Applications of Genetic Algorithm to Portfolio Optimization with Practical Transaction Constraints

The portfolio optimization model, initially proposed by Markowitz in 1952 and known as mean-variance model (MV model), is applied to find the optimized allocation among assets to get higher investment return and lower investment risk. However, the MV model did not consider some practical limitations of financial market, including: (1) transaction cost and (2) minimal transaction lots. While these constraints are not considered in the model, the practicability of the model will be restrained. But when they are included in the model, the model will become an NP hard problem, which cannot obtain global optimal solution by traditional mathematics programming techniques. In this research, besides proposing various models to include afore-mentioned consideration in the MV model, genetic algorithms are applied to solve these models. Empirical tests in the Taiwan stock market are provided to prove the applicability of the techniques.