




摘 要: 考慮帶參數(shù)敏感度的最優(yōu)權(quán)衡投資組合問題,其模型是一個非凸非可微優(yōu)化問題,其中目標(biāo)函數(shù)含有極大和極小函數(shù)。將該優(yōu)化問題變換為一個等價的非凸二次約束二次規(guī)劃問題,提出了等價變換問題的一個緊的半定規(guī)劃松弛,并估計(jì)了其與原問題之間的間隙。數(shù)值結(jié)果表明,該半定規(guī)劃松弛可以有效找到大多數(shù)測試問題的全局最優(yōu)解,且計(jì)算時間優(yōu)于求解器GUROBI,從而為尋求問題的一個好的近似解提供方法。
關(guān)鍵詞: 參數(shù)敏感度;投資組合;非凸二次約束二次規(guī)劃;半定規(guī)劃松弛;GUROBI
中圖分類號: O224
文獻(xiàn)標(biāo)志碼: A
文章編號: 1673-3851 (2024)11-0861-06
引文格式:王琳,洪陳春,羅和治. 帶參數(shù)敏感度的最優(yōu)權(quán)衡投資組合問題的半定規(guī)劃松弛[J]. 浙江理工大學(xué)學(xué)報(bào)(自然科學(xué)),2024,51(6):861-866.
Reference Format:" WANG Lin, HONG Chenchun,LUO Hezhi. Semi-definite programming relaxation for optimal trade-off portfolio selection with sensitivity of parameters[J]. Journal of Zhejiang Sci-Tech University,2024,51(6):861-866.
Semi-definite programming relaxation for optimal trade-off portfolio selection with sensitivity of parameters
WANG Lin1, HONG Chenchun2, LUO Hezhi1
(1.School of Science, Zhejiang Sci-Tech University, Hangzhou 310018, China; 2.Huaxin Consulting Co., Ltd., Hangzhou 310014, China)
Abstract: In this paper, we consider the optimal trade-off portfolio problem with parameter sensitivity. For this problem, the model is a non-convex and non-differentiable optimization problem in which the objective function contains the maximum and minimum functions. This optimization problem is transformed into an equivalent non-convex quadratically constrained quadratic programming problem. A tight semi-definite programming relaxation for the equivalent transformation problem is proposed and the gap between it and the original problem is estimated. The numerical results show that the semi-definite programming relaxation can effectively find the global optimal solution of most test problems, and the computational time is less than that of the solver GUROBI. It can provide a method for finding a good approximate solution to the problem.
Key words: sensitivity of parameters; portfolio selection; non-convex quadratically constrained quadratic programming; semi-definite programming relaxation; GUROBI
0 引 言
1952年,Markowitz[1]提出了投資組合選擇的均值-方差(Mean-variance,MV)模型,開辟了投資組合量化分析的新時代,MV模型開創(chuàng)性的貢獻(xiàn)是用均值和方差衡量投資組合的預(yù)期收益和風(fēng)險。MV模型中的基本參數(shù)是期望收益和協(xié)方差矩陣,通常來源于歷史數(shù)據(jù),然而歷史數(shù)據(jù)的缺失可能會導(dǎo)致MV模型的參數(shù)估計(jì)不準(zhǔn)確,參數(shù)估計(jì)的準(zhǔn)確性直接影響收益和風(fēng)險的測量,因此如何應(yīng)對模型中參數(shù)的估計(jì)誤差成為投資組合選擇問題的一個重要研究方向。
