Efficient estimation in portfolio management

This thesis investigates whether estimating the inputs of the Markowitz (1952) Mean-Variance framework using various econometric techniques leads to improved optimalportfolio allocations at the country, sector and stock levels over a number of timeperiods. We build upon previous work by using various combinations of conventionaland Bayesian expected returns and covariance matrix estimators in a Mean-Varianceframework that incorporates a benchmark reference, an allowable deviation range fromthe benchmark weights and short-selling constraints so as to achieve meaningful andrealistic outcomes. We found that models based on the classical maximum likelihoodmethod performed just as well as the more sophisticated Bayesian return estimators inthe study. We also found that the covariance matrix estimators analysed createdcovariance matrices that were similar to one another and, as a result, did not seem tohave a large effect on the overall portfolio allocation. A sensitivity analysis on the levelof risk aversion confirmed that the simulation results were robust for the different levelsof risk aversion.