Convex optimization approaches to maximally predictable portfolio selection

Jun Ya Gotoh, Katsuki Fujisawa

Research output: Contribution to journalArticle

3 Citations (Scopus)

Abstract

In this article we propose a simple heuristic algorithm for approaching the maximally predictable portfolio, which is constructed so that return model of the resulting portfolio would attain the largest goodness-of-fit. It is obtained by solving a fractional program in which a ratio of two convex quadratic functions is maximized, and the number of variables associated with its nonconcavity has been a bottleneck in spite of continuing endeavour for its global optimization. The proposed algorithm can be implemented by simply solving a series of convex quadratic programs, and computational results show that it yields within a few seconds a (near) Karush-Kuhn-Tucker solution to each of the instances which were solved via a global optimization method in [H. Konno, Y. Takaya and R. Yamamoto, A maximal predictability portfolio using dynamic factor selection strategy, Int. J. Theor. Appl. Fin. 13 (2010) pp. 355-366]. In order to confirm the solution accuracy, we also pose a semidefinite programming relaxation approach, which succeeds in ensuring a near global optimality of the proposed approach. Our findings through computational experiments encourage us not to employ the global optimization approach, but to employ the local search algorithm for solving the fractional program of much larger size.

Original languageEnglish
Pages (from-to)1713-1735
Number of pages23
JournalOptimization
Volume63
Issue number11
DOIs
Publication statusPublished - Nov 2014
Externally publishedYes

Fingerprint

Portfolio Selection
Convex optimization
Global optimization
Convex Optimization
Global Optimization
Fractional
Semidefinite Programming Relaxation
Global Optimality
Convex Program
Quadratic Program
Local Search Algorithm
Predictability
Heuristic algorithms
Goodness of fit
Quadratic Function
Computational Experiments
Heuristic algorithm
Convex function
Optimization Methods
Computational Results

All Science Journal Classification (ASJC) codes

  • Control and Optimization
  • Management Science and Operations Research
  • Applied Mathematics

Cite this

Convex optimization approaches to maximally predictable portfolio selection. / Gotoh, Jun Ya; Fujisawa, Katsuki.

In: Optimization, Vol. 63, No. 11, 11.2014, p. 1713-1735.

Research output: Contribution to journalArticle

@article{a4282600925841c2a1d4d1c587e4a06c,
title = "Convex optimization approaches to maximally predictable portfolio selection",
abstract = "In this article we propose a simple heuristic algorithm for approaching the maximally predictable portfolio, which is constructed so that return model of the resulting portfolio would attain the largest goodness-of-fit. It is obtained by solving a fractional program in which a ratio of two convex quadratic functions is maximized, and the number of variables associated with its nonconcavity has been a bottleneck in spite of continuing endeavour for its global optimization. The proposed algorithm can be implemented by simply solving a series of convex quadratic programs, and computational results show that it yields within a few seconds a (near) Karush-Kuhn-Tucker solution to each of the instances which were solved via a global optimization method in [H. Konno, Y. Takaya and R. Yamamoto, A maximal predictability portfolio using dynamic factor selection strategy, Int. J. Theor. Appl. Fin. 13 (2010) pp. 355-366]. In order to confirm the solution accuracy, we also pose a semidefinite programming relaxation approach, which succeeds in ensuring a near global optimality of the proposed approach. Our findings through computational experiments encourage us not to employ the global optimization approach, but to employ the local search algorithm for solving the fractional program of much larger size.",
author = "Gotoh, {Jun Ya} and Katsuki Fujisawa",
year = "2014",
month = "11",
doi = "10.1080/02331934.2012.741237",
language = "English",
volume = "63",
pages = "1713--1735",
journal = "Optimization",
issn = "0233-1934",
publisher = "Taylor and Francis Ltd.",
number = "11",

}

TY - JOUR

T1 - Convex optimization approaches to maximally predictable portfolio selection

AU - Gotoh, Jun Ya

AU - Fujisawa, Katsuki

PY - 2014/11

Y1 - 2014/11

N2 - In this article we propose a simple heuristic algorithm for approaching the maximally predictable portfolio, which is constructed so that return model of the resulting portfolio would attain the largest goodness-of-fit. It is obtained by solving a fractional program in which a ratio of two convex quadratic functions is maximized, and the number of variables associated with its nonconcavity has been a bottleneck in spite of continuing endeavour for its global optimization. The proposed algorithm can be implemented by simply solving a series of convex quadratic programs, and computational results show that it yields within a few seconds a (near) Karush-Kuhn-Tucker solution to each of the instances which were solved via a global optimization method in [H. Konno, Y. Takaya and R. Yamamoto, A maximal predictability portfolio using dynamic factor selection strategy, Int. J. Theor. Appl. Fin. 13 (2010) pp. 355-366]. In order to confirm the solution accuracy, we also pose a semidefinite programming relaxation approach, which succeeds in ensuring a near global optimality of the proposed approach. Our findings through computational experiments encourage us not to employ the global optimization approach, but to employ the local search algorithm for solving the fractional program of much larger size.

AB - In this article we propose a simple heuristic algorithm for approaching the maximally predictable portfolio, which is constructed so that return model of the resulting portfolio would attain the largest goodness-of-fit. It is obtained by solving a fractional program in which a ratio of two convex quadratic functions is maximized, and the number of variables associated with its nonconcavity has been a bottleneck in spite of continuing endeavour for its global optimization. The proposed algorithm can be implemented by simply solving a series of convex quadratic programs, and computational results show that it yields within a few seconds a (near) Karush-Kuhn-Tucker solution to each of the instances which were solved via a global optimization method in [H. Konno, Y. Takaya and R. Yamamoto, A maximal predictability portfolio using dynamic factor selection strategy, Int. J. Theor. Appl. Fin. 13 (2010) pp. 355-366]. In order to confirm the solution accuracy, we also pose a semidefinite programming relaxation approach, which succeeds in ensuring a near global optimality of the proposed approach. Our findings through computational experiments encourage us not to employ the global optimization approach, but to employ the local search algorithm for solving the fractional program of much larger size.

UR - http://www.scopus.com/inward/record.url?scp=84905188331&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84905188331&partnerID=8YFLogxK

U2 - 10.1080/02331934.2012.741237

DO - 10.1080/02331934.2012.741237

M3 - Article

AN - SCOPUS:84905188331

VL - 63

SP - 1713

EP - 1735

JO - Optimization

JF - Optimization

SN - 0233-1934

IS - 11

ER -