Chopra等[2]研究了均值、方差和協(xié)方差的誤差對最優(yōu)投資組合選擇的影響,指出一個小的參數(shù)誤差都可能會導(dǎo)致結(jié)果與真實(shí)最優(yōu)組合間產(chǎn)生很大的偏差。為了克服參數(shù)估計(jì)誤差的影響,常用的方法是魯棒優(yōu)化,Goldfarb等[3]提出了一個投資組合選擇的魯棒MV模型,并將其等價變換為一個二階錐規(guī)劃問題。然而,Scherer[4]指出,魯棒MV模型不僅不能帶來額外的回報(bào),反而增加了計(jì)算成本。最近,Cui等[5]考慮投資組合對單個資產(chǎn)期望收益和標(biāo)準(zhǔn)差的敏感度,利用導(dǎo)數(shù)作為敏感度的度量,將一組參數(shù)敏感度約束集成到傳統(tǒng)的MV模型中,提出了一種具有參數(shù)敏感度控制的MV模型,并通過分支定界算法求解,但當(dāng)分支變量維數(shù)較大時算法的收斂速度較慢,并且該模型中參數(shù)敏感度的上下界選擇比較困難。為克服參數(shù)敏感度上下界的選擇,Bai等[6]提出了一個帶參數(shù)敏感度的最優(yōu)權(quán)衡投資組合選擇問題,將該問題等價變換為無約束復(fù)合優(yōu)化問題,給出了一個改進(jìn)的加速梯度算法,證明了該算法收斂到無約束復(fù)合優(yōu)化問題的一個穩(wěn)定點(diǎn)。
本文研究帶參數(shù)敏感度的最優(yōu)權(quán)衡投資組合選擇問題的半定規(guī)劃(Semi-definite programming,SDP)松弛方法,以尋求其近似最優(yōu)解。眾所周知,SDP松弛可以為非凸二次約束二次規(guī)劃(Quadratically constrained quadratic programming,QCQP)問題提供更緊的下界或近似解[7-9]。Zheng等[10-11]對非凸QCQP問題給出了基于最佳DC(Difference of convex function,DC)分解、矩陣錐分解和多胞形逼近技術(shù)的SDP松弛。Luo等[12]對非凸QCQP問題提出了基于罰方法的SDP松弛。Luo等[13]對凸二次約束非凸QP問題利用Disjunctive割技術(shù)改進(jìn)了SDP松弛界。章顯業(yè)等[14]針對帶凸二次約束非凸QP問題提出了一個緊的雙非負(fù)規(guī)劃松弛及其解法。丁曉東等[15]給出了帶邊際風(fēng)險控制的投資組合優(yōu)化問題的一個緊的SDP松弛。受以上文獻(xiàn)[12-15]的啟發(fā),本文首先將帶參數(shù)敏感度的最優(yōu)權(quán)衡投資組合選擇問題變換為一個等價的非凸QCQP問題,進(jìn)而提出基于Secant割的SDP松弛,并給出它與原問題之間的間隙估計(jì),最后給出數(shù)值結(jié)果驗(yàn)證SDP松弛的緊性。
表1為GUROBI求解問題(4)與SDR對10個隨機(jī)測試?yán)拥钠骄鶖?shù)值結(jié)果,其中n為資產(chǎn)的數(shù)目,m為參數(shù)敏感度需要控制的資產(chǎn)數(shù)目。表1中“vOpt”“vLB”“vGap”“時間”分別表示10個測試問題得到的平均全局最優(yōu)值、平均下界、平均相對間隙和平均時間(單位:s),“(*)”表示GUROBI在1000 s內(nèi)找到全局最優(yōu)解的實(shí)例數(shù)。從表1的數(shù)值結(jié)果中可知,當(dāng)n=10,m=10時,SDR提供的下界與全局最優(yōu)值的平均相對間隙小于7%,對于20~200維的所有測試?yán)覵DR可以在13 s內(nèi)快速地找到原問題的一個全局最優(yōu)解,而GUROBI在n=50,m=40和n=200,m=80時無法在1000 s以內(nèi)找到問題(4)的10個例子的全局最優(yōu)解。此外,在表1所列的維數(shù)中,除n=100,m=20和n=200,m=20這兩組維數(shù)外,SDR的計(jì)算時間都優(yōu)于求解器GUROBI。
4 結(jié) 論
本文考慮了帶參數(shù)敏感度的最優(yōu)權(quán)衡投資組合問題,其模型是一個非凸優(yōu)化問題,求解它的全局最優(yōu)解是非常困難的。對該問題提出了一個等價的非凸QCQP變換問題,給出了該問題的一個緊的半定規(guī)劃松弛。初步的數(shù)值結(jié)果表明,本文提出的半定規(guī)劃松弛可以快速得到20至200維帶參數(shù)敏感度的最優(yōu)權(quán)衡投資組合問題的全局最優(yōu)解,并且對于大部分測試?yán)樱攵ㄒ?guī)劃松弛的求解時間小于求解器GUROBI。然而,目前只通過數(shù)值實(shí)驗(yàn)證明了該半定規(guī)劃松弛可求得原問題的全局最優(yōu)解,后續(xù)將針對更一般的問題研究基于該半定規(guī)劃松弛求解原問題的全局算法。
參考文獻(xiàn):
[1]Markowitz H M. Portfolio selection[J]. The Journal of Finance, 1952, 7(1): 77-91.
