Title:Bilevel Cutting-Plane Algorithm for Solving Cardinality-Constrained Mean-CVaR Portfolio Optimization Problems

Abstract:This paper studies the mean-risk portfolio optimization problems with a constraint of the number of assets to be invested. We employ conditional value-at-risk (CVaR) as a risk measure. While several studies aim at efficiently solving mean-CVaR model, it has been computationally expensive to solve cardinality-constrained mean-CVaR model, and especially for exact algorithms, the size of a problem that can be handled has been limited. In this paper, we devised an exact and efficient algorithm to handle a large-sized mean-CVaR optimization problem with a cardinality constraint. For computational efficiency, our proposed method integrates two cutting-plane algorithms, upper-level cutting-plane algorithm and lower-level one, and each algorithm solves the master problem and the subproblem, respectively. Numerical experiments demonstrated that our algorithms can attain optimal solutions to large-sized problems in a reasonable amount of time.