Asset strongAllocationstrong for Robust Portfolios - PortfolioConstruction.pdfVIP

Asset Allocation for Robust Portfolios Working Paper: March 2004 Tim Farrelly Visiting Fellow University of Technology Sydney Summary A new risk return optimisation model is described that overcomes much of the instability inherent in Mean Variance optimisers, with minimal sacrifice of efficiency. The Robust Frontier model identifies portfolios that are close to the efficient frontier, less likely to produce extreme results and, most importantly, are much more stable to changes in input assumptions. Practitioners using this model can approach the task of asset allocation with confidence that small errors in forecasts will not be the primary driver of asset allocation outputs and that, if the model is used over time, it will not produce huge swings in allocations with consequent high transaction costs. Difficulties in implementing Mean Variance style optimisers Various authors have documented the difficulties that many practitioners have had in implementing Mean Variance (MV) style optimisers. At the heart of these difficulties is the instability inherent in the way the MV models trade off risk and return. Michaud (1989) and others have referred to the extreme sensitivity of Mean Variance style optimisers to imput assumptions. Small changes in input variables can have quite profound impacts on the structure of Efficient Portfolios, so much so that Michaud refers to MV optimisers as error maximisers. This instability is then compounded by the difficulty in developing precise imput data. • Estimates for future expected returns carry a large measure of uncertainty at the best of times. • There has been considerable debate over what is the best measure of risk, let alone how one should then estimate that measure. Balzer (1994) and others reviewed this issue at some length. Which measure should we use, and how best to gain a forward looking estimate of that measure? • Correlation co