Optimization · 2009 · 30 citations · 14 references
BusinessEconomicsFinanceFinancial EconomicsOptimizationPortfolio ManagementAsset AllocationOptimization Under UncertaintyPortfolio OptimizationPortfolio TheoryRobust OptimizationRobust StatisticsEfficiency-robustness TradeoffEfficient Frontier AnalysisEllipsoidal Uncertainty SetsParameter RobustnessPortfolio EfficiencyRobust Portfolio OptimizationRobustnessUncertainty Sets
Abstract In recent years, new ideas for the robustification of the traditional mean-variance frontier have appeared in the literature. One of these approaches is based upon the robust counterpart idea, introduced by Ben-Tal and Nemirovski. To make this idea work in practice, realistic uncertainty sets have to be specified. In this article, we are investigating two popular uncertainty sets together with their impact on the efficient frontier. As a main contribution we will provide two theorems stating that the corresponding robust efficient frontiers are identical to the classical efficient frontier up to some risk level. This especially implies that the classical efficient portfolios are already robust themselves. We will briefly discuss the relevance of this finding for practical applications. Keywords: portfolio optimizationefficient frontierconfidence ellipsoidAMS Subject Classifications:: 90C2590C9091B2862H12 Acknowledgements We are most thankful for the comments of three anonymous referees, who helped to improve the quality of the paper. In addition, fruitful comments by an early version by Rudi Zagst and Jan Kallsen are much appreciated. Notes Notes 1. With 'lower part' of the efficient frontier we mean the part when starting in the minimum variance portfolio as the lower left end of the frontier and moving towards more riskier allocations, hence points with higher risk and expected returns lying higher and more to the right on the efficient frontier. 2. The crucial part of the extension lies in the proper choice of the estimator for the covariance matrix. This estimator has to be uncorrelated to the estimator of the mean. Otherwise, results as in Section 3.2 cannot be obtained. 3. In line with [23], we are considering the German DAX, i.e. its 30 underlying stocks, as realistic example. Stock prices of all 30 assets from 01/08/2006 to 01/08/2007 are used as the basis for all investigations. Numbers have only been annualized in figures for illustrating purposes. A maximum weight of 10% for each asset was applied in accordance with the definition of the strategy indices. 4. For elliptical distributions, the sample estimator of the mean coincides with the maximum likelihood estimator, see e.g. Citation6, Section 4.1]. Thus, the analysis in this section is valid for general elliptical distributions as well. 5. The sample and the maximum likelihood estimator for μ coincide, see above, and the sample and the maximum likelihood estimator for Σ only differ by the constant . For general elliptical distributions, it is important that the estimators of the mean and the covariance remain uncorrelated. 6. This fact is not illustrated in any figure, since for comparison with the classical efficient frontier, the point estimates and are used for plotting the robust efficient frontier as well. The robust return and covariance are merely necessary to determine the robust portfolio allocations.
14
Optimization by Vector Space Methods
Students Quarterly Journal · 1970 · 6.1K citations
Natural Sciences, Industrial Engineering, Design Optimization +6
Harry M. Markowitz · The Journal of Finance · 1952 · 4.5K citations
Aharon Ben‐Tal, Arkadi Nemirovski · Mathematics of Operations Research · 1998 · 2.6K citations