A0891
Title: On minimum variance high-dimensional portfolio: How to shrink a covariance matrix?
Authors: Cy Sin - National Tsing Hua University (Taiwan) [presenting]
Abstract: The global minimum variance (GMV) portfolio forms the basis of the so-called mean-variance analysis pioneered by past studies. However, variance is unobserved in neither the training set nor the test set. Worse still, it is well-known that unreliable results are obtained should the population variances and covariances be replaced by the sample variances and covariances. The lines of a recent study are followed, and consider the double-shrinkage portfolio in the sense that both the estimated covariance matrix and the portfolio weight are shrunk. The weights are shrunk towards equal weights while (correspondingly) the estimated covariance matrix is shrunk towards the identity matrix. Monte-Carlo simulations are performed to assess (a) the threshold covariance matrix suggested by prior studies, (b) the linear shrinkage covariance matrix suggested by another study, and (c) the non-linear shrinkage covariance matrix suggested by a past study.
