Análise de desempenho das metaheurísticas algoritmo genético e enxame de partículas aplicadas ao problema de otimização de portfólio financeiro

Abstract

The Financial Portfolio Optimization Problem involves determining the optimal proportion and the best allocation of resources in financial assets, aiming to maximize profits and minimize risks. Over the years, the classical model for this problem, introduced by Markowitz (1952), has been improved with different additions of restrictions that sought to make it more aligned with the investor’s reality, such as, for example, the cardinality restrictions of the portfolio, the floor and ceiling of the investment for each asset and the budget. Thus, the complexity of the problem increases when it comes to finding exact solutions, which suggests the use of metaheuristic procedures, which stand out for their ability to find approximate and flexible solutions to adapt to different models of the problem. In this context, the present work investigates the performance and compares two metaheuristics when applied to the Portfolio Optimization Problem, the Particle Swarm (PSO) and the Genetic Algorithm (GA), the latter being improved with a component procedure inspired by the literature. To perform the experiments, classical instances and the metrics of efficient frontier, mean Euclidean distance, mean return error, mean variance error and execution time were used. After 75 cases of analysis, it was found that the Genetic Algorithm demonstrated better performance.