An entropy-driven expert system shell applied to portfolio selection

摘要

In modern portfolio theory like that of Markowitz or Sharpe the investor follows a mean/variance rationality. Even the founders of this theory observed unsatisfactory results because of symmetrical risk measures such as variance and standard deviation. Post-modern theory then considers downside risk measures and takes into consideration the investor’s specific goals. In this contribution we follow these ideas, but use an information theoretical inference mechanism under maximum entropy and minimum relative entropy, respectively. The approach results in a high performance expert system under the shell SPIRIT, combining an index model with the new method. For three DAX-listed blue chips and for varying risk attitudes of the investor the system’s portfolio selection capacity is compared to that of classical Markowitz and Sharpe optimization.