Statistical Inference for Markowitz Efficient Portfolios

This dissertation, "Statistical Inference for Markowitz Efficient Portfolios" by Yuanyuan, Zhu, 朱淵遠, was obtained from The University of Hong Kong (Pokfulam, Hong Kong) and is being sold pursuant to Creative Commons: Attribution 3.0 Hong Kong License. The content of this dissertation has not been altered in any way. We have altered the formatting in order to facilitate the ease of printing and reading of the dissertation. All rights not granted by the above license are retained by the author. Abstract: Abstract of the thesis entitled ST A TISTICAL INFERENCE FOR MARKOWITZ EFFICIENT POR TFOLIOS Submitted by ZHU, YUANYUAN for the degree of Do ctor of Philosophy at The University of Hong Kong in September 2015 Markowitz mean-v ariance mo del has been the foundation of modern portfolio theory . The Markowitz model attempts to maximize the portfolio expected return for a given level of portfolio risk, or equiv alently to minimize portfolio risk for a given level of expected return. Assuming multivariate normality of the asset returns, the optimal portfolio weights can be treated as a function of the unknown mean vector and covariance matrix. However it has b een criti- cized by many researchers the ineective and unstable performance of the op- timal portfolio under the model. This thesis intends to improve the Markowitz mean-variance model through two new methods. The rst method is to make use of generalized pivotal quantity (GPQ). The GPQ approach is widely used in constructing hypothesis tests and condence interv als. In this thesis, the GPQ approach is used to make statistical inference on the optimal portfolio weights. Dierent approaches are proposed for con- structing point estimator and simultaneous condence interv als for the optimal portfolio weights. Simulation studies has been conducted to compare the GPQ estimators with existing estimators based on Markowitz model, bootstrap andshrinkage methods. The results show that the GPQ based approach results in a smallest mean squared error for the point estimate of the portfolio weights in most cases and satisfactory coverage rate for the simultaneous condence interv als. F urthermore, an application on portfolio re-balancing problem is considered. Results show that the condence intervals help investors decide whether or not to update the p ortfolio weights so as to achieve a higher prot. This thesis not only focuses on the portfolio optimal weights, but also proposes a new estimator for the Sharpe ratio. Sharpe ratio serves as an important measure of the portfolio performance measure. Some researches have been done on the estimation of the distribution of Sharpe ratio when the number of assets is not too large but the sample size is big. This thesis makes use of GPQ to estimate the Sharpe ratio for high-dimensional data or small-sample-size data. The second method attempts to improve the estimation of the unknown cov ariance matrix. Note that the plug-in estimator for the optimal portfolio weights is biased and p erforms po orly due to the estimation error, especially in the cases of high dimensions. Instead of the sample covariance matrix, we consider the scaled sample cov ariance matrix to construct the new estimator for weights. The explicit formulae for both the mean and v ariance of the new estimator are derived. T wo approaches are prop osed to determine the optimal scale parameter of the covariance matrix estimator. Simulation studies show that the new estimators outperform the existing ones, especially when the number of assets is large. In addition, we illustrate the new estimators with an example from the US stock market. DOI: 10.5353/th_b5689290 Subjects: Portfolio management - Statistical methods