A learning-guided multi-objective evolutionary algorithm for constrained portfolio optimization

Khin Lwin, Rong Qu, Graham Kendall

Research output: Contribution to journalArticle

Abstract

Portfoliooptimizationinvolvestheoptimalassignmentoflimitedcapitaltodifferentavailablefinancialassetstoachieveareasonabletrade-offbetweenprofitandriskobjectives.Inthispaper,westudiedtheextendedMarkowitz’smean-varianceportfoliooptimizationmodel.Weconsideredthecardinality,quan-tity,pre-assignmentandroundlotconstraintsintheextendedmodel.Thesefourreal-worldconstraintslimitthenumberofassetsinaportfolio,restricttheminimumandmaximumproportionsofassetsheldintheportfolio,requiresomespecificassetstobeincludedintheportfolioandrequiretoinvesttheassetsinunitsofacertainsizerespectively.Anefficientlearning-guidedhybridmulti-objectiveevolutionaryalgo-rithmisproposedtosolvetheconstrainedportfoliooptimizationproblemintheextendedmean-varianceframework.Alearning-guidedsolutiongenerationstrategyisincorporatedintothemulti-objectiveopti-mizationprocesstopromotetheefficientconvergencebyguidingtheevolutionarysearchtowardsthepromisingregionsofthesearchspace.Theproposedalgorithmiscomparedagainstfourexistingstate-of-the-artmulti-objectiveevolutionaryalgorithms,namelyNon-dominatedSortingGeneticAlgorithm(NSGA-II),StrengthParetoEvolutionaryAlgorithm(SPEA-2),ParetoEnvelope-basedSelectionAlgorithm(PESA-II)andParetoArchivedEvolutionStrategy(PAES).ComputationalresultsarereportedforpubliclyavailableOR-librarydatasetsfromsevenmarketindicesinvolvingupto1318assets.Experimentalresultsontheconstrainedportfoliooptimizationproblemdemonstratethattheproposedalgorithmsignificantlyoutperformsthefourwell-knownmulti-objectiveevolutionaryalgorithmswithrespecttothequalityofobtainedefficientfrontierintheconductedexperiments
Original languageEnglish
Pages (from-to)757-772
JournalApplied Soft Computing Journal
Volume24
Early online date27 Aug 2014
DOIs
Publication statusPublished - Nov 2014

Fingerprint Dive into the research topics of 'A learning-guided multi-objective evolutionary algorithm for constrained portfolio optimization'. Together they form a unique fingerprint.

  • Cite this