sell trades. The next values used is the Stop Loss and Take Profit for these hypothetical trades The values are set in pips and will set the stop and limit levels for each of the hypothetical trades. The final parameter is the trade duration This variable is measured in hours and will ensure that only trades that are complete within this maximum duration will be deemed successful The reason for including this variable is to avoid the support vector machine signalling trades in a slow moving sideways market. Considerations to Make When Choosing Inputs. It is important to put some thought into the input selection when implementing support vector machines in your trading Similar the Schnick example, it is important to choose an input that would be expected to have similar across difference incidences For example, you may be tempted to use a moving average as an input, however since the long term average price tends to change quite dramatically over time, a moving average in isolation may not be the best input to use This is because there won t be any significant similarity between the moving average value today and the moving average values six months ago. Assume we are trading EURUSD and using a support vector machine with a moving average input to signal buy trades Say the current price is 1 10, however it is generating training data from six months ago when the price was 0 55 When training the support vector machine, the pattern it finds may only lead to a trade being signaled when the price is around 0 55, since this is the only data it knows Therefore, your support vector machine may never signal a trade until the price drops back down to 0 55.Instead, a better input to use for the support vector machine may be a MACD or a similar oscillator because the value of the MACD is independent of the average price level and only signals relative movement I recommend you experiment with this to see what produces the best results for you. Another consideration to make when choosing inputs is ensurin g that the support vector machine has an adequate snapshot of an indicator to signal a new trade You may find in your own trading experience that a MACD is only useful when you have the past five bars to look at, as this will show a trend A single bar of the MACD may be useless in isolation unless you can tell if it is heading up or down Therefore, it may be necessary to pass the past few bars of the MACD indicator to the support vector are two possible ways you can do this. You can create a new custom indicator that uses the past five bars of the MACD indicator to calculate a trend as a single value This custom indicator can then be passed to the support vector machine as a single input, or. You can use the previous five bars of the MACD indicator in the support vector machine as five separate inputs The way to do this is to initialize five different instances of the MACD indicator Each of the indicators can be initialized with a different offset from the current bar Then the five handl es from the separate indicators can be passed to the support vector machine It should be noted, that option 2 will tend to cause longer execution times for your Expert Advisor The more inputs you have, the longer it will take to successfully train. Implementing Support Vector Machines in and Expert Advisor. I have prepared an Expert Advisor that is an example of how someone could potentially use support vector machines in their own trading a copy of this can be downloaded by following this link Hopefully the Expert Advisor will allow you to experiment a little with support vector machines I recommend you copy change modify the Expert Advisor to suit your own trading style The EA works as follows. Two new support vector machines are created using the svMachineTool library One is set up to signal new Buy trades and the other is set up to signal new Sell trades. Seven standard indicators are initialized with each of their handles stored to an integer array Note any combination of indicators c an be used as inputs, they just need to be passed to the SVM in a single integer array. The array of indicator handles is passed to the new support vector machines. Using the array of indicator handles and other parameters, historical price data is used to generate accurate inputs and outputs to be used for training the support vector machines. Once all of the inputs and outputs have been generated, both of the support vector machines are trained. The trained support vector machines are used in the EA to signal new buy and sell trades When a new buy or sell trade is signaled, the trade opens along with manual Stop Loss and Take Profit orders. The initialization and training of the support vector machine are executed within the onInit function For your reference, this segment of the svTrader EA has been included below with notes. Advanced Support Vector Machine Trading. Additional capability was built into the support vector machine learning tool for the more advanced users out there The tool allows users to pass in their own custom input data and output data as in the Schnick example This allows you to custom design your own criteria for support vector machine inputs and outputs, and manually pass in this data to train it This opens up the opportunity to use support vector machines in any aspect of your trading. It is not only possible to use support vector machines to signal new trades, but it can also be used to signal the closing of trades, money management, new advanced indicators etc However to ensure you don t receive errors, it is important to understand how these inputs and outputs are to be structured. Inputs Inputs are passed to SVM as a 1 dimensional array of double values Please note that any input you create must be passed in as a double value Boolean, integer, etc must all be converted into a double value before being passed into the support vector machine The inputs are required in the following form For example, assume we are passing in inputs with 3 inputs x 5 training points To achieve this, our double array must be 15 units long in the format. A 1 B 1 C 1 A 2 B 2 C 2 A 3 B 3 C 3 A 4 B 4 C 4 A 5 B 5 C 5.It is also necessary to pass in a value for the number of inputs In the case, NInputs 3.Outputs outputs are passed in as an array of Boolean values These boolean values are the desired output of the SVM corresponded to each of the sets of inputs passed in