Mean-VaR portfolio selection under real constraints

This paper is concerned with asset allocation under real constraints when VaR is the risk measure to minimize. Our paper makes a contribution in several ways, we use a risk measure that is not linear programming solvable, we introduce real constraints, such as minimum transaction units and non-linear cost structure and, finally, we avoid the use of smoothing techniques. The approach we propose is based on multi-objective genetic algorithms. The results presented show the adequacy of the method for the portfolio optimization problem and emphasize the importance of dealing with real constraints during the optimization process.