A Novel Initialization of PSO for Costly Portfolio Selection Problems
Posted: 23 Aug 2015
Date Written: July 2015
In this paper we propose an efficient initialization of a deterministic Particle Swarm Optimization (PSO) scheme. PSO has showed to be promising for solving several unconstrained global optimization problems from real applications, where derivatives are unavailable and the evaluation of the objective function tends to be costly. Here we provide a theoretical framework which motivates the use of a deterministic version of PSO, in place of the standard stochastic iteration currently adopted in the literature. Then, in order to test our proposal, we include a numerical experience using a realistic complex portfolio selection problem. This numerical experience includes the application of PSO to a parameter dependent unconstrained reformulation of the considered portfolio selection problem. The parameters are either adaptively updated as in an exact penalty framework, or they are tuned by the code REVAC. We show that in both these settings our PSO initialization is preferable with respect to the standard proposal from the literature.
Keywords: Deterministic PSO, Global Optimization, Portfolio Selection Problems, Exact Penalty functions
JEL Classification: G11, C44, C61
Suggested Citation: Suggested Citation