The Traditional Approach to Asset Allocation

After presenting the preliminary definitions and statistics that are necessary to correctly formulate portfolio risk and return, the chapter illustrates the appropriate way of constructing portfolios according to the Markowitz model, also named Mean-Variance Optimization. Its application requires optimization inputs (expected returns, risks, correlations) that are not observable ex-ante, thus they have to be estimated usually from past data. Then the standard implementation of Markowitz’s model follows the “plug-in” rule. In practice, the estimated parameters are processed into the mean-variance optimizer that treats them as if they were the true parameters. This renders the portfolio construction process deterministic as the parameters uncertainty is completely neglected. This chapter gives evidence and clarification of the different undesirable features and deficiencies of the optimized portfolios (counter-intuitive nature, instability, erroneously supposed uniqueness and poor out-of-sample performance) triggered by implementing the Mean-Variance Optimization without recognizing the existence of estimation risk inherent in the input parameters. Amongst other, estimation errors in expected returns are the most crucial and costly ones.