A Fast Converging Evolutionary Algorithm for Constrained Multiobjective Portfolio Optimization

摘要

Portfolio optimization is a well-known problem in the domain of finance with reports dating as far back as 1952. It aims to find a tradeoff between risk and expected return for the investors, who want to invest finite capital in a set of available assets. Furthermore, constrained portfolio optimization problems are of particular interest in real-world scenarios where practical aspects such as cardinality (among others) are considered. Both mathematical programming and meta-heuristic approaches have been employed for handling this problem. Evolutionary Algorithms (EAs) are often preferred for constrained portfolio optimization problems involving non-convex models. In this paper, we propose an EA with a tailored variable representation and initialization scheme to solve the problem. The proposed approach uses a short variable vector, regardless of the size of the assets available to choose from, making it more scalable. The solutions generated do not need to be repaired and satisfy some of the constraints implicitly rather than requiring a dedicated technique. Empirical experiments on 20 instances with the numbers of assets, ranging from 31 to 2235, indicate that the proposed components can significantly expedite the convergence of the algorithm towards the Pareto front.