Adaptive evolutionary algorithms for portfolio selection problems

In this contribution we propose to solve complex portfolio selection problems via Evolutionary Algorithms (EAs) that resort to adaptive parameter control to manage the Exploration versus Exploitation balance and to find (near)-optimal solutions. This strategy modifies the algorithm's parameters during execution, and relies on continuous feedbacks provided to the EA with respect to some user-defined criteria. In particular, our study aims to understand whether a standard EA can benefit from a robust method that iteratively selects the crossover operator out of a predefined set, in the context of optimised portfolio choices. We apply this approach to large-scale optimization problems, by tackling a number of NP-hard mixed-integer programming problems. Our results show that generic EAs equipped with single crossover operator do not perform homogeneously across problem instances, whereas the adaptive policy leads to robust (and improved) solutions, by alternating exploration and exploitation on the basis of the features of the current search space.