1: \begin{abstract} % 

2: This paper studies a distributionally robust portfolio optimization model with a cardinality constraint for limiting the number of invested assets. 

3: We formulate this model as a mixed-integer semidefinite optimization (MISDO) problem by means of the moment-based uncertainty set of probability distributions of asset returns. 

4: To exactly solve large-scale problems, we propose a specialized cutting-plane algorithm that is based on bilevel optimization reformulation. 

5: We prove the finite convergence of the algorithm. 

6: We also apply a matrix completion technique to lower-level SDO problems to make their problem sizes much smaller. 

7: Numerical experiments demonstrate that our cutting-plane algorithm is significantly faster than the state-of-the-art MISDO solver SCIP-SDP. 

8: We also show that our portfolio optimization model can achieve good investment performance compared with the conventional mean-variance model. 

9: \end{abstract}

10: