Modified generalized simulated annealing algorithm used in data driven portfolio management

In industry, fund managers combine their expertise with quantitative tools to construct portfolios. They research on companies with fundamental as well as technical analysis. After choosing proper stocks, certain quantitative models, e.g. Markowitz model, SIM model, Multi-group model, are applied to determine the optimal weight for each stocks. This general method has a vital draw-back: it relies too much on the portfolio manager's subjective idea and expertise. This is one of the reason why 86% of the active managed funds fail to beat the market index in 2014. In this paper, we propose a pure data-driven method to construct optimal portfolio. We develop a modified generalized simulated annealing algorithm used in the global optimization process which selects the most suitable stocks. We adopt the multi-group model in the decision of each stocks' weights. We use the data of all S&P 500 composites from Jan 2009 to May 2014 to construct the portfolio and use the data from June 2014 to May 2015 to test the performance. We show that our model clearly outperforms the S&P 500 both towards the average return and the Sharpe Ratio.