1: \begin{abstract}

2: This paper studies mean-risk portfolio optimization models using the conditional value-at-risk (CVaR) as a risk measure. 

3: We also employ a cardinality constraint for limiting the number of invested assets.

4: Solving such a cardinality-constrained mean-CVaR model is computationally challenging for two main reasons. 

5: First, this model is formulated as a mixed-integer optimization (MIO) problem because of the cardinality constraint, so solving it exactly is very hard when the number of investable assets is large. 

6: Second, the problem size depends on the number of asset return scenarios, and the computational efficiency decreases when the number of scenarios is large. 

7: To overcome these challenges, we propose a high-performance algorithm named the \emph{bilevel cutting-plane algorithm} for exactly solving the cardinality-constrained mean-CVaR portfolio optimization problem.

8: We begin by reformulating the problem as a bilevel optimization problem and then develop a cutting-plane algorithm for solving the upper-level problem.

9: To speed up computations for cut generation, we apply to the lower-level problem another cutting-plane algorithm for efficiently minimizing CVaR with a large number of scenarios. 

10: Moreover, we prove the convergence properties of our bilevel cutting-plane algorithm. 

11: Numerical experiments demonstrate that, compared with other MIO approaches, our algorithm can provide optimal solutions to large problem instances faster.

12: \keywords{%% keywords here, in the form: keyword \sep keyword

13: Mixed-integer optimization \and Portfolio optimization \and Cardinality constraint \and Conditional value-at-risk \and Cutting-plane algorithm

14: }

15: % \PACS{PACS code1 \and PACS code2 \and more}

16: % \subclass{MSC code1 \and MSC code2 \and more}

17: \end{abstract}

18: