Archive - Central European Conference on Information and Intelligent Systems, CECIIS - 2009

Font Size: 
Cardinality Constrained Portfolio Optimization by Means of Genetic Algorithms
Ivica Martinjak

Last modified: 2009-08-30

Abstract


When  applying  the  standard
Markowitz  mean-variance  model  on  a  real
portfolio  selection  problem,  we  are  faced  with
certain  limitations  like  cardinality  of  chosen
assets,  discrete  nature  of  trading  variables  etc.
While  the classical mean-variance model can be
successfully  solved  by  standard  algorithms
(quadratic  programming),  modelling  an  actual
investment  leads  to  NP-hard  optimization
problem.  In  such  circumstances  heuristic
methods appear as the only way out.
This  paper  aims  at  finding  efficient
evolutionary  inspired  algorithm  for  cardinality
constrained  portfolio  optimization.  Among
developed  algorithms  which  were  able  to  solve
problems  with  very  many  possible  assets,  the
algorithm with hybrid crossover presents itself as
the  most  effective.  In  order  to  make  obtained
results comparable, test sample was chosen from
databases  that  serve  as  a  benchmark  for  this
problem class.

Full Text: PDF