Constrained Portfolio Optimization in Liability-Driven Investing

Abstract

In this thesis we formulate and implement a multi-stage portfolio optimization model, and solve it using a genetic algorithm. The goals of the thesis are, apart from formulating and implementing the problem, to estimate suitable parameters for the scenario generation, and to make sure that the problem is solved in a computationally efficient manner. Lastly, we investigate and discuss the performance of the complete system, including financial aspects of the produced solutions, the stability of the solutions, and the computational complexity of the model. We find that our problem formulation is useful, and that it it allows for great flexibility with regards to adding new constraints. We also find that our genetic algorithm solves the problem in reasonable time. Before the model can be used in practice however, results show that it needs to be improved with regards to stability in the solutions.