Following the above example, say we have 5 training points In this scenario, we will pass in a Boolean array of output values that is 5 units long. When generating your own inputs and outputs, be sure that the length of your arrays matches the values you pass in If they don t match, an error will be generated notifying you of the discrepancy For example, if we have passed in NInputs 3, and inputs is an array of length 16, an error will be thrown since, a Ninputs value of 3 will mean that the length of any input array will need to be a multiple of 3 Similarly, ensure that the number of sets of inputs and the number of outputs that you pass in are equal Again, if you have NInput s 3, length of inputs of 15 and a length of outputs of 6, another error will be thrown as you have 5 sets of inputs and 6 outputs. Try to ensure you have enough variation in your training outputs For example, if you pass in 100 training points, which means an output array of length 100, and all of the values are false with only one true, then the differentiation between the true case and the false case is not sufficient enough This will tend to lead to the SVM training very fast, but the final solution being very poor A more diverse training set will often lead to a more affective SVM. Support Vector Machine Learning Tool. This is an easy-to-use tool for implementing Support Vector Machine Learning in your Expert Advisors, Indicators and other MetaTrader 5 projects. Until now, the use of support vector machine classification has been limited only by advanced coders via external java and c dll libraries This tool has been developed using only the standard MetaTrader 5 tools and provides adv anced support vector machine functionality using a very simple interface. Please note this product is not an Expert Advisor or Indicator This is a library that allows users to implement support vector machine classification in their own Expert Advisors and Indicators. What is a Support Vector Machine. Support vector machines svm are a form of machine learning that use a supervised learning algorithm to analyze data and recognize patterns to be used for classification They are used most prominently in fields such as bioinformatics and mathematics, however this library has been specifically developed with the intention to use support vector machine learning to analyze historical price data and extract patterns that can be used to generate signals. If you want to find out more about the support vector machines mechanics and how they work, I suggest you start with the Wikipedia page The article provides good overview and further links if you are interested in looking into it further. How the Li brary Works. The basic process for any support vector machine is as follows. Gather historical market price and indicator data. Use historical data to generate a set of training inputs and outputs. Use these historical inputs and outputs to train the support vector machine. Use the trained support vector machine to analyze current market price and indicator data to signal new trades. A support vector machine is basically an input output machine The user passes input s to the machine and it produces an output of either true or false If the support vector machine has not yet been trained, it will usually give only a random output for any given input To have the support vector machine produce a useful output, it must first be trained. The training of a support vector machine is done by passing in a set of inputs with a set of corresponding desired outputs The support vector machine algorithm will then use this combined dataset to extract patterns In the case of this tool, the inputs used are ind icators inputs can be any combination of standard or custom indicators selected by the user and the outputs are either true or false corresponding to whether a new trade should be opened. Once the indicators to be used as inputs have been selected by the user along with the parameters for determining outputs, the tool will generate a set of inputs and outputs to be used for training the support vector machine Once this is done, training can be commenced Once the support vector machine has been successfully trained, it can be used to take current indicator values as inputs and signal the Expert Advisor to either make a new trade, or not. Advanced Users additional functions have been included to allow users to manually create and set your training inputs and outputs This can be used for more complex applications such as signalling when to exit a trade or for money management To do this, see details below on the setInputs and setOutputs functions. A variety of functions have been included fo r both basic and advanced users These are outlined below. Training the support vector machine can consume a significant amount of memory This option sets the maximum memory footprint you want the support vector machine to take The value given is measured in MB If a memory value of 1000 MB is set, then the training algorithm will manage its memory to keep its foot print below this level This should be considered particularly if you choose to perform back testing across multiple cores For example, if I have a quad core computer corresponding to 4 local testing agents and I have 8 GB RAM, I will generally set my memory value to about 1250 MB This will mean that when training is being executed in parallel across all local agents, only a maximum of 5000 MB 4 x 1250 MB will be used leaving 3000 MB for the operating system and other programs without causing problems. This will limit the maximum number of training cycles that will occur The reason for this is to avoid the scenario where training never stops This can occasionally happen because it is trying to achieve an impossible solution Unless you have a specific reason, I recommend you don t manually change this value. This value sets the maximum error you are willing to accept from the final support vector machine The input for this is a percentage i e 0 1 is 10 error, 0 15 is 15 error If you are finding that your training doesn t converge on a solution, I recommend you increase the acceptable error tolerance value. Example How to Use the Support Vector Machine Tool to Signal Trades. An example of Expert Advisor svmTrader has been written to show a typical use of the support vector machine learning tool You can download it for free from Code Base.
No comments:
Post a Comment