Inverse covariance matrix estimation for the global minimum variance portfolio

Abstract

The estimation of inverse covariance matrices plays a major role in portfolio optimization, for the global minimum variance portfolio in mean-variance analysis it is the only parameter used to determine the asset allocation. In this thesis I propose to of use the graphical lasso methodology to directly estimate the inverse covariance matrix, and apply it to the global minimum variance portfolio. The results indicate that the graphical lasso provides better out-of-sample portfolio variance than the traditional sample estimator.