[2]Chopra V K, Ziemba W T. The effect of errors in means, variances, and covariances on optimal portfoliochoice[J]. The Journal of Portfolio Management, 1993, 19(2): 6-11.
[3]Goldfarb D, Iyengar G. Robust portfolio selection problems[J]. Mathematics of Operations Research, 2003, 28(1): 1-38.
[4]Scherer B. Can robust portfolio optimization help to build better portfolios[J]. Journal of Asset Management, 2007, 7(6): 374-387.
[5]Cui X T, Zhu S S, Li D, et al. Mean-variance portfolio optimization with parameter sensitivity control[J]. Optimization Methods and Software, 2016, 31(4): 755-774.
[6]Bai Y Q, Wei Y D, Li Q. An optimal trade-off model for portfolio selection with sensitivity of parameters[J]." Journal of Industrial amp; Management Optimization, 2017, 13(2): 947-965.
[7]Peng J M, Zhu T, Luo H Z, et al. Semi-definite programming relaxation of quadratic assignment problems based on nonredundant matrix splitting[J]. Computational Optimization and Applications, 2015, 60(1): 171-198.
[8]Parrilo P A. Semidefinite programming relaxations for semialgebraic problems[J]. Mathematical Programming, 2003, 96(2): 293-320.
[9]Anstreicher K M. Semidefinite programming versus the reformulation-linearization technique for nonconvex quadratically constrained quadratic programming[J]. Journal of Global Optimization, 2009, 43(2): 471-484.
[10]Zheng X J, Sun X L, Li D. Nonconvex quadratically constrained quadratic programming: best D.C. decompositions and their SDP representations[J]. Journal of Global Optimization, 2011, 50(4): 695-712.
[11]Zheng X J, Sun X L, Li D. Convex relaxations for nonconvex quadratically constrained quadratic programming: matrix cone decomposition and polyhedral approximation[J]. Mathematical Programming, 2011, 129(2): 301-329.
[12]Luo H Z, Bai X D, Peng J M. Enhancing semidefinite relaxation for quadratically constrained quadratic programming via penalty methods[J]. Journal of Optimization Theory and Applications, 2019, 180(3): 964-992.
[13]Luo H Z, Chen S K, Wu H X. A new branch-and-cut algorithm for non-convex quadratic programming via alternative direction method and semidefinite relaxation[J]. Numerical Algorithms, 2021, 88(2): 993-1024.
[14]章顯業(yè), 羅和治. 帶凸二次約束非凸二次規(guī)劃的雙非負(fù)規(guī)劃松弛及其解法[J]. 浙江理工大學(xué)學(xué)報(bào)(自然科學(xué)版), 2022, 47(4): 601-607.
[15]丁曉東, 肖琳燦, 羅和治. 帯邊際風(fēng)險控制的投資組合問題的半定規(guī)劃松弛[J]. 浙江工業(yè)大學(xué)學(xué)報(bào), 2017, 45(1): 64-68.
[16]Saxena A, Bonami P, Lee J. Convex relaxations of non-convex mixed integer quadratically constrained programs: Projected formulations[J]. Mathematical Programming, 2011, 130(2): 359-413.
[17]Zhu S S, Li D, Sun X L. Portfolio selection with marginal risk control[J]. The Journal of Computational Finance, 2010, 14(1): 3-28.
(責(zé)任編輯:康 鋒)