2002年日本オペレーションズ・リサーチ学会 秋季研究発表会 2−E−1
Portfo1ioOptimizationunderShortSaleOpportunity
01102370 中央大学 02103820 中央大学 02702020 中央大学1.Introduction
今野浩 ⅩONNOHiroshi *越塚知事 ⅩOSHIZtTXATbmoyuki山本零 YAMAMOTORei
▼lmaximi次f(u,V,y)=∑77(叫−り)
メ=1 T れ 一入∑扉/ア−∑吋り+すり) t=1 ブ=1 nSubjectto仇−∑み(り−り)=0,
J=1 f=1,2,…,r †l †l∑り+7∑り=1
メ=1 j=1 †l∑α‘血−り)≧わ‘,
j=1 宜=1,2,…,mO≦り≦α,0≦り≦α′
りり=0,ゴ=1,2,…,れ しW占consider a portfo1io optimization problem under short sale opportunity.When we sellas− sets short we must pay deposit and commission
feetothethirdpartywholendstheaぉets,and theca5l10btainedbytheshortsaleisheldatthe party.Wbhavetouseuptoallamountoffundat thetimeofportfo1ioconstruction.Inthiscasethe investablesetisanonconvexset,SOthatthemean− variancemodelbecomesanonconvexminimization problem.ThiskindofproblemcannotbesoIved
bystandardnonlinearprogrammlngmethodology・
SoweproposeabranChandboundalgorithmex− ploitingthespecialstruCtureOfthisproblem・Itis demonstratedthatthisalgorithmcansoIvevirtu− allya11testproblemsinaverye侃cientmannCr.2.Fbrmulation
Let there be n舶SetSち,j=1,2,…,nin
the marketandletちbe the random variable
repr尊entingthe rate ofreturn ofち・Ako,1et Xj,j=1,2,…,nbctheproportionoffund(ei− therp(冶itiveornegative)to beallocated toち・Thetotalc舶houtflowis(7isapositiveco吋ant) れ ∑(l頑++7lりト)=1 メ=1 (1) wbere Whererjt,t=1,2,…,Taretherateofreturnof thej・thassetduringthepast t−thperiod.AIso, み=りモーり・ TbconstruCtapraCticalalgorithm,Werelaxthe equalityconstraint(1)asfo1lows: れ ▼1
1一方≦∑り+7∑り≦1
ブ=1 メ=1 (2) forsomepositiveconstant6>0.3・ABranchandBoundAlgoritlm
Thefirst and naturalstepfor soIving this nonconvex problemis to relax the complemen−
tarity condition ujVj=0,j=1,2,・・・,n and
SOIvetheresultingquadraticprogrammlngprOb− 1em・Let(u書,V■,y+)be an optimalsolution of
thequadraticprogrammlngprOblem・Ifu;v;=0
,j=1,2,…,n,thenitisobviouslyanOptimalso−1utionoftheorlglnalproblem・Ifthereexistany /
jssuchthatviblatecomplementaritycondition,
WeuSeabranchandboundmethod.Ⅵ屯consider thefo1lowlnglinearsystem托
汀∬ブ≧0 0therwise ifごJ≧0 0tbeⅣke l頑+= I勺ト= Letu5introduceujandvjSuChthat り−り=勺,り≧0,り≧0,りり=0, J=1,2,…,m. ThenLxjI+=uj,lxjト=Vj・Weemplqythehis− torical data to represent the return structure of theasset.TherneanーVariancemodelcanberepre− sented asfo1lows:ー238−
lOIf9=¢thengoto70. 20chooseonesubproblem乃∈gandletg\挿). 30soIve(乃).If(乃)isinfeaLSible,thengobacktolO. Otherwiselet(uZ,VZ,yZ)beitsoptimalsolution andletfzbeitsoptimalvalue. 40culculateabasicfbasiblesolution(材,均l)ofthe linearsystemQl(yl).
50If the complementarity conditions are satis丘ed, thenleりIbeitsoptimalvalue.Otherwisegener− ateanewsolutionby(3).Ifthesolutionsatisfies (2),thenletflbetheassociatedoptimalvalue.If fi>/’,thenJ’:=flandremoveallsubproblem 凡sucbtbat九≦J■. 60 Let 軋+1=仇∪伺max(如ま>E)) 叱+1=Ⅵ∪伺max(如三>e)) andgeneratestlbproblemsPL+1and Et+2.Let 伊=ク∪(J㌔十1,J㌔+2),andretumtolO. 70sbp・
4.ComputationalResultand
Conclusions
WもtestedtheAlgorithmforthisproblemwith
andwithoutupperboundoneashvaLriable,uSlng themonthlydataoftheTbkyoStockExchange.Wb Showcomputationalresultsinpresentation.Theproblemwithoutupperbound(Classl)is
easiersincethesol11tiongeneratedatthefirststageCOntainsatmostonepairofⅦriableviolatingthe
COmplementaritycondition.AlltestprdblemswereSOIvedwithoutusingbranChandboundprocedure・
Whiletheproblemwithupperbound(Claぉ2)
ismorecomplicated.However,WeShowthatonly OneOutOfsixtestproblemsrequiredbranchand boundprocedure.Wb demonst,rated that a mean−Variance model undershortsaleopportunitycanbesoIvedvery
kLSt・Thoughnonconvex,itisusual1ynotmorediト ficultthansoIvingstandardmean−VarianCemOdel
Withoutshortsaleopportunity.AIsoevenwhen
We need to apply branch and bound pr∝edure, the numberofiterationsisnotexcessivesincethe
numberofⅦriablesviolatingthecomplementarity
COnditionisusllal1yverysmall.WearecurrentlyplannlngeXtenSivcexperimentstocomparethereト
ativeadvantageoftheMVmodelwithandwithout Shortsaleopportunityuslngthefactormodel. Reference.今野浩「理財工学Ⅰ,Ⅱ」日科技連出 版,1995,1998 Tl n∑凍明−∑減明=瑳,f=1,2,…,ア
ブ=1 ブ=1 n Tl∑α肋−∑町り≧あ‘,宜=1,2,…m
メ=1 j=1 ▼l ▼11一方≦∑り+7∑り≦1
メ=1 ブ=1/ 0≦り≦α,0≦り≦α, J=1,2,…,れ.
Obviouslythisproblemhasafeasiblesolution,SO thatithasaba5icfeasiblesolution(滋,i)). Letl左>0,碗>0.Iftik疾>e,thenlet us redefine t左一塊 汀l砧≧穐 0 0therwise, O if tん>鴫 鳴一晩 otberwi5e. uた= Vた= (3)Thereforeifthe new solution satisfythe condi−
tion(2)thenwearedone・Otherwisewesplitthe
probleminto two subproblems byimposlng the conditionuk=00rVk=0. kt usde丘ne maximi2;e f(u,V,y) ▼l