Application And Research On Genetic Algorithm In Security Investment

Investment in securities have many characteristics, such as quick changes of market,complex factors and risk uncertainty. In order to offset the risk, we need to choose anumber of different securities with different proportions, which is a portfolio approach.Portfolio problem is a complex optimization problem, and it is difficult to find the globaloptimal solution using the conventional algorithm in a short time. The genetic algorithmhas a strong applicability, and it almost have no requires on the nature of objective function,which is a feasible method for solving the portfolio problem. This paper mainly discusseshow to build a portfolio model and how to design algorithm to solve the models for thesecurity investment problem.This paper uses theory of utility, assumption of rational people in economics andportfolio variance formula in statistical theory. According to the CAPM and the sharpe ratio,I proposed a new portfolio model: excess returns based on the unit systemic risk model.This model measures risk using standard deviation of portfolio, and uses maximization ofthe excess returns under unit systemic risk as the objective function. It compares the meritsof different portfolios according to the return of unit risk. This model is more in line withthe investment demand of real market compared with the pure maximizing revenue model.First I give the detailed derivation of the model, and then design a genetic algorithm tosolve the model.This paper designs double genetic algorithm to solve the portfolio problem. The firstlayer genetic algorithm filters out the sample range for the model. The second layer geneticalgorithm designs the weight coding for the model, so that the final results determine acombination of securities investment ratio.Finally, the algorithm implements using the data of Shanghai and Shenzhen A-sharemarket as empirical analysis. The results verify that this model can disperse investmentrisks effectively, and the holding period yield is higher than the average level. Model andalgorithm have achieved a good result.