摘要
Using the concepts from information theory, it is possible to improve the traditional methodologies of asset allocation. In this paper, it was studied and extended the two existent approaches： the first is based on the Shannon entropy concept and the second on the Kullback-Leibler distance. In modem portfolio theory, the investor has two basic procedures： the choice of a portfolio that maximizes its risk-adjusted excess return or the mixed allocation between the maximum Sharpe portfolio and the risk-free asset. The first procedure was already addressed in the related literature. One important contribution of this paper is the consideration of the second procedure in the information theory context. The performance of these approaches was compared with the three traditional asset allocation methodologies： the Markowitz＇s mean-variance, the resampled mean-variance and the equally weighted portfolio. It was used simulated and real data from Brazilian stocks. The information theory-based methodologies were verified to be more robust when dealing with the estimation errors.
