A Genetic Algorithm for the Asset Paring Problem in Portfolio Optimization

摘要

We have developed and tested a genetic algorithm (GA) for helping portfolio managers construct optimal (mean-variance) portfolios subject to a constraint on number of assets that may be in the portfolio. This is a practical problem of importance to many quantitative portfolio managers. Our results are encouraging, showing that GAs can often find very good solutions. The GA's performance was found to be sensitive to the crossover rate, population size, and to the choice to use an elitist strategy. Currently, however, the GA converges to "good" solutions much more slowly than some specially developed heuristics.