Portfolio Optimization and Ambiguity Aversion

This thesis analyses whether considering ambiguity aversion in portfolio optimization improves the out-of-sample performance of portfolio optimization approaches. Furthermore, it is assessed which role ambiguity aversion plays in improving the portfolio performance, especially compared with the role of estimation errors. This is done by evaluating the out-of-sample performance of the approach of Garlappi, Uppal and Wang for an investor with multiples priors and aversion to ambiguity compared to other portfolio optimization strategies from the literature not taking ambiguity aversion into account. It is shown that considering ambiguity aversion in portfolio optimization can improve the out-of-sample performance compared to the sample based mean-variance model and the Bayes-Stein model. However, the minimum-variance model and the model of naïve diversification, which are both independent of expected returns, outperform the approach considering ambiguity aversion for most of the empirical applications shown in this thesis. These results indicate that ambiguity aversion does play a role in portfolio optimization, however, estimation errors regarding expected returns overshadow the benefits of optimal asset allocation.