Mean-variance model based on filters of minimum spanning tree

This study aims to reduce the statistical uncertainty of the correlation coefficient matrix in the mean-variance model of Markowitz. A filtering algorithm based on minimum spanning tree (MST) is proposed. Daily data of the 30 stocks of the Hang Seng Index (HSI) and Dow Jones Index (DJI) from 2004 to 2009 are selected as the base dataset. The proposed algorithm is compared with the Markowitz method in terms of risk, reliability, and effective size of the portfolio. Results show that (1) although the predicted risk of portfolio built with the MST is slightly higher than that of Markowitz, the realized risk of MST filtering algorithm is much smaller; and (2) the reliability and the effective size of filtering algorithm based on MST is apparently better than that of the Markowitz portfolio. Therefore, conclusion is that filtering algorithm based on MST improves the mean-variance model of Markowitz.