Mean-Variance Analysis with Pattern Recognition

Story:

What are the differences between Black-Litterman model and traditional mean-variance analysis? Maybe I should not state the question in this way. Since mean-variance analysis is a generic parametric framework for portfolio optimization, Black-Litterman is a specific portfolio optimization model. The uncertainties of expected returns and expected variance-covariance matrix together with the mean-variance framework's sensitivity regarding its inputs usually lead to the bad performance of mean-variance analysis. Then Black-Litterman model addressed the issue by replacing the expected returns with CAPM equilibrium returns and incorporating investor's views.

In this simple research, the expected returns in the mean-variance analysis were replaced with series pattern recognition forecasted returns.

Data:

Randomly draw 50 stocks from SP500 to construct a monthly portfolio.

Sample the process by 20 times (Cannot do more. It is kind of slow....).

Year Range: 2008-2016

Results:

Figure 1. Mean annual returns of portfolios implemented with various strategies.

Of course, equal weighted portfolios outperform the value weighted portfolios. Mean-Variance analysis with historical mean and covariance matrix as inputs has poor performance. If the expected returns were replaced with series pattern recognition forecasted returns, the results are slightly better.

Below I also attached several cumulative returns' plots of sample portfolios.

Sample 1:

Sample 2:

Sample 3: