Second order stochastic dominance is an optimal rule for portfolio selection of risk averse investors when we only know that the investors' utility function is increasing concave. The main advantage of SSD is that it makes no assumptions regarding the return distributions of investment assets and has been proven to lead to utility maximization for the class of increasing concave utility functions. A number of different SSD models have emerged in the literature for portfolio selection based on SSD. However, current SSD models produce the same SSD efficient portfolio for all risk averse investors, regardless of their risk aversion degree. In this thesis, we have developed a new SSD efficiency model, SSD-DP, which unlike existing SSD efficiency models in the literature, provides an SSD efficient portfolio as a function of investors' risk aversion degrees. The SSD-DP model is based on the linear programming technique and finds an SSD efficient portfolio by minimizing the dual power transform (DP) of a weighted portfolio of assets for a given risk aversion degree. We show that the optimal portfolio of the proposed model is SSD efficient, i.e. it is not dominated by SSD by any other portfolio, and, through empirical studies of historical data, we show that the method is a promising tool for constructing trading strategies.

When portfolio selection is implemented by using the past sample values, parameter uncertainty may lead to suboptimal portfolios. Previous studies of portfolio selection demonstrate that classical approach based on the simple mean estimator is less reliable cause of inherent estimation error. In this paper, we investigate a shrinkage estimator based on Stein’s idea in measuring the expected returns. We apply the research of Jorion (1985) to Taiwan Stock market, present the effects of estimation error on the portfolio selection and demonstrate that the shrinkage estimator is robust and dominates the classical estimator on the MSE criterion. In addition, we also examine the effect of different shrinkage target on the performance of the Bayes-Stein estimator and find that this estimator still has lower risk than the classical sample mean.

Zugl.: Karlsruhe, Univ., Diss., 2004

Zugl.: Darmstadt, Techn. Univ., Diss., 2005

Thesis (Maste). / Includes bibliographical references.

Zugl.: Masterarbeit

Leipzig, Universiẗat, Diss., 2009.

Thesis (doctoral)--Universiẗat, Köln, 2003.

Thesis (doctoral)--Universität St. Gallen, 2004.

Thesis (doctoral)--Universität St. Gallen, 2002.