This paper describes a new model for portfolio optimization (PO), using entropy and mutual information instead of variance and covariance as measurements of risk. We also compare the performance in and out of sample of the original Markowitz model against the proposed model and against other state of the art shrinkage methods. It was found that ME (mean-entropy) models do not always outperform their MV (mean-variance) and robust counterparts, although presenting an edge in terms of portfolio diversity measures, especially for portfolio weight entropy. It further shows that when increasing return constraints on portfolio optimization, ME models were more stable overall, showing dampened responses in cumulative returns and Sharpe indexes in comparison to MV and robust methods, but concentrated their portfolios more rapidly as they were more evenly spread initially. Finally, the results suggest that it was also shown that, depending on the market, increasing return constraints may have positive or negative impacts on the out-of-sample performance.
Keywords: entropy; mutual information; portfolio optimization; variance and covariance.